Supported ONNX operators#
OnnxAbs#
- class mlprodict.npy.xop_auto_import_.OnnxAbs(*args, **kwargs)#
Version
name: Abs (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Absolute takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the absolute is, y = abs(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxAbs_1#
- class mlprodict.npy.xop_auto_import_.OnnxAbs_1(*args, **kwargs)#
Version
name: Abs (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Absolute takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the absolute is, y = abs(x), is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAbs_13#
- class mlprodict.npy.xop_auto_import_.OnnxAbs_13(*args, **kwargs)#
Version
name: Abs (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Absolute takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the absolute is, y = abs(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxAbs_6#
- class mlprodict.npy.xop_auto_import_.OnnxAbs_6(*args, **kwargs)#
Version
name: Abs (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Absolute takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the absolute is, y = abs(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxAcos#
- class mlprodict.npy.xop_auto_import_.OnnxAcos(*args, **kwargs)#
Version
name: Acos (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the arccosine (inverse of cosine) of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The arccosine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAcos_7#
- class mlprodict.npy.xop_auto_import_.OnnxAcos_7(*args, **kwargs)#
Version
name: Acos (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the arccosine (inverse of cosine) of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The arccosine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAcosh#
- class mlprodict.npy.xop_auto_import_.OnnxAcosh(*args, **kwargs)#
Version
name: Acosh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic arccosine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic arccosine values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAcosh_9#
- class mlprodict.npy.xop_auto_import_.OnnxAcosh_9(*args, **kwargs)#
Version
name: Acosh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic arccosine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic arccosine values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAdd#
- class mlprodict.npy.xop_auto_import_.OnnxAdd(*args, **kwargs)#
Version
name: Add (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary addition (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxAdd_1#
- class mlprodict.npy.xop_auto_import_.OnnxAdd_1(*args, **kwargs)#
Version
name: Add (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Performs element-wise binary addition (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.consumed_inputs: legacy optimization attribute.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAdd_13#
- class mlprodict.npy.xop_auto_import_.OnnxAdd_13(*args, **kwargs)#
Version
name: Add (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Performs element-wise binary addition (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxAdd_14#
- class mlprodict.npy.xop_auto_import_.OnnxAdd_14(*args, **kwargs)#
Version
name: Add (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary addition (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxAdd_6#
- class mlprodict.npy.xop_auto_import_.OnnxAdd_6(*args, **kwargs)#
Version
name: Add (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Performs element-wise binary addition (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxAdd_7#
- class mlprodict.npy.xop_auto_import_.OnnxAdd_7(*args, **kwargs)#
Version
name: Add (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Performs element-wise binary addition (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxAiOnnxMlArrayFeatureExtractor#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlArrayFeatureExtractor(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Select elements of the input tensor based on the indices passed.
The indices are applied to the last axes of the tensor.
Inputs
X (heterogeneous) - T: Data to be selected
Y (heterogeneous) - tensor(int64): The indices, based on 0 as the first index of any dimension.
Outputs
Z (heterogeneous) - T: Selected output data as an array
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64), tensor(string) ): The input must be a tensor of a numeric type or string. The output will be of the same tensor type.
OnnxAiOnnxMlArrayFeatureExtractor_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlArrayFeatureExtractor_1(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Select elements of the input tensor based on the indices passed.
The indices are applied to the last axes of the tensor.
Inputs
X (heterogeneous) - T: Data to be selected
Y (heterogeneous) - tensor(int64): The indices, based on 0 as the first index of any dimension.
Outputs
Z (heterogeneous) - T: Selected output data as an array
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64), tensor(string) ): The input must be a tensor of a numeric type or string. The output will be of the same tensor type.
OnnxAiOnnxMlBinarizer#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlBinarizer(*args, **kwargs)#
Version
name: Binarizer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Maps the values of the input tensor to either 0 or 1, element-wise, based on the outcome of a comparison against a threshold value.
Attributes
threshold: Values greater than this are mapped to 1, others to 0. Default value is
0.0.
Inputs
X (heterogeneous) - T: Data to be binarized
Outputs
Y (heterogeneous) - T: Binarized output data
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type. The output will be of the same tensor type.
OnnxAiOnnxMlBinarizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlBinarizer_1(*args, **kwargs)#
Version
name: Binarizer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Maps the values of the input tensor to either 0 or 1, element-wise, based on the outcome of a comparison against a threshold value.
Attributes
threshold: Values greater than this are mapped to 1, others to 0. Default value is
0.0.
Inputs
X (heterogeneous) - T: Data to be binarized
Outputs
Y (heterogeneous) - T: Binarized output data
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type. The output will be of the same tensor type.
OnnxAiOnnxMlCastMap#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlCastMap(*args, **kwargs)#
Version
name: CastMap (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Converts a map to a tensor. The map key must be an int64 and the values will be ordered in ascending order based on this key. The operator supports dense packing or sparse packing. If using sparse packing, the key cannot exceed the max_map-1 value.
Attributes
cast_to: A string indicating the desired element type of the output tensor, one of ‘TO_FLOAT’, ‘TO_STRING’, ‘TO_INT64’. Default value is
'TO_FLOAT'.map_form: Indicates whether to only output as many values as are in the input (dense), or position the input based on using the key of the map as the index of the output (sparse).<br>One of ‘DENSE’, ‘SPARSE’. Default value is
'DENSE'.max_map: If the value of map_form is ‘SPARSE,’ this attribute indicates the total length of the output tensor. Default value is
1.
Inputs
X (heterogeneous) - T1: The input map that is to be cast to a tensor
Outputs
Y (heterogeneous) - T2: A tensor representing the same data as the input map, ordered by their keys
Type Constraints
T1 in ( map(int64, float), map(int64, string) ): The input must be an integer map to either string or float.
T2 in ( tensor(float), tensor(int64), tensor(string) ): The output is a 1-D tensor of string, float, or integer.
OnnxAiOnnxMlCastMap_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlCastMap_1(*args, **kwargs)#
Version
name: CastMap (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Converts a map to a tensor. The map key must be an int64 and the values will be ordered in ascending order based on this key. The operator supports dense packing or sparse packing. If using sparse packing, the key cannot exceed the max_map-1 value.
Attributes
cast_to: A string indicating the desired element type of the output tensor, one of ‘TO_FLOAT’, ‘TO_STRING’, ‘TO_INT64’. Default value is
'TO_FLOAT'.map_form: Indicates whether to only output as many values as are in the input (dense), or position the input based on using the key of the map as the index of the output (sparse).<br>One of ‘DENSE’, ‘SPARSE’. Default value is
'DENSE'.max_map: If the value of map_form is ‘SPARSE,’ this attribute indicates the total length of the output tensor. Default value is
1.
Inputs
X (heterogeneous) - T1: The input map that is to be cast to a tensor
Outputs
Y (heterogeneous) - T2: A tensor representing the same data as the input map, ordered by their keys
Type Constraints
T1 in ( map(int64, float), map(int64, string) ): The input must be an integer map to either string or float.
T2 in ( tensor(float), tensor(int64), tensor(string) ): The output is a 1-D tensor of string, float, or integer.
OnnxAiOnnxMlCategoryMapper#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlCategoryMapper(*args, **kwargs)#
Version
name: CategoryMapper (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Converts strings to integers and vice versa.
Two sequences of equal length are used to map between integers and strings, with strings and integers at the same index detailing the mapping.
Each operator converts either integers to strings or strings to integers, depending on which default value attribute is provided. Only one default value attribute should be defined.
If the string default value is set, it will convert integers to strings. If the int default value is set, it will convert strings to integers.
Attributes
cats_int64s: The integers of the map. This sequence must be the same length as the ‘cats_strings’ sequence.
cats_strings: The strings of the map. This sequence must be the same length as the ‘cats_int64s’ sequence
default_int64: An integer to use when an input string value is not found in the map.<br>One and only one of the ‘default_*’ attributes must be defined. Default value is
-1.default_string: A string to use when an input integer value is not found in the map.<br>One and only one of the ‘default_*’ attributes must be defined. Default value is
'_Unused'.
Inputs
X (heterogeneous) - T1: Input data
Outputs
Y (heterogeneous) - T2: Output data. If strings are input, the output values are integers, and vice versa.
Type Constraints
T1 in ( tensor(int64), tensor(string) ): The input must be a tensor of strings or integers, either [N,C] or [C].
T2 in ( tensor(int64), tensor(string) ): The output is a tensor of strings or integers. Its shape will be the same as the input shape.
OnnxAiOnnxMlCategoryMapper_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlCategoryMapper_1(*args, **kwargs)#
Version
name: CategoryMapper (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Converts strings to integers and vice versa.
Two sequences of equal length are used to map between integers and strings, with strings and integers at the same index detailing the mapping.
Each operator converts either integers to strings or strings to integers, depending on which default value attribute is provided. Only one default value attribute should be defined.
If the string default value is set, it will convert integers to strings. If the int default value is set, it will convert strings to integers.
Attributes
cats_int64s: The integers of the map. This sequence must be the same length as the ‘cats_strings’ sequence.
cats_strings: The strings of the map. This sequence must be the same length as the ‘cats_int64s’ sequence
default_int64: An integer to use when an input string value is not found in the map.<br>One and only one of the ‘default_*’ attributes must be defined. Default value is
-1.default_string: A string to use when an input integer value is not found in the map.<br>One and only one of the ‘default_*’ attributes must be defined. Default value is
'_Unused'.
Inputs
X (heterogeneous) - T1: Input data
Outputs
Y (heterogeneous) - T2: Output data. If strings are input, the output values are integers, and vice versa.
Type Constraints
T1 in ( tensor(int64), tensor(string) ): The input must be a tensor of strings or integers, either [N,C] or [C].
T2 in ( tensor(int64), tensor(string) ): The output is a tensor of strings or integers. Its shape will be the same as the input shape.
OnnxAiOnnxMlDictVectorizer#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlDictVectorizer(*args, **kwargs)#
Version
name: DictVectorizer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Uses an index mapping to convert a dictionary to an array.
Given a dictionary, each key is looked up in the vocabulary attribute corresponding to the key type. The index into the vocabulary array at which the key is found is then used to index the output 1-D tensor ‘Y’ and insert into it the value found in the dictionary ‘X’.
The key type of the input map must correspond to the element type of the defined vocabulary attribute. Therefore, the output array will be equal in length to the index mapping vector parameter. All keys in the input dictionary must be present in the index mapping vector. For each item in the input dictionary, insert its value in the output array. Any keys not present in the input dictionary, will be zero in the output array.
For example: if the
string_vocabularyparameter is set to["a", "c", "b", "z"], then an input of{"a": 4, "c": 8}will produce an output of[4, 8, 0, 0].Attributes
int64_vocabulary: An integer vocabulary array.<br>One and only one of the vocabularies must be defined.
string_vocabulary: A string vocabulary array.<br>One and only one of the vocabularies must be defined.
Inputs
X (heterogeneous) - T1: A dictionary.
Outputs
Y (heterogeneous) - T2: A 1-D tensor holding values from the input dictionary.
Type Constraints
T1 in ( map(int64, double), map(int64, float), map(int64, string), map(string, double), map(string, float), map(string, int64) ): The input must be a map from strings or integers to either strings or a numeric type. The key and value types cannot be the same.
T2 in ( tensor(double), tensor(float), tensor(int64), tensor(string) ): The output will be a tensor of the value type of the input map. It’s shape will be [1,C], where C is the length of the input dictionary.
OnnxAiOnnxMlDictVectorizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlDictVectorizer_1(*args, **kwargs)#
Version
name: DictVectorizer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Uses an index mapping to convert a dictionary to an array.
Given a dictionary, each key is looked up in the vocabulary attribute corresponding to the key type. The index into the vocabulary array at which the key is found is then used to index the output 1-D tensor ‘Y’ and insert into it the value found in the dictionary ‘X’.
The key type of the input map must correspond to the element type of the defined vocabulary attribute. Therefore, the output array will be equal in length to the index mapping vector parameter. All keys in the input dictionary must be present in the index mapping vector. For each item in the input dictionary, insert its value in the output array. Any keys not present in the input dictionary, will be zero in the output array.
For example: if the
string_vocabularyparameter is set to["a", "c", "b", "z"], then an input of{"a": 4, "c": 8}will produce an output of[4, 8, 0, 0].Attributes
int64_vocabulary: An integer vocabulary array.<br>One and only one of the vocabularies must be defined.
string_vocabulary: A string vocabulary array.<br>One and only one of the vocabularies must be defined.
Inputs
X (heterogeneous) - T1: A dictionary.
Outputs
Y (heterogeneous) - T2: A 1-D tensor holding values from the input dictionary.
Type Constraints
T1 in ( map(int64, double), map(int64, float), map(int64, string), map(string, double), map(string, float), map(string, int64) ): The input must be a map from strings or integers to either strings or a numeric type. The key and value types cannot be the same.
T2 in ( tensor(double), tensor(float), tensor(int64), tensor(string) ): The output will be a tensor of the value type of the input map. It’s shape will be [1,C], where C is the length of the input dictionary.
OnnxAiOnnxMlFeatureVectorizer#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlFeatureVectorizer(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Concatenates input tensors into one continuous output.
All input shapes are 2-D and are concatenated along the second dimention. 1-D tensors are treated as [1,C]. Inputs are copied to the output maintaining the order of the input arguments.
All inputs must be integers or floats, while the output will be all floating point values.
Attributes
inputdimensions: The size of each input in the input list
Inputs
Between 1 and 2147483647 inputs.
X (variadic, heterogeneous) - T1: An ordered collection of tensors, all with the same element type.
Outputs
Y (heterogeneous) - tensor(float): The output array, elements ordered as the inputs.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
OnnxAiOnnxMlFeatureVectorizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlFeatureVectorizer_1(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Concatenates input tensors into one continuous output.
All input shapes are 2-D and are concatenated along the second dimention. 1-D tensors are treated as [1,C]. Inputs are copied to the output maintaining the order of the input arguments.
All inputs must be integers or floats, while the output will be all floating point values.
Attributes
inputdimensions: The size of each input in the input list
Inputs
Between 1 and 2147483647 inputs.
X (variadic, heterogeneous) - T1: An ordered collection of tensors, all with the same element type.
Outputs
Y (heterogeneous) - tensor(float): The output array, elements ordered as the inputs.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
OnnxAiOnnxMlImputer#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlImputer(*args, **kwargs)#
Version
name: Imputer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Replaces inputs that equal one value with another, leaving all other elements alone.
This operator is typically used to replace missing values in situations where they have a canonical representation, such as -1, 0, NaN, or some extreme value.
One and only one of imputed_value_floats or imputed_value_int64s should be defined – floats if the input tensor holds floats, integers if the input tensor holds integers. The imputed values must all fit within the width of the tensor element type. One and only one of the replaced_value_float or replaced_value_int64 should be defined, which one depends on whether floats or integers are being processed.
The imputed_value attribute length can be 1 element, or it can have one element per input feature. In other words, if the input tensor has the shape [*,F], then the length of the attribute array may be 1 or F. If it is 1, then it is broadcast along the last dimension and applied to each feature.
Attributes
imputed_value_floats: Value(s) to change to
imputed_value_int64s: Value(s) to change to.
replaced_value_float: A value that needs replacing. Default value is
0.0.replaced_value_int64: A value that needs replacing. Default value is
0.
Inputs
X (heterogeneous) - T: Data to be processed.
Outputs
Y (heterogeneous) - T: Imputed output data
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type, either [N,C] or [C]. The output type will be of the same tensor type and shape.
OnnxAiOnnxMlImputer_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlImputer_1(*args, **kwargs)#
Version
name: Imputer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Replaces inputs that equal one value with another, leaving all other elements alone.
This operator is typically used to replace missing values in situations where they have a canonical representation, such as -1, 0, NaN, or some extreme value.
One and only one of imputed_value_floats or imputed_value_int64s should be defined – floats if the input tensor holds floats, integers if the input tensor holds integers. The imputed values must all fit within the width of the tensor element type. One and only one of the replaced_value_float or replaced_value_int64 should be defined, which one depends on whether floats or integers are being processed.
The imputed_value attribute length can be 1 element, or it can have one element per input feature. In other words, if the input tensor has the shape [*,F], then the length of the attribute array may be 1 or F. If it is 1, then it is broadcast along the last dimension and applied to each feature.
Attributes
imputed_value_floats: Value(s) to change to
imputed_value_int64s: Value(s) to change to.
replaced_value_float: A value that needs replacing. Default value is
0.0.replaced_value_int64: A value that needs replacing. Default value is
0.
Inputs
X (heterogeneous) - T: Data to be processed.
Outputs
Y (heterogeneous) - T: Imputed output data
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type, either [N,C] or [C]. The output type will be of the same tensor type and shape.
OnnxAiOnnxMlLabelEncoder#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLabelEncoder(*args, **kwargs)#
Version
name: LabelEncoder (GitHub)
domain: ai.onnx.ml
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2 of domain ai.onnx.ml.
Summary
Maps each element in the input tensor to another value.
The mapping is determined by the two parallel attributes, ‘keys_*’ and ‘values_*’ attribute. The i-th value in the specified ‘keys_*’ attribute would be mapped to the i-th value in the specified ‘values_*’ attribute. It implies that input’s element type and the element type of the specified ‘keys_*’ should be identical while the output type is identical to the specified ‘values_*’ attribute. If an input element can not be found in the specified ‘keys_*’ attribute, the ‘default_*’ that matches the specified ‘values_*’ attribute may be used as its output value.
Let’s consider an example which maps a string tensor to an integer tensor. Assume and ‘keys_strings’ is [“Amy”, “Sally”], ‘values_int64s’ is [5, 6], and ‘default_int64’ is ‘-1’. The input [“Dori”, “Amy”, “Amy”, “Sally”, “Sally”] would be mapped to [-1, 5, 5, 6, 6].
Since this operator is an one-to-one mapping, its input and output shapes are the same. Notice that only one of ‘keys_*’/’values_*’ can be set.
For key look-up, bit-wise comparison is used so even a float NaN can be mapped to a value in ‘values_*’ attribute.
Attributes
default_float: A float. Default value is
-0.0.default_int64: An integer. Default value is
-1.default_string: A string. Default value is
'_Unused'.keys_floats: A list of floats.
keys_int64s: A list of ints.
keys_strings: A list of strings. One and only one of ‘keys_*’s should be set.
values_floats: A list of floats.
values_int64s: A list of ints.
values_strings: A list of strings. One and only one of ‘value_*’s should be set.
Inputs
X (heterogeneous) - T1: Input data. It can be either tensor or scalar.
Outputs
Y (heterogeneous) - T2: Output data.
Type Constraints
T1 in ( tensor(float), tensor(int64), tensor(string) ): The input type is a tensor of any shape.
T2 in ( tensor(float), tensor(int64), tensor(string) ): Output type is determined by the specified ‘values_*’ attribute.
OnnxAiOnnxMlLabelEncoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLabelEncoder_1(*args, **kwargs)#
Version
name: LabelEncoder (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Converts strings to integers and vice versa.
If the string default value is set, it will convert integers to strings. If the int default value is set, it will convert strings to integers.
Each operator converts either integers to strings or strings to integers, depending on which default value attribute is provided. Only one default value attribute should be defined.
When converting from integers to strings, the string is fetched from the ‘classes_strings’ list, by simple indexing.
When converting from strings to integers, the string is looked up in the list and the index at which it is found is used as the converted value.
Attributes
classes_strings: A list of labels.
default_int64: An integer to use when an input string value is not found in the map.<br>One and only one of the ‘default_*’ attributes must be defined. Default value is
-1.default_string: A string to use when an input integer value is not found in the map.<br>One and only one of the ‘default_*’ attributes must be defined. Default value is
'_Unused'.
Inputs
X (heterogeneous) - T1: Input data.
Outputs
Y (heterogeneous) - T2: Output data. If strings are input, the output values are integers, and vice versa.
Type Constraints
T1 in ( tensor(int64), tensor(string) ): The input type must be a tensor of integers or strings, of any shape.
T2 in ( tensor(int64), tensor(string) ): The output type will be a tensor of strings or integers, and will have the same shape as the input.
OnnxAiOnnxMlLabelEncoder_2#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLabelEncoder_2(*args, **kwargs)#
Version
name: LabelEncoder (GitHub)
domain: ai.onnx.ml
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2 of domain ai.onnx.ml.
Summary
Maps each element in the input tensor to another value.
The mapping is determined by the two parallel attributes, ‘keys_*’ and ‘values_*’ attribute. The i-th value in the specified ‘keys_*’ attribute would be mapped to the i-th value in the specified ‘values_*’ attribute. It implies that input’s element type and the element type of the specified ‘keys_*’ should be identical while the output type is identical to the specified ‘values_*’ attribute. If an input element can not be found in the specified ‘keys_*’ attribute, the ‘default_*’ that matches the specified ‘values_*’ attribute may be used as its output value.
Let’s consider an example which maps a string tensor to an integer tensor. Assume and ‘keys_strings’ is [“Amy”, “Sally”], ‘values_int64s’ is [5, 6], and ‘default_int64’ is ‘-1’. The input [“Dori”, “Amy”, “Amy”, “Sally”, “Sally”] would be mapped to [-1, 5, 5, 6, 6].
Since this operator is an one-to-one mapping, its input and output shapes are the same. Notice that only one of ‘keys_*’/’values_*’ can be set.
For key look-up, bit-wise comparison is used so even a float NaN can be mapped to a value in ‘values_*’ attribute.
Attributes
default_float: A float. Default value is
-0.0.default_int64: An integer. Default value is
-1.default_string: A string. Default value is
'_Unused'.keys_floats: A list of floats.
keys_int64s: A list of ints.
keys_strings: A list of strings. One and only one of ‘keys_*’s should be set.
values_floats: A list of floats.
values_int64s: A list of ints.
values_strings: A list of strings. One and only one of ‘value_*’s should be set.
Inputs
X (heterogeneous) - T1: Input data. It can be either tensor or scalar.
Outputs
Y (heterogeneous) - T2: Output data.
Type Constraints
T1 in ( tensor(float), tensor(int64), tensor(string) ): The input type is a tensor of any shape.
T2 in ( tensor(float), tensor(int64), tensor(string) ): Output type is determined by the specified ‘values_*’ attribute.
OnnxAiOnnxMlLinearClassifier#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLinearClassifier(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Linear classifier
Attributes
classlabels_ints: Class labels when using integer labels. One and only one ‘classlabels’ attribute must be defined.
classlabels_strings: Class labels when using string labels. One and only one ‘classlabels’ attribute must be defined.
coefficients (required): A collection of weights of the model(s).
intercepts: A collection of intercepts.
multi_class: Indicates whether to do OvR or multinomial (0=OvR is the default). Default value is
0.post_transform: Indicates the transform to apply to the scores vector.<br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.
Inputs
X (heterogeneous) - T1: Data to be classified.
Outputs
Y (heterogeneous) - T2: Classification outputs (one class per example).
Z (heterogeneous) - tensor(float): Classification scores ([N,E] - one score for each class and example
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type, and of shape [N,C] or [C]. In the latter case, it will be treated as [1,C]
T2 in ( tensor(int64), tensor(string) ): The output will be a tensor of strings or integers.
OnnxAiOnnxMlLinearClassifier_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLinearClassifier_1(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Linear classifier
Attributes
classlabels_ints: Class labels when using integer labels. One and only one ‘classlabels’ attribute must be defined.
classlabels_strings: Class labels when using string labels. One and only one ‘classlabels’ attribute must be defined.
coefficients (required): A collection of weights of the model(s).
intercepts: A collection of intercepts.
multi_class: Indicates whether to do OvR or multinomial (0=OvR is the default). Default value is
0.post_transform: Indicates the transform to apply to the scores vector.<br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.
Inputs
X (heterogeneous) - T1: Data to be classified.
Outputs
Y (heterogeneous) - T2: Classification outputs (one class per example).
Z (heterogeneous) - tensor(float): Classification scores ([N,E] - one score for each class and example
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type, and of shape [N,C] or [C]. In the latter case, it will be treated as [1,C]
T2 in ( tensor(int64), tensor(string) ): The output will be a tensor of strings or integers.
OnnxAiOnnxMlLinearRegressor#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLinearRegressor(*args, **kwargs)#
Version
name: LinearRegressor (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Generalized linear regression evaluation.
If targets is set to 1 (default) then univariate regression is performed.
If targets is set to M then M sets of coefficients must be passed in as a sequence and M results will be output for each input n in N.
The coefficients array is of length n, and the coefficients for each target are contiguous. Intercepts are optional but if provided must match the number of targets.
Attributes
coefficients: Weights of the model(s).
intercepts: Weights of the intercepts, if used.
post_transform: Indicates the transform to apply to the regression output vector.<br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.targets: The total number of regression targets, 1 if not defined. Default value is
1.
Inputs
X (heterogeneous) - T: Data to be regressed.
Outputs
Y (heterogeneous) - tensor(float): Regression outputs (one per target, per example).
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlLinearRegressor_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlLinearRegressor_1(*args, **kwargs)#
Version
name: LinearRegressor (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Generalized linear regression evaluation.
If targets is set to 1 (default) then univariate regression is performed.
If targets is set to M then M sets of coefficients must be passed in as a sequence and M results will be output for each input n in N.
The coefficients array is of length n, and the coefficients for each target are contiguous. Intercepts are optional but if provided must match the number of targets.
Attributes
coefficients: Weights of the model(s).
intercepts: Weights of the intercepts, if used.
post_transform: Indicates the transform to apply to the regression output vector.<br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.targets: The total number of regression targets, 1 if not defined. Default value is
1.
Inputs
X (heterogeneous) - T: Data to be regressed.
Outputs
Y (heterogeneous) - tensor(float): Regression outputs (one per target, per example).
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlNormalizer#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlNormalizer(*args, **kwargs)#
Version
name: Normalizer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Normalize the input. There are three normalization modes, which have the corresponding formulas, defined using element-wise infix operators ‘/’ and ‘^’ and tensor-wide functions ‘max’ and ‘sum’:
Max: Y = X / max(X)
L1: Y = X / sum(X)
L2: Y = sqrt(X^2 / sum(X^2)}
In all modes, if the divisor is zero, Y == X.
For batches, that is, [N,C] tensors, normalization is done along the C axis. In other words, each row of the batch is normalized independently.
Attributes
norm: One of ‘MAX,’ ‘L1,’ ‘L2’ Default value is
'MAX'.
Inputs
X (heterogeneous) - T: Data to be encoded, a tensor of shape [N,C] or [C]
Outputs
Y (heterogeneous) - tensor(float): Encoded output data
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlNormalizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlNormalizer_1(*args, **kwargs)#
Version
name: Normalizer (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Normalize the input. There are three normalization modes, which have the corresponding formulas, defined using element-wise infix operators ‘/’ and ‘^’ and tensor-wide functions ‘max’ and ‘sum’:
Max: Y = X / max(X)
L1: Y = X / sum(X)
L2: Y = sqrt(X^2 / sum(X^2)}
In all modes, if the divisor is zero, Y == X.
For batches, that is, [N,C] tensors, normalization is done along the C axis. In other words, each row of the batch is normalized independently.
Attributes
norm: One of ‘MAX,’ ‘L1,’ ‘L2’ Default value is
'MAX'.
Inputs
X (heterogeneous) - T: Data to be encoded, a tensor of shape [N,C] or [C]
Outputs
Y (heterogeneous) - tensor(float): Encoded output data
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlOneHotEncoder#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlOneHotEncoder(*args, **kwargs)#
Version
name: OneHotEncoder (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Replace each input element with an array of ones and zeros, where a single one is placed at the index of the category that was passed in. The total category count will determine the size of the extra dimension of the output array Y.
For example, if we pass a tensor with a single value of 4, and a category count of 8, the output will be a tensor with
[0,0,0,0,1,0,0,0].This operator assumes every input feature is from the same set of categories.
If the input is a tensor of float, int32, or double, the data will be cast to integers and the cats_int64s category list will be used for the lookups.
Attributes
cats_int64s: List of categories, ints.<br>One and only one of the ‘cats_*’ attributes must be defined.
cats_strings: List of categories, strings.<br>One and only one of the ‘cats_*’ attributes must be defined.
zeros: If true and category is not present, will return all zeros; if false and a category if not found, the operator will fail. Default value is
1.
Inputs
X (heterogeneous) - T: Data to be encoded.
Outputs
Y (heterogeneous) - tensor(float): Encoded output data, having one more dimension than X.
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64), tensor(string) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlOneHotEncoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlOneHotEncoder_1(*args, **kwargs)#
Version
name: OneHotEncoder (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Replace each input element with an array of ones and zeros, where a single one is placed at the index of the category that was passed in. The total category count will determine the size of the extra dimension of the output array Y.
For example, if we pass a tensor with a single value of 4, and a category count of 8, the output will be a tensor with
[0,0,0,0,1,0,0,0].This operator assumes every input feature is from the same set of categories.
If the input is a tensor of float, int32, or double, the data will be cast to integers and the cats_int64s category list will be used for the lookups.
Attributes
cats_int64s: List of categories, ints.<br>One and only one of the ‘cats_*’ attributes must be defined.
cats_strings: List of categories, strings.<br>One and only one of the ‘cats_*’ attributes must be defined.
zeros: If true and category is not present, will return all zeros; if false and a category if not found, the operator will fail. Default value is
1.
Inputs
X (heterogeneous) - T: Data to be encoded.
Outputs
Y (heterogeneous) - tensor(float): Encoded output data, having one more dimension than X.
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64), tensor(string) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlSVMClassifier#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlSVMClassifier(*args, **kwargs)#
Version
name: SVMClassifier (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Support Vector Machine classifier
Attributes
classlabels_ints: Class labels if using integer labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: Class labels if using string labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
coefficients:
kernel_params: List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
kernel_type: The kernel type, one of ‘LINEAR,’ ‘POLY,’ ‘RBF,’ ‘SIGMOID’. Default value is
'LINEAR'.post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.prob_a: First set of probability coefficients.
prob_b: Second set of probability coefficients. This array must be same size as prob_a.<br>If these are provided then output Z are probability estimates, otherwise they are raw scores.
rho:
support_vectors:
vectors_per_class:
Inputs
X (heterogeneous) - T1: Data to be classified.
Outputs
Y (heterogeneous) - T2: Classification outputs (one class per example).
Z (heterogeneous) - tensor(float): Class scores (one per class per example), if prob_a and prob_b are provided they are probabilities for each class, otherwise they are raw scores.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type, either [C] or [N,C].
T2 in ( tensor(int64), tensor(string) ): The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used. Its size will match the bactch size of the input.
OnnxAiOnnxMlSVMClassifier_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlSVMClassifier_1(*args, **kwargs)#
Version
name: SVMClassifier (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Support Vector Machine classifier
Attributes
classlabels_ints: Class labels if using integer labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: Class labels if using string labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
coefficients:
kernel_params: List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
kernel_type: The kernel type, one of ‘LINEAR,’ ‘POLY,’ ‘RBF,’ ‘SIGMOID’. Default value is
'LINEAR'.post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.prob_a: First set of probability coefficients.
prob_b: Second set of probability coefficients. This array must be same size as prob_a.<br>If these are provided then output Z are probability estimates, otherwise they are raw scores.
rho:
support_vectors:
vectors_per_class:
Inputs
X (heterogeneous) - T1: Data to be classified.
Outputs
Y (heterogeneous) - T2: Classification outputs (one class per example).
Z (heterogeneous) - tensor(float): Class scores (one per class per example), if prob_a and prob_b are provided they are probabilities for each class, otherwise they are raw scores.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type, either [C] or [N,C].
T2 in ( tensor(int64), tensor(string) ): The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used. Its size will match the bactch size of the input.
OnnxAiOnnxMlSVMRegressor#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlSVMRegressor(*args, **kwargs)#
Version
name: SVMRegressor (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Support Vector Machine regression prediction and one-class SVM anomaly detection.
Attributes
coefficients: Support vector coefficients.
kernel_params: List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
kernel_type: The kernel type, one of ‘LINEAR,’ ‘POLY,’ ‘RBF,’ ‘SIGMOID’. Default value is
'LINEAR'.n_supports: The number of support vectors. Default value is
0.one_class: Flag indicating whether the regression is a one-class SVM or not. Default value is
0.post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT.’ Default value is
'NONE'.rho:
support_vectors: Chosen support vectors
Inputs
X (heterogeneous) - T: Data to be regressed.
Outputs
Y (heterogeneous) - tensor(float): Regression outputs (one score per target per example).
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type, either [C] or [N,C].
OnnxAiOnnxMlSVMRegressor_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlSVMRegressor_1(*args, **kwargs)#
Version
name: SVMRegressor (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Support Vector Machine regression prediction and one-class SVM anomaly detection.
Attributes
coefficients: Support vector coefficients.
kernel_params: List of 3 elements containing gamma, coef0, and degree, in that order. Zero if unused for the kernel.
kernel_type: The kernel type, one of ‘LINEAR,’ ‘POLY,’ ‘RBF,’ ‘SIGMOID’. Default value is
'LINEAR'.n_supports: The number of support vectors. Default value is
0.one_class: Flag indicating whether the regression is a one-class SVM or not. Default value is
0.post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT.’ Default value is
'NONE'.rho:
support_vectors: Chosen support vectors
Inputs
X (heterogeneous) - T: Data to be regressed.
Outputs
Y (heterogeneous) - tensor(float): Regression outputs (one score per target per example).
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type, either [C] or [N,C].
OnnxAiOnnxMlScaler#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlScaler(*args, **kwargs)#
Version
name: Scaler (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Rescale input data, for example to standardize features by removing the mean and scaling to unit variance.
Attributes
offset: First, offset by this.<br>Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.
scale: Second, multiply by this.<br>Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.<br>Must be same length as ‘offset’
Inputs
X (heterogeneous) - T: Data to be scaled.
Outputs
Y (heterogeneous) - tensor(float): Scaled output data.
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlScaler_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlScaler_1(*args, **kwargs)#
Version
name: Scaler (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Rescale input data, for example to standardize features by removing the mean and scaling to unit variance.
Attributes
offset: First, offset by this.<br>Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.
scale: Second, multiply by this.<br>Can be length of features in an [N,F] tensor or length 1, in which case it applies to all features, regardless of dimension count.<br>Must be same length as ‘offset’
Inputs
X (heterogeneous) - T: Data to be scaled.
Outputs
Y (heterogeneous) - tensor(float): Scaled output data.
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input must be a tensor of a numeric type.
OnnxAiOnnxMlTreeEnsembleClassifier#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlTreeEnsembleClassifier(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 3
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 3 of domain ai.onnx.ml.
Summary
Tree Ensemble classifier. Returns the top class for each of N inputs.
The attributes named ‘nodes_X’ form a sequence of tuples, associated by index into the sequences, which must all be of equal length. These tuples define the nodes.
Similarly, all fields prefixed with ‘class_’ are tuples of votes at the leaves. A leaf may have multiple votes, where each vote is weighted by the associated class_weights index.
One and only one of classlabels_strings or classlabels_int64s will be defined. The class_ids are indices into this list. All fields ending with <i>_as_tensor</i> can be used instead of the same parameter without the suffix if the element type is double and not float.
Attributes
base_values: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
base_values_as_tensor: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
class_ids: The index of the class list that each weight is for.
class_nodeids: node id that this weight is for.
class_treeids: The id of the tree that this node is in.
class_weights: The weight for the class in class_id.
class_weights_as_tensor: The weight for the class in class_id.
classlabels_int64s: Class labels if using integer labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: Class labels if using string labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
nodes_falsenodeids: Child node if expression is false.
nodes_featureids: Feature id for each node.
nodes_hitrates: Popularity of each node, used for performance and may be omitted.
nodes_hitrates_as_tensor: Popularity of each node, used for performance and may be omitted.
nodes_missing_value_tracks_true: For each node, define what to do in the presence of a missing value: if a value is missing (NaN), use the ‘true’ or ‘false’ branch based on the value in this array.<br>This attribute may be left undefined, and the defalt value is false (0) for all nodes.
nodes_modes: The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.<br>One of ‘BRANCH_LEQ’, ‘BRANCH_LT’, ‘BRANCH_GTE’, ‘BRANCH_GT’, ‘BRANCH_EQ’, ‘BRANCH_NEQ’, ‘LEAF’
nodes_nodeids: Node id for each node. Ids may restart at zero for each tree, but it not required to.
nodes_treeids: Tree id for each node.
nodes_truenodeids: Child node if expression is true.
nodes_values: Thresholds to do the splitting on for each node.
nodes_values_as_tensor: Thresholds to do the splitting on for each node.
post_transform: Indicates the transform to apply to the score. <br> One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT.’ Default value is
'NONE'.
Inputs
X (heterogeneous) - T1: Input of shape [N,F]
Outputs
Y (heterogeneous) - T2: N, Top class for each point
Z (heterogeneous) - tensor(float): The class score for each class, for each point, a tensor of shape [N,E].
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
T2 in ( tensor(int64), tensor(string) ): The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used.
OnnxAiOnnxMlTreeEnsembleClassifier_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlTreeEnsembleClassifier_1(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Tree Ensemble classifier. Returns the top class for each of N inputs.
The attributes named ‘nodes_X’ form a sequence of tuples, associated by index into the sequences, which must all be of equal length. These tuples define the nodes.
Similarly, all fields prefixed with ‘class_’ are tuples of votes at the leaves. A leaf may have multiple votes, where each vote is weighted by the associated class_weights index.
One and only one of classlabels_strings or classlabels_int64s will be defined. The class_ids are indices into this list.
Attributes
base_values: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
class_ids: The index of the class list that each weight is for.
class_nodeids: node id that this weight is for.
class_treeids: The id of the tree that this node is in.
class_weights: The weight for the class in class_id.
classlabels_int64s: Class labels if using integer labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: Class labels if using string labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
nodes_falsenodeids: Child node if expression is false.
nodes_featureids: Feature id for each node.
nodes_hitrates: Popularity of each node, used for performance and may be omitted.
nodes_missing_value_tracks_true: For each node, define what to do in the presence of a missing value: if a value is missing (NaN), use the ‘true’ or ‘false’ branch based on the value in this array.<br>This attribute may be left undefined, and the defalt value is false (0) for all nodes.
nodes_modes: The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.<br>One of ‘BRANCH_LEQ’, ‘BRANCH_LT’, ‘BRANCH_GTE’, ‘BRANCH_GT’, ‘BRANCH_EQ’, ‘BRANCH_NEQ’, ‘LEAF’
nodes_nodeids: Node id for each node. Ids may restart at zero for each tree, but it not required to.
nodes_treeids: Tree id for each node.
nodes_truenodeids: Child node if expression is true.
nodes_values: Thresholds to do the splitting on for each node.
post_transform: Indicates the transform to apply to the score. <br> One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT.’ Default value is
'NONE'.
Inputs
X (heterogeneous) - T1: Input of shape [N,F]
Outputs
Y (heterogeneous) - T2: N, Top class for each point
Z (heterogeneous) - tensor(float): The class score for each class, for each point, a tensor of shape [N,E].
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
T2 in ( tensor(int64), tensor(string) ): The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used.
OnnxAiOnnxMlTreeEnsembleClassifier_3#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlTreeEnsembleClassifier_3(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 3
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 3 of domain ai.onnx.ml.
Summary
Tree Ensemble classifier. Returns the top class for each of N inputs.
The attributes named ‘nodes_X’ form a sequence of tuples, associated by index into the sequences, which must all be of equal length. These tuples define the nodes.
Similarly, all fields prefixed with ‘class_’ are tuples of votes at the leaves. A leaf may have multiple votes, where each vote is weighted by the associated class_weights index.
One and only one of classlabels_strings or classlabels_int64s will be defined. The class_ids are indices into this list. All fields ending with <i>_as_tensor</i> can be used instead of the same parameter without the suffix if the element type is double and not float.
Attributes
base_values: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
base_values_as_tensor: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
class_ids: The index of the class list that each weight is for.
class_nodeids: node id that this weight is for.
class_treeids: The id of the tree that this node is in.
class_weights: The weight for the class in class_id.
class_weights_as_tensor: The weight for the class in class_id.
classlabels_int64s: Class labels if using integer labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: Class labels if using string labels.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
nodes_falsenodeids: Child node if expression is false.
nodes_featureids: Feature id for each node.
nodes_hitrates: Popularity of each node, used for performance and may be omitted.
nodes_hitrates_as_tensor: Popularity of each node, used for performance and may be omitted.
nodes_missing_value_tracks_true: For each node, define what to do in the presence of a missing value: if a value is missing (NaN), use the ‘true’ or ‘false’ branch based on the value in this array.<br>This attribute may be left undefined, and the defalt value is false (0) for all nodes.
nodes_modes: The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.<br>One of ‘BRANCH_LEQ’, ‘BRANCH_LT’, ‘BRANCH_GTE’, ‘BRANCH_GT’, ‘BRANCH_EQ’, ‘BRANCH_NEQ’, ‘LEAF’
nodes_nodeids: Node id for each node. Ids may restart at zero for each tree, but it not required to.
nodes_treeids: Tree id for each node.
nodes_truenodeids: Child node if expression is true.
nodes_values: Thresholds to do the splitting on for each node.
nodes_values_as_tensor: Thresholds to do the splitting on for each node.
post_transform: Indicates the transform to apply to the score. <br> One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT.’ Default value is
'NONE'.
Inputs
X (heterogeneous) - T1: Input of shape [N,F]
Outputs
Y (heterogeneous) - T2: N, Top class for each point
Z (heterogeneous) - tensor(float): The class score for each class, for each point, a tensor of shape [N,E].
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
T2 in ( tensor(int64), tensor(string) ): The output type will be a tensor of strings or integers, depending on which of the classlabels_* attributes is used.
OnnxAiOnnxMlTreeEnsembleRegressor#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlTreeEnsembleRegressor(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 3
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 3 of domain ai.onnx.ml.
Summary
Tree Ensemble regressor. Returns the regressed values for each input in N.
All args with nodes_ are fields of a tuple of tree nodes, and it is assumed they are the same length, and an index i will decode the tuple across these inputs. Each node id can appear only once for each tree id.
All fields prefixed with target_ are tuples of votes at the leaves.
A leaf may have multiple votes, where each vote is weighted by the associated target_weights index.
All fields ending with <i>_as_tensor</i> can be used instead of the same parameter without the suffix if the element type is double and not float. All trees must have their node ids start at 0 and increment by 1.
Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF
Attributes
aggregate_function: Defines how to aggregate leaf values within a target. <br>One of ‘AVERAGE,’ ‘SUM,’ ‘MIN,’ ‘MAX.’ Default value is
'SUM'.base_values: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
base_values_as_tensor: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
n_targets: The total number of targets.
nodes_falsenodeids: Child node if expression is false
nodes_featureids: Feature id for each node.
nodes_hitrates: Popularity of each node, used for performance and may be omitted.
nodes_hitrates_as_tensor: Popularity of each node, used for performance and may be omitted.
nodes_missing_value_tracks_true: For each node, define what to do in the presence of a NaN: use the ‘true’ (if the attribute value is 1) or ‘false’ (if the attribute value is 0) branch based on the value in this array.<br>This attribute may be left undefined and the defalt value is false (0) for all nodes.
nodes_modes: The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.<br>One of ‘BRANCH_LEQ’, ‘BRANCH_LT’, ‘BRANCH_GTE’, ‘BRANCH_GT’, ‘BRANCH_EQ’, ‘BRANCH_NEQ’, ‘LEAF’
nodes_nodeids: Node id for each node. Node ids must restart at zero for each tree and increase sequentially.
nodes_treeids: Tree id for each node.
nodes_truenodeids: Child node if expression is true
nodes_values: Thresholds to do the splitting on for each node.
nodes_values_as_tensor: Thresholds to do the splitting on for each node.
post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.target_ids: The index of the target that each weight is for
target_nodeids: The node id of each weight
target_treeids: The id of the tree that each node is in.
target_weights: The weight for each target
target_weights_as_tensor: The weight for each target
Inputs
X (heterogeneous) - T: Input of shape [N,F]
Outputs
Y (heterogeneous) - tensor(float): N classes
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
OnnxAiOnnxMlTreeEnsembleRegressor_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlTreeEnsembleRegressor_1(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Tree Ensemble regressor. Returns the regressed values for each input in N.
All args with nodes_ are fields of a tuple of tree nodes, and it is assumed they are the same length, and an index i will decode the tuple across these inputs. Each node id can appear only once for each tree id.
All fields prefixed with target_ are tuples of votes at the leaves.
A leaf may have multiple votes, where each vote is weighted by the associated target_weights index.
All trees must have their node ids start at 0 and increment by 1.
Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF
Attributes
aggregate_function: Defines how to aggregate leaf values within a target. <br>One of ‘AVERAGE,’ ‘SUM,’ ‘MIN,’ ‘MAX.’ Default value is
'SUM'.base_values: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
n_targets: The total number of targets.
nodes_falsenodeids: Child node if expression is false
nodes_featureids: Feature id for each node.
nodes_hitrates: Popularity of each node, used for performance and may be omitted.
nodes_missing_value_tracks_true: For each node, define what to do in the presence of a NaN: use the ‘true’ (if the attribute value is 1) or ‘false’ (if the attribute value is 0) branch based on the value in this array.<br>This attribute may be left undefined and the defalt value is false (0) for all nodes.
nodes_modes: The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.<br>One of ‘BRANCH_LEQ’, ‘BRANCH_LT’, ‘BRANCH_GTE’, ‘BRANCH_GT’, ‘BRANCH_EQ’, ‘BRANCH_NEQ’, ‘LEAF’
nodes_nodeids: Node id for each node. Node ids must restart at zero for each tree and increase sequentially.
nodes_treeids: Tree id for each node.
nodes_truenodeids: Child node if expression is true
nodes_values: Thresholds to do the splitting on for each node.
post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.target_ids: The index of the target that each weight is for
target_nodeids: The node id of each weight
target_treeids: The id of the tree that each node is in.
target_weights: The weight for each target
Inputs
X (heterogeneous) - T: Input of shape [N,F]
Outputs
Y (heterogeneous) - tensor(float): N classes
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
OnnxAiOnnxMlTreeEnsembleRegressor_3#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlTreeEnsembleRegressor_3(*args, **kwargs)#
Version
domain: ai.onnx.ml
since_version: 3
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 3 of domain ai.onnx.ml.
Summary
Tree Ensemble regressor. Returns the regressed values for each input in N.
All args with nodes_ are fields of a tuple of tree nodes, and it is assumed they are the same length, and an index i will decode the tuple across these inputs. Each node id can appear only once for each tree id.
All fields prefixed with target_ are tuples of votes at the leaves.
A leaf may have multiple votes, where each vote is weighted by the associated target_weights index.
All fields ending with <i>_as_tensor</i> can be used instead of the same parameter without the suffix if the element type is double and not float. All trees must have their node ids start at 0 and increment by 1.
Mode enum is BRANCH_LEQ, BRANCH_LT, BRANCH_GTE, BRANCH_GT, BRANCH_EQ, BRANCH_NEQ, LEAF
Attributes
aggregate_function: Defines how to aggregate leaf values within a target. <br>One of ‘AVERAGE,’ ‘SUM,’ ‘MIN,’ ‘MAX.’ Default value is
'SUM'.base_values: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
base_values_as_tensor: Base values for classification, added to final class score; the size must be the same as the classes or can be left unassigned (assumed 0)
n_targets: The total number of targets.
nodes_falsenodeids: Child node if expression is false
nodes_featureids: Feature id for each node.
nodes_hitrates: Popularity of each node, used for performance and may be omitted.
nodes_hitrates_as_tensor: Popularity of each node, used for performance and may be omitted.
nodes_missing_value_tracks_true: For each node, define what to do in the presence of a NaN: use the ‘true’ (if the attribute value is 1) or ‘false’ (if the attribute value is 0) branch based on the value in this array.<br>This attribute may be left undefined and the defalt value is false (0) for all nodes.
nodes_modes: The node kind, that is, the comparison to make at the node. There is no comparison to make at a leaf node.<br>One of ‘BRANCH_LEQ’, ‘BRANCH_LT’, ‘BRANCH_GTE’, ‘BRANCH_GT’, ‘BRANCH_EQ’, ‘BRANCH_NEQ’, ‘LEAF’
nodes_nodeids: Node id for each node. Node ids must restart at zero for each tree and increase sequentially.
nodes_treeids: Tree id for each node.
nodes_truenodeids: Child node if expression is true
nodes_values: Thresholds to do the splitting on for each node.
nodes_values_as_tensor: Thresholds to do the splitting on for each node.
post_transform: Indicates the transform to apply to the score. <br>One of ‘NONE,’ ‘SOFTMAX,’ ‘LOGISTIC,’ ‘SOFTMAX_ZERO,’ or ‘PROBIT’ Default value is
'NONE'.target_ids: The index of the target that each weight is for
target_nodeids: The node id of each weight
target_treeids: The id of the tree that each node is in.
target_weights: The weight for each target
target_weights_as_tensor: The weight for each target
Inputs
X (heterogeneous) - T: Input of shape [N,F]
Outputs
Y (heterogeneous) - tensor(float): N classes
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64) ): The input type must be a tensor of a numeric type.
OnnxAiOnnxMlZipMap#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlZipMap(*args, **kwargs)#
Version
name: ZipMap (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Creates a map from the input and the attributes.
The values are provided by the input tensor, while the keys are specified by the attributes. Must provide keys in either classlabels_strings or classlabels_int64s (but not both).
The columns of the tensor correspond one-by-one to the keys specified by the attributes. There must be as many columns as keys.
Attributes
classlabels_int64s: The keys when using int keys.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: The keys when using string keys.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
Inputs
X (heterogeneous) - tensor(float): The input values
Outputs
Z (heterogeneous) - T: The output map
Type Constraints
T in ( seq(map(int64, float)), seq(map(string, float)) ): The output will be a sequence of string or integer maps to float.
OnnxAiOnnxMlZipMap_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxMlZipMap_1(*args, **kwargs)#
Version
name: ZipMap (GitHub)
domain: ai.onnx.ml
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.ml.
Summary
Creates a map from the input and the attributes.
The values are provided by the input tensor, while the keys are specified by the attributes. Must provide keys in either classlabels_strings or classlabels_int64s (but not both).
The columns of the tensor correspond one-by-one to the keys specified by the attributes. There must be as many columns as keys.
Attributes
classlabels_int64s: The keys when using int keys.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
classlabels_strings: The keys when using string keys.<br>One and only one of the ‘classlabels_*’ attributes must be defined.
Inputs
X (heterogeneous) - tensor(float): The input values
Outputs
Z (heterogeneous) - T: The output map
Type Constraints
T in ( seq(map(int64, float)), seq(map(string, float)) ): The output will be a sequence of string or integer maps to float.
OnnxAiOnnxPreviewTrainingAdagrad#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingAdagrad(*args, **kwargs)#
Version
name: Adagrad (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Compute one iteration of ADAGRAD, a stochastic gradient based optimization algorithm. This operator can conduct the optimization of multiple tensor variables.
Let’s define the behavior of this operator. As you can imagine, ADAGRAD requires some parameters:
The initial learning-rate “R”.
The update count “T”. That is, the number of training iterations conducted.
A L2-norm regularization coefficient “norm_coefficient”.
A learning-rate decay factor “decay_factor”.
A small constant “epsilon” to avoid dividing-by-zero.
At each ADAGRAD iteration, the optimized tensors are moved along a direction computed based on their estimated gradient and accumulated squared gradient. Assume that only a single tensor “X” is updated by this operator. We need the value of “X”, its gradient “G”, and its accumulated squared gradient “H”. Therefore, variables in this operator’s input list are sequentially “R”, “T”, “X”, “G”, and “H”. Other parameters are given as attributes because they are usually constants. Also, the corresponding output tensors are the new value of “X” (called “X_new”), and then the new accumulated squared gradient (called “H_new”). Those outputs are computed from the given inputs following the pseudo code below.
Let “+”, “-”, “*”, and “/” are all element-wise arithmetic operations with numpy-style broadcasting support. The pseudo code to compute those outputs is:
// Compute a scalar learning-rate factor. At the first update of X, T is generally // 0 (0-based update index) or 1 (1-based update index). r = R / (1 + T * decay_factor);
// Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. G_regularized = norm_coefficient * X + G;
// Compute new accumulated squared gradient. H_new = H + G_regularized * G_regularized;
// Compute the adaptive part of per-coordinate learning rate. Note that Sqrt(…) // computes element-wise square-root. H_adaptive = Sqrt(H_new) + epsilon
// Compute the new value of “X”. X_new = X - r * G_regularized / H_adaptive;
If one assign this operators to optimize multiple inputs, for example, “X_1” and “X_2”, the same pseudo code may be extended to handle all tensors jointly. More specifically, we can view “X” as a concatenation of “X_1” and “X_2” (of course, their gradient and accumulate gradient should be concatenated too) and then just reuse the entire pseudo code.
Note that ADAGRAD was first proposed in http://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf. In that reference paper, this operator is a special case of the Figure 1’s composite mirror descent update.
Attributes
decay_factor: The decay factor of learning rate after one update.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn’t reduce the learning rate. Default value is
0.0.epsilon: Small scalar to avoid dividing by zero. Default value is
9.999999974752427e-07.norm_coefficient: Regularization coefficient in 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization. Default value is
0.0.
Inputs
Between 3 and 2147483647 inputs.
R (heterogeneous) - T1: The initial learning rate.
T (heterogeneous) - T2: The update count of “X”. It should be a scalar.
inputs (variadic) - T3: The current values of optimized tensors, followed by their respective gradients, followed by their respective accumulated squared gradients.For example, if two tensor “X_1” and “X_2” are optimized, The input list would be [“X_1”, “X_2”, gradient of “X_1”, gradient of “X_2”, accumulated squared gradient of “X_1”, accumulated squared gradient of “X_2”].
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T3: Updated values of optimized tensors, followed by their updated values of accumulated squared gradients. For example, if two tensor “X_1” and “X_2” are optimized, the output list would be [new value of “X_1,” new value of “X_2” new accumulated squared gradient of “X_1”, new accumulated squared gradient of “X_2”].
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float scalars.
T2 in ( tensor(int64) ): Constrain input types to 64-bit integer scalars.
T3 in ( tensor(double), tensor(float) ): Constrain input and output types to float tensors.
OnnxAiOnnxPreviewTrainingAdagrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingAdagrad_1(*args, **kwargs)#
Version
name: Adagrad (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Compute one iteration of ADAGRAD, a stochastic gradient based optimization algorithm. This operator can conduct the optimization of multiple tensor variables.
Let’s define the behavior of this operator. As you can imagine, ADAGRAD requires some parameters:
The initial learning-rate “R”.
The update count “T”. That is, the number of training iterations conducted.
A L2-norm regularization coefficient “norm_coefficient”.
A learning-rate decay factor “decay_factor”.
A small constant “epsilon” to avoid dividing-by-zero.
At each ADAGRAD iteration, the optimized tensors are moved along a direction computed based on their estimated gradient and accumulated squared gradient. Assume that only a single tensor “X” is updated by this operator. We need the value of “X”, its gradient “G”, and its accumulated squared gradient “H”. Therefore, variables in this operator’s input list are sequentially “R”, “T”, “X”, “G”, and “H”. Other parameters are given as attributes because they are usually constants. Also, the corresponding output tensors are the new value of “X” (called “X_new”), and then the new accumulated squared gradient (called “H_new”). Those outputs are computed from the given inputs following the pseudo code below.
Let “+”, “-”, “*”, and “/” are all element-wise arithmetic operations with numpy-style broadcasting support. The pseudo code to compute those outputs is:
// Compute a scalar learning-rate factor. At the first update of X, T is generally // 0 (0-based update index) or 1 (1-based update index). r = R / (1 + T * decay_factor);
// Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. G_regularized = norm_coefficient * X + G;
// Compute new accumulated squared gradient. H_new = H + G_regularized * G_regularized;
// Compute the adaptive part of per-coordinate learning rate. Note that Sqrt(…) // computes element-wise square-root. H_adaptive = Sqrt(H_new) + epsilon
// Compute the new value of “X”. X_new = X - r * G_regularized / H_adaptive;
If one assign this operators to optimize multiple inputs, for example, “X_1” and “X_2”, the same pseudo code may be extended to handle all tensors jointly. More specifically, we can view “X” as a concatenation of “X_1” and “X_2” (of course, their gradient and accumulate gradient should be concatenated too) and then just reuse the entire pseudo code.
Note that ADAGRAD was first proposed in http://jmlr.org/papers/volume12/duchi11a/duchi11a.pdf. In that reference paper, this operator is a special case of the Figure 1’s composite mirror descent update.
Attributes
decay_factor: The decay factor of learning rate after one update.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn’t reduce the learning rate. Default value is
0.0.epsilon: Small scalar to avoid dividing by zero. Default value is
9.999999974752427e-07.norm_coefficient: Regularization coefficient in 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization. Default value is
0.0.
Inputs
Between 3 and 2147483647 inputs.
R (heterogeneous) - T1: The initial learning rate.
T (heterogeneous) - T2: The update count of “X”. It should be a scalar.
inputs (variadic) - T3: The current values of optimized tensors, followed by their respective gradients, followed by their respective accumulated squared gradients.For example, if two tensor “X_1” and “X_2” are optimized, The input list would be [“X_1”, “X_2”, gradient of “X_1”, gradient of “X_2”, accumulated squared gradient of “X_1”, accumulated squared gradient of “X_2”].
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T3: Updated values of optimized tensors, followed by their updated values of accumulated squared gradients. For example, if two tensor “X_1” and “X_2” are optimized, the output list would be [new value of “X_1,” new value of “X_2” new accumulated squared gradient of “X_1”, new accumulated squared gradient of “X_2”].
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float scalars.
T2 in ( tensor(int64) ): Constrain input types to 64-bit integer scalars.
T3 in ( tensor(double), tensor(float) ): Constrain input and output types to float tensors.
OnnxAiOnnxPreviewTrainingAdam#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingAdam(*args, **kwargs)#
Version
name: Adam (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Compute one iteration of Adam, a stochastic gradient based optimization algorithm. This operator can conduct the optimization of multiple tensor variables.
Let’s define the behavior of this operator. First of all, Adam requires some parameters:
The learning-rate “R”.
The update count “T”. That is, the number of training iterations conducted.
A L2-norm regularization coefficient “norm_coefficient”.
A small constant “epsilon” to avoid dividing-by-zero.
Two coefficients, “alpha” and “beta”.
At each Adam iteration, the optimized tensors are moved along a direction computed based on their exponentially-averaged historical gradient and exponentially-averaged historical squared gradient. Assume that only a tensor “X” is being optimized. The rest of required information is
the value of “X”,
“X“‘s gradient (denoted by “G”),
“X“‘s exponentially-averaged historical gradient (denoted by “V”), and
“X“‘s exponentially-averaged historical squared gradient (denoted by “H”).
Some of those parameters are passed into this operator as input tensors and others are stored as this operator’s attributes. Specifically, this operator’s input tensor list is [“R”, “T”, “X”, “G”, “V”, “H”]. That is, “R” is the first input, “T” is the second input, and so on. Other parameters are given as attributes because they are constants. Moreover, the corresponding output tensors are
the new value of “X” (called “X_new”),
the new exponentially-averaged historical gradient (denoted by “V_new”), and
the new exponentially-averaged historical squared gradient (denoted by “H_new”).
Those outputs are computed following the pseudo code below.
Let “+”, “-”, “*”, and “/” are all element-wise arithmetic operations with numpy-style broadcasting support. The pseudo code to compute those outputs is:
// Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. G_regularized = norm_coefficient * X + G
// Update exponentially-averaged historical gradient. V_new = alpha * V + (1 - alpha) * G_regularized
// Update exponentially-averaged historical squared gradient. H_new = beta * H + (1 - beta) * G_regularized * G_regularized
// Compute the element-wise square-root of H_new. V_new will be element-wisely // divided by H_sqrt for a better update direction. H_sqrt = Sqrt(H_new) + epsilon
// Compute learning-rate. Note that “alpha**T”/”beta**T” is alpha’s/beta’s T-th power. R_adjusted = T > 0 ? R * Sqrt(1 - beta**T) / (1 - alpha**T) : R
// Compute new value of “X”. X_new = X - R_adjusted * V_new / H_sqrt
// Post-update regularization. X_final = (1 - norm_coefficient_post) * X_new
If there are multiple inputs to be optimized, the pseudo code will be applied independently to each of them.
Attributes
alpha: Coefficient of previously accumulated gradient in running average. Default to 0.9. Default value is
0.8999999761581421.beta: Coefficient of previously accumulated squared-gradient in running average. Default to 0.999. Default value is
0.9990000128746033.epsilon: Small scalar to avoid dividing by zero. Default value is
9.999999974752427e-07.norm_coefficient: Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization. Default value is
0.0.norm_coefficient_post: Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization. Default value is
0.0.
Inputs
Between 3 and 2147483647 inputs.
R (heterogeneous) - T1: The initial learning rate.
T (heterogeneous) - T2: The update count of “X”. It should be a scalar.
inputs (variadic) - T3: The tensors to be optimized, followed by their respective gradients, followed by their respective accumulated gradients (aka momentum), followed by their respective accumulated squared gradients. For example, to optimize tensors “X_1” and “X_2,”, the input list would be [“X_1”, “X_2”, gradient of “X_1”, gradient of “X_2”, accumulated gradient of “X_1”, accumulated gradient of “X_2”, accumulated squared gradient of “X_1”, accumulated squared gradient of “X_2”].
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T3: New values of optimized tensors, followed by their respective new accumulated gradients, followed by their respective new accumulated squared gradients. For example, if two tensors “X_1” and “X_2” are optimized, the outputs list would be [new value of “X_1”, new value of “X_2”, new accumulated gradient of “X_1”, new accumulated gradient of “X_2”, new accumulated squared gradient of “X_1”, new accumulated squared gradient of “X_2”].
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float scalars.
T2 in ( tensor(int64) ): Constrain input types to 64-bit integer scalars.
T3 in ( tensor(double), tensor(float) ): Constrain input and output types to float tensors.
OnnxAiOnnxPreviewTrainingAdam_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingAdam_1(*args, **kwargs)#
Version
name: Adam (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Compute one iteration of Adam, a stochastic gradient based optimization algorithm. This operator can conduct the optimization of multiple tensor variables.
Let’s define the behavior of this operator. First of all, Adam requires some parameters:
The learning-rate “R”.
The update count “T”. That is, the number of training iterations conducted.
A L2-norm regularization coefficient “norm_coefficient”.
A small constant “epsilon” to avoid dividing-by-zero.
Two coefficients, “alpha” and “beta”.
At each Adam iteration, the optimized tensors are moved along a direction computed based on their exponentially-averaged historical gradient and exponentially-averaged historical squared gradient. Assume that only a tensor “X” is being optimized. The rest of required information is
the value of “X”,
“X“‘s gradient (denoted by “G”),
“X“‘s exponentially-averaged historical gradient (denoted by “V”), and
“X“‘s exponentially-averaged historical squared gradient (denoted by “H”).
Some of those parameters are passed into this operator as input tensors and others are stored as this operator’s attributes. Specifically, this operator’s input tensor list is [“R”, “T”, “X”, “G”, “V”, “H”]. That is, “R” is the first input, “T” is the second input, and so on. Other parameters are given as attributes because they are constants. Moreover, the corresponding output tensors are
the new value of “X” (called “X_new”),
the new exponentially-averaged historical gradient (denoted by “V_new”), and
the new exponentially-averaged historical squared gradient (denoted by “H_new”).
Those outputs are computed following the pseudo code below.
Let “+”, “-”, “*”, and “/” are all element-wise arithmetic operations with numpy-style broadcasting support. The pseudo code to compute those outputs is:
// Add gradient of 0.5 * norm_coefficient * ||X||_2^2, where ||X||_2 is the 2-norm. G_regularized = norm_coefficient * X + G
// Update exponentially-averaged historical gradient. V_new = alpha * V + (1 - alpha) * G_regularized
// Update exponentially-averaged historical squared gradient. H_new = beta * H + (1 - beta) * G_regularized * G_regularized
// Compute the element-wise square-root of H_new. V_new will be element-wisely // divided by H_sqrt for a better update direction. H_sqrt = Sqrt(H_new) + epsilon
// Compute learning-rate. Note that “alpha**T”/”beta**T” is alpha’s/beta’s T-th power. R_adjusted = T > 0 ? R * Sqrt(1 - beta**T) / (1 - alpha**T) : R
// Compute new value of “X”. X_new = X - R_adjusted * V_new / H_sqrt
// Post-update regularization. X_final = (1 - norm_coefficient_post) * X_new
If there are multiple inputs to be optimized, the pseudo code will be applied independently to each of them.
Attributes
alpha: Coefficient of previously accumulated gradient in running average. Default to 0.9. Default value is
0.8999999761581421.beta: Coefficient of previously accumulated squared-gradient in running average. Default to 0.999. Default value is
0.9990000128746033.epsilon: Small scalar to avoid dividing by zero. Default value is
9.999999974752427e-07.norm_coefficient: Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization. Default value is
0.0.norm_coefficient_post: Regularization coefficient of 0.5 * norm_coefficient * ||X||_2^2. Default to 0, which means no regularization. Default value is
0.0.
Inputs
Between 3 and 2147483647 inputs.
R (heterogeneous) - T1: The initial learning rate.
T (heterogeneous) - T2: The update count of “X”. It should be a scalar.
inputs (variadic) - T3: The tensors to be optimized, followed by their respective gradients, followed by their respective accumulated gradients (aka momentum), followed by their respective accumulated squared gradients. For example, to optimize tensors “X_1” and “X_2,”, the input list would be [“X_1”, “X_2”, gradient of “X_1”, gradient of “X_2”, accumulated gradient of “X_1”, accumulated gradient of “X_2”, accumulated squared gradient of “X_1”, accumulated squared gradient of “X_2”].
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T3: New values of optimized tensors, followed by their respective new accumulated gradients, followed by their respective new accumulated squared gradients. For example, if two tensors “X_1” and “X_2” are optimized, the outputs list would be [new value of “X_1”, new value of “X_2”, new accumulated gradient of “X_1”, new accumulated gradient of “X_2”, new accumulated squared gradient of “X_1”, new accumulated squared gradient of “X_2”].
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float scalars.
T2 in ( tensor(int64) ): Constrain input types to 64-bit integer scalars.
T3 in ( tensor(double), tensor(float) ): Constrain input and output types to float tensors.
OnnxAiOnnxPreviewTrainingGradient#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingGradient(*args, **kwargs)#
Version
name: Gradient (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Gradient operator computes the partial derivatives of a specific tensor w.r.t. some other tensors. This operator is widely used in gradient-based training algorithms. To illustrate its use, let’s consider a computation graph,
X -----. | v W --> Conv --> H --> Gemm --> Y ^ | Z
, where W and Z are trainable tensors. Note that operators’ attributes are omitted for the sake of simplicity. Let dY/dW (dY/dZ) be the gradient of Y with respect to W (Z). The user can compute gradient by inserting Gradient operator to form another graph shown below.
W --> Conv --> H --> Gemm --> Y | ^ ^ | | | | X Z | | | | | .----------' | | | (W/Z/X is the 1st/2nd/3rd input of Gradient as shown in | | | "xs" followed by "zs") | v v '---> Gradient(xs=["W", "Z"], zs=["X"], y="Y") | | | '-----------------------------------> dY/dW (1st output of Gradient) | '---------------------------------------> dY/dZ (2nd output of Gradient)
By definition, the tensor “y” is a function of independent variables in “xs” and “zs”. Since we only compute the gradient of “y” w.r.t. the differentiable variables in “xs”, this Gradient only outputs dY/dW and dY/dZ. Note that “H” cannot appear in “xs” and “zs”. The reason is that “H” can be determined by tensors “W” and “X” and therefore “H” is not an independent variable.
All outputs are optional. If needed, for example, user can assign an empty string to the 1st output name of that Gradient to skip the generation of dY/dW. Note that the concept of optional outputs can also be found in ONNX’s RNN, GRU, and LSTM.
Gradient operator can compute derivative against intermediate tensors. For example, the gradient of Y with respect to H can be done via
W --> Conv --> H --> Gemm --> Y ^ | ^ | | | X | Z .-------' | | .----------' | | (H/Z is the 1st/2nd input of Gradient as shown in "xs") v v Gradient(xs=["H", "Z"], y="Y") | | | '-----------------------------------> dY/dH (1st output of Gradient) | '---------------------------------------> dY/dZ (2nd output of Gradient)
It is possible to represent high-order differentiation using Gradient operators. For example, given the following linear model:
W --> Gemm --> Y --> Loss --> O ^ ^ | | X L
To compute the 2nd order derivative of O with respect to W (denoted by d^2O/dW^2), one can do
W --> Gemm --> Y --> Loss --> O | ^ ^ | | | | X .------------L | | | | | | | v +------+-+> Gradient(xs=["X", "W"], zs=["L"], y="O") ---> dO/dX (1st output of Gradient) | | | | | | | '---> dO/dW (2nd output of Gradient) | v v '---> Gradient(xs=["X", "W"], zs=["L"], y="dO/dW") ---> d(dO/dW)dX (1st output of | Gradient) | | '---> d^2O/dW^2 (2nd output of Gradient)
The tensors named in attributes “xs”, “zs”, and “y” define the differentiated computation graph, and the inputs to Gradient node define the values at which the gradient is computed. We can feed different tensors to the identified graph. For example, one can compute the gradient of Y with respect to H at a specific value of H, H_1, by providing that value as an input to the Gradient node.
W --> Conv --> H --> Gemm --> Y ^ ^ | | X Z Z_1 (2nd input of Gradient) | v H_1 --> Gradient(xs=["H", "Z"], y="Y") ---> dY/dH when H = H_1 and Y = Y_1. | '------------------------------> dY/dZ (2nd output of Gradient)
When the inputs of Gradient are the tensors named in “xs” and “zs”, the computation can be optimized. More specifically, intermediate variables in forward pass can be reused if the gradient is computed via reverse-mode auto-differentiation.
Attributes
xs (required): Input tensor names of the differentiated sub-graph. It contains only the necessary differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
y (required): The targeted tensor. It can be viewed as the output of the differentiated function. The attribute “xs” and attribute “zs” are the minimal independent variable set that determines the value of “y”.
zs: Input tensor names of the differentiated sub-graph. It contains only the necessary non-differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
Inputs
Between 1 and 2147483647 inputs.
Inputs (variadic) - T1: The values fed into graph identified by the attributes. The i-th input is the value of the i-th tensor specified in the concatenated list of the attribute “xs” and the attribute “zs”. For example, if xs=[“A”, “B”] and zs=[“C”], the first input is used as the value of symbol “A” and the 3rd input is substituted for all the occurrences of “C”.
Outputs
Between 1 and 2147483647 outputs.
Outputs (variadic) - T2: The gradient of the tensor specified by the attribute “y” with respect to each of tensors specified in the attribute “xs”. The i-th output is the gradient of “y” with respect to the i-th tensor specified in the attribute “xs”.
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Allow outputs to be any kind of tensor.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Allow inputs to be any kind of floating-point tensor.
OnnxAiOnnxPreviewTrainingGradient_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingGradient_1(*args, **kwargs)#
Version
name: Gradient (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Gradient operator computes the partial derivatives of a specific tensor w.r.t. some other tensors. This operator is widely used in gradient-based training algorithms. To illustrate its use, let’s consider a computation graph,
X -----. | v W --> Conv --> H --> Gemm --> Y ^ | Z
, where W and Z are trainable tensors. Note that operators’ attributes are omitted for the sake of simplicity. Let dY/dW (dY/dZ) be the gradient of Y with respect to W (Z). The user can compute gradient by inserting Gradient operator to form another graph shown below.
W --> Conv --> H --> Gemm --> Y | ^ ^ | | | | X Z | | | | | .----------' | | | (W/Z/X is the 1st/2nd/3rd input of Gradient as shown in | | | "xs" followed by "zs") | v v '---> Gradient(xs=["W", "Z"], zs=["X"], y="Y") | | | '-----------------------------------> dY/dW (1st output of Gradient) | '---------------------------------------> dY/dZ (2nd output of Gradient)
By definition, the tensor “y” is a function of independent variables in “xs” and “zs”. Since we only compute the gradient of “y” w.r.t. the differentiable variables in “xs”, this Gradient only outputs dY/dW and dY/dZ. Note that “H” cannot appear in “xs” and “zs”. The reason is that “H” can be determined by tensors “W” and “X” and therefore “H” is not an independent variable.
All outputs are optional. If needed, for example, user can assign an empty string to the 1st output name of that Gradient to skip the generation of dY/dW. Note that the concept of optional outputs can also be found in ONNX’s RNN, GRU, and LSTM.
Gradient operator can compute derivative against intermediate tensors. For example, the gradient of Y with respect to H can be done via
W --> Conv --> H --> Gemm --> Y ^ | ^ | | | X | Z .-------' | | .----------' | | (H/Z is the 1st/2nd input of Gradient as shown in "xs") v v Gradient(xs=["H", "Z"], y="Y") | | | '-----------------------------------> dY/dH (1st output of Gradient) | '---------------------------------------> dY/dZ (2nd output of Gradient)
It is possible to represent high-order differentiation using Gradient operators. For example, given the following linear model:
W --> Gemm --> Y --> Loss --> O ^ ^ | | X L
To compute the 2nd order derivative of O with respect to W (denoted by d^2O/dW^2), one can do
W --> Gemm --> Y --> Loss --> O | ^ ^ | | | | X .------------L | | | | | | | v +------+-+> Gradient(xs=["X", "W"], zs=["L"], y="O") ---> dO/dX (1st output of Gradient) | | | | | | | '---> dO/dW (2nd output of Gradient) | v v '---> Gradient(xs=["X", "W"], zs=["L"], y="dO/dW") ---> d(dO/dW)dX (1st output of | Gradient) | | '---> d^2O/dW^2 (2nd output of Gradient)
The tensors named in attributes “xs”, “zs”, and “y” define the differentiated computation graph, and the inputs to Gradient node define the values at which the gradient is computed. We can feed different tensors to the identified graph. For example, one can compute the gradient of Y with respect to H at a specific value of H, H_1, by providing that value as an input to the Gradient node.
W --> Conv --> H --> Gemm --> Y ^ ^ | | X Z Z_1 (2nd input of Gradient) | v H_1 --> Gradient(xs=["H", "Z"], y="Y") ---> dY/dH when H = H_1 and Y = Y_1. | '------------------------------> dY/dZ (2nd output of Gradient)
When the inputs of Gradient are the tensors named in “xs” and “zs”, the computation can be optimized. More specifically, intermediate variables in forward pass can be reused if the gradient is computed via reverse-mode auto-differentiation.
Attributes
xs (required): Input tensor names of the differentiated sub-graph. It contains only the necessary differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
y (required): The targeted tensor. It can be viewed as the output of the differentiated function. The attribute “xs” and attribute “zs” are the minimal independent variable set that determines the value of “y”.
zs: Input tensor names of the differentiated sub-graph. It contains only the necessary non-differentiated inputs of a (sub-)graph. Variables (usually called intermediate variables) that can be generated from inputs cannot be included in this attribute.
Inputs
Between 1 and 2147483647 inputs.
Inputs (variadic) - T1: The values fed into graph identified by the attributes. The i-th input is the value of the i-th tensor specified in the concatenated list of the attribute “xs” and the attribute “zs”. For example, if xs=[“A”, “B”] and zs=[“C”], the first input is used as the value of symbol “A” and the 3rd input is substituted for all the occurrences of “C”.
Outputs
Between 1 and 2147483647 outputs.
Outputs (variadic) - T2: The gradient of the tensor specified by the attribute “y” with respect to each of tensors specified in the attribute “xs”. The i-th output is the gradient of “y” with respect to the i-th tensor specified in the attribute “xs”.
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Allow outputs to be any kind of tensor.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Allow inputs to be any kind of floating-point tensor.
OnnxAiOnnxPreviewTrainingMomentum#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingMomentum(*args, **kwargs)#
Version
name: Momentum (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Compute one iteration of stochastic gradient update with momentum. This operator can conduct the optimization of multiple tensor variables.
Let’s define the behavior of this operator. As you can imagine, SG with momentum requires several parameters:
The learning-rate “R”.
The update count “T”. That is, the number of conducted training iterations. It should be zero in the first training iteration.
A L2-norm regularization coefficient “norm_coefficient”.
A decay coefficient of previous accumulated gradient (i.e., momentum) “alpha”.
The scaling coefficient of current gradient “beta”.
An attribute to choose either standard momentum or Nesterov’s momentum “mode” should be used.
For the sake of simplicity, assume that there is only one tensor (called “X”) to be optimized. Other necessary inputs are “X“‘s gradient (called “G”) and “X“‘s momentum (called “V”). This Momentum operator maps all these inputs to the new value of “X” (called “X_new”) and its new momentum (called “V_new”).
This operator supports two different momentum algorithms. Set the attribute “mode” to “nesterov” if Nesterov’s momentum is desired. Otherwise, set the attribute “model” to “standard” to use standard momentum. Computation details are described subsequently.
Let “+”, “-”, “*”, and “/” are all element-wise operations with numpy-style broadcasting.
Pseudo code for SG with standard momentum:
// Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared // values of all elements in X. G_regularized = norm_coefficient * X + G
// In the first training iteration, beta should always be 1. beta_adjusted = T > 0 ? beta : 1
// Compute the current momentum based on previous momentum and the current gradient. V_new = alpha * V + beta_adjusted * G_regularized
// Update X. X_new = X - R * V_new
Pseudo code for SG with Nesterov’s momentum:
// Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared // values of all elements in X. G_regularized = norm_coefficient * X + G;
// In the first training iteration, beta should always be 1. beta_adjusted = T > 0 ? beta : 1
// Compute the current momentum based on previous momentum and the current gradient. V_new = alpha * V + beta_adjusted * G_regularized;
// Compute final update direction and then update X. X_new = X - R * (G_regularized + alpha * V_new)
If one assign this operators to optimize multiple inputs, for example, “X_1” and “X_2”. The same pseudo code would be extended to handle all tensors jointly. More specifically, we can view “X” as a concatenation of “X_1” and “X_2” (of course, their gradient and accumulate gradient should be concatenated too) and then our pseudo code becomes applicable.
Attributes
alpha (required): The decay factor of momentum. It should be a scalar.
beta (required): The coefficient of gradient in computing new momentum. It should be a scalar.
mode (required): Its value should be either “nesterov” or “standard”. The value “nesterov” leads to the use of Nesterov’s momentum while “standard” invokes stochastic gradient method using standard momentum
norm_coefficient (required): Coefficient of 0.5 * norm_coefficient * ||X||^2.
Inputs
Between 3 and 2147483647 inputs.
R (heterogeneous) - T1: The learning rate.
T (heterogeneous) - T2: Update count of “X”. It should be a scalar.
inputs (variadic) - T3: It sequentially contains the current values of optimized tensors, then their gradient tensors, and finally their momentum tensors. For example, if two tensors “X_1” and “X_2” are optimized, The expected input list would be [“X_1”, “X_2”, gradient of “X_1”, gradient of “X_2”, momentum of “X_1”, momentum of “X_2”].
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T3: It sequentially contains the new values of optimized tensors and then the new values of their momentum tensors. For example, if two tensors “X_1” and “X_2” are optimized, the output list would be [new value of “X_1,” new value of “X_2” new momentum of “X_1”, new momentum of “X_2”].
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float scalars.
T2 in ( tensor(int64) ): Constrain input types to 64-bit integer scalars.
T3 in ( tensor(double), tensor(float) ): Constrain input types to float tensors.
OnnxAiOnnxPreviewTrainingMomentum_1#
- class mlprodict.npy.xop_auto_import_.OnnxAiOnnxPreviewTrainingMomentum_1(*args, **kwargs)#
Version
name: Momentum (GitHub)
domain: ai.onnx.preview.training
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1 of domain ai.onnx.preview.training.
Summary
Compute one iteration of stochastic gradient update with momentum. This operator can conduct the optimization of multiple tensor variables.
Let’s define the behavior of this operator. As you can imagine, SG with momentum requires several parameters:
The learning-rate “R”.
The update count “T”. That is, the number of conducted training iterations. It should be zero in the first training iteration.
A L2-norm regularization coefficient “norm_coefficient”.
A decay coefficient of previous accumulated gradient (i.e., momentum) “alpha”.
The scaling coefficient of current gradient “beta”.
An attribute to choose either standard momentum or Nesterov’s momentum “mode” should be used.
For the sake of simplicity, assume that there is only one tensor (called “X”) to be optimized. Other necessary inputs are “X“‘s gradient (called “G”) and “X“‘s momentum (called “V”). This Momentum operator maps all these inputs to the new value of “X” (called “X_new”) and its new momentum (called “V_new”).
This operator supports two different momentum algorithms. Set the attribute “mode” to “nesterov” if Nesterov’s momentum is desired. Otherwise, set the attribute “model” to “standard” to use standard momentum. Computation details are described subsequently.
Let “+”, “-”, “*”, and “/” are all element-wise operations with numpy-style broadcasting.
Pseudo code for SG with standard momentum:
// Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared // values of all elements in X. G_regularized = norm_coefficient * X + G
// In the first training iteration, beta should always be 1. beta_adjusted = T > 0 ? beta : 1
// Compute the current momentum based on previous momentum and the current gradient. V_new = alpha * V + beta_adjusted * G_regularized
// Update X. X_new = X - R * V_new
Pseudo code for SG with Nesterov’s momentum:
// Add gradient of 0.5 * norm_coefficient * ||X||^2, where ||X|| is the sum of squared // values of all elements in X. G_regularized = norm_coefficient * X + G;
// In the first training iteration, beta should always be 1. beta_adjusted = T > 0 ? beta : 1
// Compute the current momentum based on previous momentum and the current gradient. V_new = alpha * V + beta_adjusted * G_regularized;
// Compute final update direction and then update X. X_new = X - R * (G_regularized + alpha * V_new)
If one assign this operators to optimize multiple inputs, for example, “X_1” and “X_2”. The same pseudo code would be extended to handle all tensors jointly. More specifically, we can view “X” as a concatenation of “X_1” and “X_2” (of course, their gradient and accumulate gradient should be concatenated too) and then our pseudo code becomes applicable.
Attributes
alpha (required): The decay factor of momentum. It should be a scalar.
beta (required): The coefficient of gradient in computing new momentum. It should be a scalar.
mode (required): Its value should be either “nesterov” or “standard”. The value “nesterov” leads to the use of Nesterov’s momentum while “standard” invokes stochastic gradient method using standard momentum
norm_coefficient (required): Coefficient of 0.5 * norm_coefficient * ||X||^2.
Inputs
Between 3 and 2147483647 inputs.
R (heterogeneous) - T1: The learning rate.
T (heterogeneous) - T2: Update count of “X”. It should be a scalar.
inputs (variadic) - T3: It sequentially contains the current values of optimized tensors, then their gradient tensors, and finally their momentum tensors. For example, if two tensors “X_1” and “X_2” are optimized, The expected input list would be [“X_1”, “X_2”, gradient of “X_1”, gradient of “X_2”, momentum of “X_1”, momentum of “X_2”].
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T3: It sequentially contains the new values of optimized tensors and then the new values of their momentum tensors. For example, if two tensors “X_1” and “X_2” are optimized, the output list would be [new value of “X_1,” new value of “X_2” new momentum of “X_1”, new momentum of “X_2”].
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float scalars.
T2 in ( tensor(int64) ): Constrain input types to 64-bit integer scalars.
T3 in ( tensor(double), tensor(float) ): Constrain input types to float tensors.
OnnxAnd#
- class mlprodict.npy.xop_auto_import_.OnnxAnd(*args, **kwargs)#
Version
name: And (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the and logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxAnd_1#
- class mlprodict.npy.xop_auto_import_.OnnxAnd_1(*args, **kwargs)#
Version
name: And (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the tensor resulted from performing the and logical operation elementwise on the input tensors A and B.
If broadcasting is enabled, the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. See the doc of Add for a detailed description of the broadcasting rules.
Attributes
axis: If set, defines the broadcast dimensions.
broadcast: Enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: Left input tensor for the logical operator.
B (heterogeneous) - T: Right input tensor for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxAnd_7#
- class mlprodict.npy.xop_auto_import_.OnnxAnd_7(*args, **kwargs)#
Version
name: And (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the and logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxArgMax#
- class mlprodict.npy.xop_auto_import_.OnnxArgMax(*args, **kwargs)#
Version
name: ArgMax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the indices of the max elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. If select_last_index is True (default False), the index of the last occurrence of the max is selected if the max appears more than once in the input. Otherwise the index of the first occurrence is selected. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.select_last_index: Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index). Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMax_1#
- class mlprodict.npy.xop_auto_import_.OnnxArgMax_1(*args, **kwargs)#
Version
name: ArgMax (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the indices of the max elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMax_11#
- class mlprodict.npy.xop_auto_import_.OnnxArgMax_11(*args, **kwargs)#
Version
name: ArgMax (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the indices of the max elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMax_12#
- class mlprodict.npy.xop_auto_import_.OnnxArgMax_12(*args, **kwargs)#
Version
name: ArgMax (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Computes the indices of the max elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. If select_last_index is True (default False), the index of the last occurrence of the max is selected if the max appears more than once in the input. Otherwise the index of the first occurrence is selected. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.select_last_index: Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index). Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMax_13#
- class mlprodict.npy.xop_auto_import_.OnnxArgMax_13(*args, **kwargs)#
Version
name: ArgMax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the indices of the max elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. If select_last_index is True (default False), the index of the last occurrence of the max is selected if the max appears more than once in the input. Otherwise the index of the first occurrence is selected. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.select_last_index: Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index). Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMin#
- class mlprodict.npy.xop_auto_import_.OnnxArgMin(*args, **kwargs)#
Version
name: ArgMin (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the indices of the min elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. If select_last_index is True (default False), the index of the last occurrence of the min is selected if the min appears more than once in the input. Otherwise the index of the first occurrence is selected. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.select_last_index: Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index). Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMin_1#
- class mlprodict.npy.xop_auto_import_.OnnxArgMin_1(*args, **kwargs)#
Version
name: ArgMin (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the indices of the min elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMin_11#
- class mlprodict.npy.xop_auto_import_.OnnxArgMin_11(*args, **kwargs)#
Version
name: ArgMin (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the indices of the min elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMin_12#
- class mlprodict.npy.xop_auto_import_.OnnxArgMin_12(*args, **kwargs)#
Version
name: ArgMin (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Computes the indices of the min elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulting tensor has the reduced dimension pruned. If select_last_index is True (default False), the index of the last occurrence of the min is selected if the min appears more than once in the input. Otherwise the index of the first occurrence is selected. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.select_last_index: Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index). Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxArgMin_13#
- class mlprodict.npy.xop_auto_import_.OnnxArgMin_13(*args, **kwargs)#
Version
name: ArgMin (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the indices of the min elements of the input tensor’s element along the provided axis. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned. If select_last_index is True (default False), the index of the last occurrence of the min is selected if the min appears more than once in the input. Otherwise the index of the first occurrence is selected. The type of the output tensor is integer.
Attributes
axis: The axis in which to compute the arg indices. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.select_last_index: Whether to select the last index or the first index if the {name} appears in multiple indices, default is False (first index). Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - tensor(int64): Reduced output tensor with integer data type.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxAsin#
- class mlprodict.npy.xop_auto_import_.OnnxAsin(*args, **kwargs)#
Version
name: Asin (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the arcsine (inverse of sine) of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The arcsine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAsin_7#
- class mlprodict.npy.xop_auto_import_.OnnxAsin_7(*args, **kwargs)#
Version
name: Asin (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the arcsine (inverse of sine) of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The arcsine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAsinh#
- class mlprodict.npy.xop_auto_import_.OnnxAsinh(*args, **kwargs)#
Version
name: Asinh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic arcsine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic arcsine values of the input tensor computed element- wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAsinh_9#
- class mlprodict.npy.xop_auto_import_.OnnxAsinh_9(*args, **kwargs)#
Version
name: Asinh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic arcsine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic arcsine values of the input tensor computed element- wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAtan#
- class mlprodict.npy.xop_auto_import_.OnnxAtan(*args, **kwargs)#
Version
name: Atan (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the arctangent (inverse of tangent) of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The arctangent of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAtan_7#
- class mlprodict.npy.xop_auto_import_.OnnxAtan_7(*args, **kwargs)#
Version
name: Atan (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the arctangent (inverse of tangent) of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The arctangent of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAtanh#
- class mlprodict.npy.xop_auto_import_.OnnxAtanh(*args, **kwargs)#
Version
name: Atanh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic arctangent of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic arctangent values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAtanh_9#
- class mlprodict.npy.xop_auto_import_.OnnxAtanh_9(*args, **kwargs)#
Version
name: Atanh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic arctangent of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic arctangent values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAveragePool#
- class mlprodict.npy.xop_auto_import_.OnnxAveragePool(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
AveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.count_include_pad: Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad. Default value is
0.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAveragePool_1#
- class mlprodict.npy.xop_auto_import_.OnnxAveragePool_1(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
AveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) * pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is divided by the number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAveragePool_10#
- class mlprodict.npy.xop_auto_import_.OnnxAveragePool_10(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
AveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.count_include_pad: Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad. Default value is
0.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAveragePool_11#
- class mlprodict.npy.xop_auto_import_.OnnxAveragePool_11(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
AveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.count_include_pad: Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad. Default value is
0.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxAveragePool_7#
- class mlprodict.npy.xop_auto_import_.OnnxAveragePool_7(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
AveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) * pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.count_include_pad: Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad. Default value is
0.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxBatchNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization(*args, **kwargs)#
Version
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, There are five required inputs ‘X’, ‘scale’, ‘B’, ‘input_mean’ and ‘input_var’. Note that ‘input_mean’ and ‘input_var’ are expected to be the estimated statistics in inference mode (training_mode=False, default), and the running statistics in training mode (training_mode=True). There are multiple cases for the number of outputs, which we list below:
Output case #1: Y, running_mean, running_var (training_mode=True) Output case #2: Y (training_mode=False)
When training_mode=False, extra outputs are invalid. The outputs are updated as follows when training_mode=True:
running_mean = input_mean * momentum + current_mean * (1 - momentum) running_var = input_var * momentum + current_var * (1 - momentum) Y = (X - current_mean) / sqrt(current_var + epsilon) * scale + B where: current_mean = ReduceMean(X, axis=all_except_channel_index) current_var = ReduceVar(X, axis=all_except_channel_index) Notice that ReduceVar refers to the population variance, and it equals to sum(sqrd(x_i - x_avg)) / N where N is the population size (this formula does not use sample size N - 1).
The computation of ReduceMean and ReduceVar uses float to avoid overflow for float16 inputs.
When training_mode=False:
Y = (X - input_mean) / sqrt(input_var + epsilon) * scale + B
For previous (depreciated) non-spatial cases, implementors are suggested to flatten the input shape to (N x C * D1 * D2 * … * Dn) before a BatchNormalization Op. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum). Default value is
0.8999999761581421.training_mode: If set to true, it indicates BatchNormalization is being used for training, and outputs 1, 2, 3, and 4 would be populated. Default value is
0.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
scale (heterogeneous) - T1: Scale tensor of shape (C).
B (heterogeneous) - T1: Bias tensor of shape (C).
input_mean (heterogeneous) - T2: running (training) or estimated (testing) mean tensor of shape (C).
input_var (heterogeneous) - T2: running (training) or estimated (testing) variance tensor of shape (C).
Outputs
Between 1 and 3 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
running_mean (optional, heterogeneous) - T2: The running mean after the BatchNormalization operator.
running_var (optional, heterogeneous) - T2: The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain scale and bias types to float tensors.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain mean and variance types to float tensors.
OnnxBatchNormalization_1#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization_1(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, there are multiple cases for the number of outputs, which we list below:
Output case #1: Y, mean, var, saved_mean, saved_var (training mode) Output case #2: Y (test mode)
Attributes
consumed_inputs (required): legacy optimization attribute.
epsilon: The epsilon value to use to avoid division by zero, default is 1e-5f. Default value is
9.999999747378752e-06.is_test: If set to nonzero, run spatial batch normalization in test mode, default is 0. Default value is
0.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum), default is 0.9f. Default value is
0.8999999761581421.spatial: If true, compute the mean and variance across all spatial elements If false, compute the mean and variance across per feature.Default is 1. Default value is
1.
Inputs
X (heterogeneous) - T: The input 4-dimensional tensor of shape NCHW.
scale (heterogeneous) - T: The scale as a 1-dimensional tensor of size C to be applied to the output.
B (heterogeneous) - T: The bias as a 1-dimensional tensor of size C to be applied to the output.
mean (heterogeneous) - T: The running mean (training) or the estimated mean (testing) as a 1-dimensional tensor of size C.
var (heterogeneous) - T: The running variance (training) or the estimated variance (testing) as a 1-dimensional tensor of size C.
Outputs
Between 1 and 5 outputs.
Y (heterogeneous) - T: The output 4-dimensional tensor of the same shape as X.
mean (optional, heterogeneous) - T: The running mean after the BatchNormalization operator. Must be in- place with the input mean. Should not be used for testing.
var (optional, heterogeneous) - T: The running variance after the BatchNormalization operator. Must be in-place with the input var. Should not be used for testing.
saved_mean (optional, heterogeneous) - T: Saved mean used during training to speed up gradient computation. Should not be used for testing.
saved_var (optional, heterogeneous) - T: Saved variance used during training to speed up gradient computation. Should not be used for testing.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxBatchNormalization_14#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization_14(*args, **kwargs)#
Version
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, There are five required inputs ‘X’, ‘scale’, ‘B’, ‘input_mean’ and ‘input_var’. Note that ‘input_mean’ and ‘input_var’ are expected to be the estimated statistics in inference mode (training_mode=False, default), and the running statistics in training mode (training_mode=True). There are multiple cases for the number of outputs, which we list below:
Output case #1: Y, running_mean, running_var (training_mode=True) Output case #2: Y (training_mode=False)
When training_mode=False, extra outputs are invalid. The outputs are updated as follows when training_mode=True:
running_mean = input_mean * momentum + current_mean * (1 - momentum) running_var = input_var * momentum + current_var * (1 - momentum) Y = (X - current_mean) / sqrt(current_var + epsilon) * scale + B where: current_mean = ReduceMean(X, axis=all_except_channel_index) current_var = ReduceVar(X, axis=all_except_channel_index) Notice that ReduceVar refers to the population variance, and it equals to sum(sqrd(x_i - x_avg)) / N where N is the population size (this formula does not use sample size N - 1).
When training_mode=False:
Y = (X - input_mean) / sqrt(input_var + epsilon) * scale + B
For previous (depreciated) non-spatial cases, implementors are suggested to flatten the input shape to (N x C * D1 * D2 * … * Dn) before a BatchNormalization Op. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum). Default value is
0.8999999761581421.training_mode: If set to true, it indicates BatchNormalization is being used for training, and outputs 1, 2, 3, and 4 would be populated. Default value is
0.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
scale (heterogeneous) - T: Scale tensor of shape (C).
B (heterogeneous) - T: Bias tensor of shape (C).
input_mean (heterogeneous) - U: running (training) or estimated (testing) mean tensor of shape (C).
input_var (heterogeneous) - U: running (training) or estimated (testing) variance tensor of shape (C).
Outputs
Between 1 and 3 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
running_mean (optional, heterogeneous) - U: The running mean after the BatchNormalization operator.
running_var (optional, heterogeneous) - U: The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
U in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain mean and variance types to float tensors. It allows all float type for U.
OnnxBatchNormalization_15#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization_15(*args, **kwargs)#
Version
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, There are five required inputs ‘X’, ‘scale’, ‘B’, ‘input_mean’ and ‘input_var’. Note that ‘input_mean’ and ‘input_var’ are expected to be the estimated statistics in inference mode (training_mode=False, default), and the running statistics in training mode (training_mode=True). There are multiple cases for the number of outputs, which we list below:
Output case #1: Y, running_mean, running_var (training_mode=True) Output case #2: Y (training_mode=False)
When training_mode=False, extra outputs are invalid. The outputs are updated as follows when training_mode=True:
running_mean = input_mean * momentum + current_mean * (1 - momentum) running_var = input_var * momentum + current_var * (1 - momentum) Y = (X - current_mean) / sqrt(current_var + epsilon) * scale + B where: current_mean = ReduceMean(X, axis=all_except_channel_index) current_var = ReduceVar(X, axis=all_except_channel_index) Notice that ReduceVar refers to the population variance, and it equals to sum(sqrd(x_i - x_avg)) / N where N is the population size (this formula does not use sample size N - 1).
The computation of ReduceMean and ReduceVar uses float to avoid overflow for float16 inputs.
When training_mode=False:
Y = (X - input_mean) / sqrt(input_var + epsilon) * scale + B
For previous (depreciated) non-spatial cases, implementors are suggested to flatten the input shape to (N x C * D1 * D2 * … * Dn) before a BatchNormalization Op. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum). Default value is
0.8999999761581421.training_mode: If set to true, it indicates BatchNormalization is being used for training, and outputs 1, 2, 3, and 4 would be populated. Default value is
0.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
scale (heterogeneous) - T1: Scale tensor of shape (C).
B (heterogeneous) - T1: Bias tensor of shape (C).
input_mean (heterogeneous) - T2: running (training) or estimated (testing) mean tensor of shape (C).
input_var (heterogeneous) - T2: running (training) or estimated (testing) variance tensor of shape (C).
Outputs
Between 1 and 3 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
running_mean (optional, heterogeneous) - T2: The running mean after the BatchNormalization operator.
running_var (optional, heterogeneous) - T2: The running variance after the BatchNormalization operator. This op uses the population size (N) for calculating variance, and not the sample size N-1.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain scale and bias types to float tensors.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain mean and variance types to float tensors.
OnnxBatchNormalization_6#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization_6(*args, **kwargs)#
Version
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, there are multiple cases for the number of outputs, which we list below:
Output case #1: Y, mean, var, saved_mean, saved_var (training mode) Output case #2: Y (test mode)
Attributes
epsilon: The epsilon value to use to avoid division by zero, default is 1e-5f. Default value is
9.999999747378752e-06.is_test: If set to nonzero, run spatial batch normalization in test mode, default is 0. Default value is
0.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum), default is 0.9f. Default value is
0.8999999761581421.spatial: If true, compute the mean and variance across all spatial elements If false, compute the mean and variance across per feature.Default is 1. Default value is
1.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
scale (heterogeneous) - T: The scale as a 1-dimensional tensor of size C to be applied to the output.
B (heterogeneous) - T: The bias as a 1-dimensional tensor of size C to be applied to the output.
mean (heterogeneous) - T: The running mean (training) or the estimated mean (testing) as a 1-dimensional tensor of size C.
var (heterogeneous) - T: The running variance (training) or the estimated variance (testing) as a 1-dimensional tensor of size C.
Outputs
Between 1 and 5 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X.
mean (optional, heterogeneous) - T: The running mean after the BatchNormalization operator. Must be in- place with the input mean. Should not be used for testing.
var (optional, heterogeneous) - T: The running variance after the BatchNormalization operator. Must be in-place with the input var. Should not be used for testing.
saved_mean (optional, heterogeneous) - T: Saved mean used during training to speed up gradient computation. Should not be used for testing.
saved_var (optional, heterogeneous) - T: Saved variance used during training to speed up gradient computation. Should not be used for testing.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxBatchNormalization_7#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization_7(*args, **kwargs)#
Version
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, there are multiple cases for the number of outputs, which we list below:
Output case #1: Y, mean, var, saved_mean, saved_var (training mode) Output case #2: Y (test mode)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum). Default value is
0.8999999761581421.spatial: If true, compute the mean and variance across per activation. If false, compute the mean and variance across per feature over each mini-batch. Default value is
1.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
scale (heterogeneous) - T: If spatial is true, the dimension of scale is (C). If spatial is false, the dimensions of scale are (C x D1 x … x Dn)
B (heterogeneous) - T: If spatial is true, the dimension of bias is (C). If spatial is false, the dimensions of bias are (C x D1 x … x Dn)
mean (heterogeneous) - T: If spatial is true, the dimension of the running mean (training) or the estimated mean (testing) is (C). If spatial is false, the dimensions of the running mean (training) or the estimated mean (testing) are (C x D1 x … x Dn).
var (heterogeneous) - T: If spatial is true, the dimension of the running variance(training) or the estimated variance (testing) is (C). If spatial is false, the dimensions of the running variance(training) or the estimated variance (testing) are (C x D1 x … x Dn).
Outputs
Between 1 and 5 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
mean (optional, heterogeneous) - T: The running mean after the BatchNormalization operator.
var (optional, heterogeneous) - T: The running variance after the BatchNormalization operator.
saved_mean (optional, heterogeneous) - T: Saved mean used during training to speed up gradient computation.
saved_var (optional, heterogeneous) - T: Saved variance used during training to speed up gradient computation.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxBatchNormalization_9#
- class mlprodict.npy.xop_auto_import_.OnnxBatchNormalization_9(*args, **kwargs)#
Version
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Carries out batch normalization as described in the paper https://arxiv.org/abs/1502.03167. Depending on the mode it is being run, there are multiple cases for the number of outputs, which we list below:
Output case #1: Y, mean, var, saved_mean, saved_var (training mode) Output case #2: Y (test mode)
For previous (depreciated) non-spatial cases, implementors are suggested to flatten the input shape to (N x C*D1*D2 ..*Dn) before a BatchNormalization Op. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.momentum: Factor used in computing the running mean and variance.e.g., running_mean = running_mean * momentum + mean * (1 - momentum). Default value is
0.8999999761581421.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size, C is the number of channels. Statistics are computed for every channel of C over N and D1 to Dn dimensions. For image data, input dimensions become (N x C x H x W). The op also accepts single dimension input of size N in which case C is assumed to be 1
scale (heterogeneous) - T: Scale tensor of shape (C).
B (heterogeneous) - T: Bias tensor of shape (C).
mean (heterogeneous) - T: running (training) or estimated (testing) mean tensor of shape (C).
var (heterogeneous) - T: running (training) or estimated (testing) variance tensor of shape (C).
Outputs
Between 1 and 5 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
mean (optional, heterogeneous) - T: The running mean after the BatchNormalization operator.
var (optional, heterogeneous) - T: The running variance after the BatchNormalization operator.
saved_mean (optional, heterogeneous) - T: Saved mean used during training to speed up gradient computation.
saved_var (optional, heterogeneous) - T: Saved variance used during training to speed up gradient computation.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxBernoulli#
- class mlprodict.npy.xop_auto_import_.OnnxBernoulli(*args, **kwargs)#
Version
name: Bernoulli (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Draws binary random numbers (0 or 1) from a Bernoulli distribution. The input tensor should be a tensor containing probabilities p (a value in the range [0,1]) to be used for drawing the binary random number, where an output of 1 is produced with probability p and an output of 0 is produced with probability (1-p).
This operator is non-deterministic and may not produce the same values in different implementations (even if a seed is specified).
Attributes
dtype: The data type for the elements of the output tensor. if not specified, we will use the data type of the input tensor.
seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: All values in input have to be in the range:[0, 1].
Outputs
output (heterogeneous) - T2: The returned output tensor only has values 0 or 1, same shape as input tensor.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to all numeric tensors and bool tensors.
OnnxBernoulli_15#
- class mlprodict.npy.xop_auto_import_.OnnxBernoulli_15(*args, **kwargs)#
Version
name: Bernoulli (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Draws binary random numbers (0 or 1) from a Bernoulli distribution. The input tensor should be a tensor containing probabilities p (a value in the range [0,1]) to be used for drawing the binary random number, where an output of 1 is produced with probability p and an output of 0 is produced with probability (1-p).
This operator is non-deterministic and may not produce the same values in different implementations (even if a seed is specified).
Attributes
dtype: The data type for the elements of the output tensor. if not specified, we will use the data type of the input tensor.
seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: All values in input have to be in the range:[0, 1].
Outputs
output (heterogeneous) - T2: The returned output tensor only has values 0 or 1, same shape as input tensor.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to all numeric tensors and bool tensors.
OnnxBitShift#
- class mlprodict.npy.xop_auto_import_.OnnxBitShift(*args, **kwargs)#
Version
name: BitShift (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- Bitwise shift operator performs element-wise operation. For each input element, if the
attribute “direction” is “RIGHT”, this operator moves its binary representation toward the right side so that the input value is effectively decreased. If the attribute “direction” is “LEFT”, bits of binary representation moves toward the left side, which results the increase of its actual value. The input X is the tensor to be shifted and another input Y specifies the amounts of shifting. For example, if “direction” is “Right”, X is [1, 4], and S is [1, 1], the corresponding output Z would be [0, 2]. If “direction” is “LEFT” with X=[1, 2] and S=[1, 2], the corresponding output Y would be [2, 8].
Because this operator supports Numpy-style broadcasting, X’s and Y’s shapes are not necessarily identical.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Attributes
direction (required): Direction of moving bits. It can be either “RIGHT” (for right shift) or “LEFT” (for left shift).
Inputs
X (heterogeneous) - T: First operand, input to be shifted.
Y (heterogeneous) - T: Second operand, amounts of shift.
Outputs
Z (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to integer tensors.
OnnxBitShift_11#
- class mlprodict.npy.xop_auto_import_.OnnxBitShift_11(*args, **kwargs)#
Version
name: BitShift (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- Bitwise shift operator performs element-wise operation. For each input element, if the
attribute “direction” is “RIGHT”, this operator moves its binary representation toward the right side so that the input value is effectively decreased. If the attribute “direction” is “LEFT”, bits of binary representation moves toward the left side, which results the increase of its actual value. The input X is the tensor to be shifted and another input Y specifies the amounts of shifting. For example, if “direction” is “Right”, X is [1, 4], and S is [1, 1], the corresponding output Z would be [0, 2]. If “direction” is “LEFT” with X=[1, 2] and S=[1, 2], the corresponding output Y would be [2, 8].
Because this operator supports Numpy-style broadcasting, X’s and Y’s shapes are not necessarily identical.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Attributes
direction (required): Direction of moving bits. It can be either “RIGHT” (for right shift) or “LEFT” (for left shift).
Inputs
X (heterogeneous) - T: First operand, input to be shifted.
Y (heterogeneous) - T: Second operand, amounts of shift.
Outputs
Z (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to integer tensors.
OnnxBitwiseAnd#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseAnd(*args, **kwargs)#
Version
name: BitwiseAnd (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the tensor resulting from performing the bitwise and operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the bitwise operator.
B (heterogeneous) - T: Second input operand for the bitwise operator.
Outputs
C (heterogeneous) - T: Result tensor.
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to integer tensors.
OnnxBitwiseAnd_18#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseAnd_18(*args, **kwargs)#
Version
name: BitwiseAnd (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the tensor resulting from performing the bitwise and operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the bitwise operator.
B (heterogeneous) - T: Second input operand for the bitwise operator.
Outputs
C (heterogeneous) - T: Result tensor.
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to integer tensors.
OnnxBitwiseNot#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseNot(*args, **kwargs)#
Version
name: BitwiseNot (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the bitwise not of the input tensor element-wise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input/output to integer tensors.
OnnxBitwiseNot_18#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseNot_18(*args, **kwargs)#
Version
name: BitwiseNot (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the bitwise not of the input tensor element-wise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input/output to integer tensors.
OnnxBitwiseOr#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseOr(*args, **kwargs)#
Version
name: BitwiseOr (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the tensor resulting from performing the bitwise or operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the bitwise operator.
B (heterogeneous) - T: Second input operand for the bitwise operator.
Outputs
C (heterogeneous) - T: Result tensor.
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to integer tensors.
OnnxBitwiseOr_18#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseOr_18(*args, **kwargs)#
Version
name: BitwiseOr (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the tensor resulting from performing the bitwise or operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the bitwise operator.
B (heterogeneous) - T: Second input operand for the bitwise operator.
Outputs
C (heterogeneous) - T: Result tensor.
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to integer tensors.
OnnxBitwiseXor#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseXor(*args, **kwargs)#
Version
name: BitwiseXor (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the tensor resulting from performing the bitwise xor operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the bitwise operator.
B (heterogeneous) - T: Second input operand for the bitwise operator.
Outputs
C (heterogeneous) - T: Result tensor.
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to integer tensors.
OnnxBitwiseXor_18#
- class mlprodict.npy.xop_auto_import_.OnnxBitwiseXor_18(*args, **kwargs)#
Version
name: BitwiseXor (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns the tensor resulting from performing the bitwise xor operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the bitwise operator.
B (heterogeneous) - T: Second input operand for the bitwise operator.
Outputs
C (heterogeneous) - T: Result tensor.
Type Constraints
T in ( tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to integer tensors.
OnnxBlackmanWindow#
- class mlprodict.npy.xop_auto_import_.OnnxBlackmanWindow(*args, **kwargs)#
Version
name: BlackmanWindow (GitHub)
domain: main
since_version: 17
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generates a Blackman window as described in the paper https://ieeexplore.ieee.org/document/1455106.
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT. Default value is
1.periodic: If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1. Default value is
1.
Inputs
size (heterogeneous) - T1: A scalar value indicating the length of the window.
Outputs
output (heterogeneous) - T2: A Blackman window with length: size. The output has the shape: [size].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain the input size to int64_t.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to numeric tensors.
OnnxBlackmanWindow_17#
- class mlprodict.npy.xop_auto_import_.OnnxBlackmanWindow_17(*args, **kwargs)#
Version
name: BlackmanWindow (GitHub)
domain: main
since_version: 17
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generates a Blackman window as described in the paper https://ieeexplore.ieee.org/document/1455106.
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT. Default value is
1.periodic: If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1. Default value is
1.
Inputs
size (heterogeneous) - T1: A scalar value indicating the length of the window.
Outputs
output (heterogeneous) - T2: A Blackman window with length: size. The output has the shape: [size].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain the input size to int64_t.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to numeric tensors.
OnnxCast#
- class mlprodict.npy.xop_auto_import_.OnnxCast(*args, **kwargs)#
Version
name: Cast (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator casts the elements of a given input tensor to a data type specified by the ‘to’ argument and returns an output tensor of the same size in the converted type. The ‘to’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Casting from string tensor in plain (e.g., “3.14” and “1000”) and scientific numeric representations (e.g., “1e-5” and “1E8”) to float types is supported. For example, converting string “100.5” to an integer may result 100. There are some string literals reserved for special floating-point values; “+INF” (and “INF”), “-INF”, and “NaN” are positive infinity, negative infinity, and not-a-number, respectively. Any string which can exactly match “+INF” in a case-insensitive way would be mapped to positive infinite. Similarly, this case-insensitive rule is applied to “INF” and “NaN”. When casting from numeric tensors to string tensors, plain floating-point representation (such as “314.15926”) would be used. Converting non-numerical-literal string such as “Hello World!” is an undefined behavior. Cases of converting string representing floating-point arithmetic value, such as “2.718”, to INT is an undefined behavior.
Conversion from a numerical type to any numerical type is always allowed. User must be aware of precision loss and value change caused by range difference between two types. For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting an integer 36 to Boolean may produce 1 because we truncate bits which can’t be stored in the targeted type.
In more detail, the conversion among numerical types should follow these rules:
Casting from floating point to: * floating point: +/- infinity if OOR (out of range). * fixed point: undefined if OOR. * bool: +/- 0.0 to False; all else to True.
Casting from fixed point to: * floating point: +/- infinity if OOR. (+ infinity in the case of uint) * fixed point: when OOR, discard higher bits and reinterpret (with respect to two’s complement representation for
- signed types). For example, 200 (int16) -> -56 (int8).
bool: zero to False; nonzero to True.
Casting from bool to: * floating point: {1.0, 0.0}. * fixed point: {1, 0}. * bool: no change.
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
Outputs
output (heterogeneous) - T2: Output tensor with the same shape as input with type specified by the ‘to’ argument
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from complex is not supported.
T2 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to complex is not supported.
OnnxCastLike#
- class mlprodict.npy.xop_auto_import_.OnnxCastLike(*args, **kwargs)#
Version
name: CastLike (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
The operator casts the elements of a given input tensor (the first input) to the same data type as the elements of the second input tensor. See documentation of the Cast operator for further details.
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
target_type (heterogeneous) - T2: The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
Outputs
output (heterogeneous) - T2: Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from complex is not supported.
T2 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to complex is not supported.
OnnxCastLike_15#
- class mlprodict.npy.xop_auto_import_.OnnxCastLike_15(*args, **kwargs)#
Version
name: CastLike (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
The operator casts the elements of a given input tensor (the first input) to the same data type as the elements of the second input tensor. See documentation of the Cast operator for further details.
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
target_type (heterogeneous) - T2: The (first) input tensor will be cast to produce a tensor of the same type as this (second input) tensor.
Outputs
output (heterogeneous) - T2: Output tensor produced by casting the first input tensor to have the same type as the second input tensor.
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from complex is not supported.
T2 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to complex is not supported.
OnnxCast_1#
- class mlprodict.npy.xop_auto_import_.OnnxCast_1(*args, **kwargs)#
Version
name: Cast (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
The operator casts the elements of a given input tensor to a data type specified by the ‘to’ argument and returns an output tensor of the same size in the converted type. The ‘to’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message. NOTE: Casting to and from strings is not supported yet.
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
Outputs
output (heterogeneous) - T2: Output tensor with the same shape as input with type specified by the ‘to’ argument
Type Constraints
T1 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from strings and complex are not supported.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to strings and complex are not supported.
OnnxCast_13#
- class mlprodict.npy.xop_auto_import_.OnnxCast_13(*args, **kwargs)#
Version
name: Cast (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator casts the elements of a given input tensor to a data type specified by the ‘to’ argument and returns an output tensor of the same size in the converted type. The ‘to’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Casting from string tensor in plain (e.g., “3.14” and “1000”) and scientific numeric representations (e.g., “1e-5” and “1E8”) to float types is supported. For example, converting string “100.5” to an integer may result 100. There are some string literals reserved for special floating-point values; “+INF” (and “INF”), “-INF”, and “NaN” are positive infinity, negative infinity, and not-a-number, respectively. Any string which can exactly match “+INF” in a case-insensitive way would be mapped to positive infinite. Similarly, this case-insensitive rule is applied to “INF” and “NaN”. When casting from numeric tensors to string tensors, plain floating-point representation (such as “314.15926”) would be used. Converting non-numerical-literal string such as “Hello World!” is an undefined behavior. Cases of converting string representing floating-point arithmetic value, such as “2.718”, to INT is an undefined behavior.
Conversion from a numerical type to any numerical type is always allowed. User must be aware of precision loss and value change caused by range difference between two types. For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting an integer 36 to Boolean may produce 1 because we truncate bits which can’t be stored in the targeted type.
In more detail, the conversion among numerical types should follow these rules:
Casting from floating point to: * floating point: +/- infinity if OOR (out of range). * fixed point: undefined if OOR. * bool: +/- 0.0 to False; all else to True.
Casting from fixed point to: * floating point: +/- infinity if OOR. (+ infinity in the case of uint) * fixed point: when OOR, discard higher bits and reinterpret (with respect to two’s complement representation for
- signed types). For example, 200 (int16) -> -56 (int8).
bool: zero to False; nonzero to True.
Casting from bool to: * floating point: {1.0, 0.0}. * fixed point: {1, 0}. * bool: no change.
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
Outputs
output (heterogeneous) - T2: Output tensor with the same shape as input with type specified by the ‘to’ argument
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from complex is not supported.
T2 in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to complex is not supported.
OnnxCast_6#
- class mlprodict.npy.xop_auto_import_.OnnxCast_6(*args, **kwargs)#
Version
name: Cast (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
The operator casts the elements of a given input tensor to a data type specified by the ‘to’ argument and returns an output tensor of the same size in the converted type. The ‘to’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message. NOTE: Casting to and from strings is not supported yet.
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
Outputs
output (heterogeneous) - T2: Output tensor with the same shape as input with type specified by the ‘to’ argument
Type Constraints
T1 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from strings and complex are not supported.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to strings and complex are not supported.
OnnxCast_9#
- class mlprodict.npy.xop_auto_import_.OnnxCast_9(*args, **kwargs)#
Version
name: Cast (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
The operator casts the elements of a given input tensor to a data type specified by the ‘to’ argument and returns an output tensor of the same size in the converted type. The ‘to’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Casting from string tensor in plain (e.g., “3.14” and “1000”) and scientific numeric representations (e.g., “1e-5” and “1E8”) to float types is supported. For example, converting string “100.5” to an integer may result 100. There are some string literals reserved for special floating-point values; “+INF” (and “INF”), “-INF”, and “NaN” are positive infinity, negative infinity, and not-a-number, respectively. Any string which can exactly match “+INF” in a case-insensitive way would be mapped to positive infinite. Similarly, this case-insensitive rule is applied to “INF” and “NaN”. When casting from numeric tensors to string tensors, plain floating-point representation (such as “314.15926”) would be used. Converting non-numerical-literal string such as “Hello World!” is an undefined behavior. Cases of converting string representing floating-point arithmetic value, such as “2.718”, to INT is an undefined behavior.
Conversion from a numerical type to any numerical type is always allowed. User must be aware of precision loss and value change caused by range difference between two types. For example, a 64-bit float 3.1415926459 may be round to a 32-bit float 3.141592. Similarly, converting an integer 36 to Boolean may produce 1 because we truncate bits which can’t be stored in the targeted type.
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto
Inputs
input (heterogeneous) - T1: Input tensor to be cast.
Outputs
output (heterogeneous) - T2: Output tensor with the same shape as input with type specified by the ‘to’ argument
Type Constraints
T1 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Casting from complex is not supported.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Casting to complex is not supported.
OnnxCeil#
- class mlprodict.npy.xop_auto_import_.OnnxCeil(*args, **kwargs)#
Version
name: Ceil (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Ceil takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the ceil is, y = ceil(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCeil_1#
- class mlprodict.npy.xop_auto_import_.OnnxCeil_1(*args, **kwargs)#
Version
name: Ceil (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Ceil takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the ceil is, y = ceil(x), is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCeil_13#
- class mlprodict.npy.xop_auto_import_.OnnxCeil_13(*args, **kwargs)#
Version
name: Ceil (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Ceil takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the ceil is, y = ceil(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCeil_6#
- class mlprodict.npy.xop_auto_import_.OnnxCeil_6(*args, **kwargs)#
Version
name: Ceil (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Ceil takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the ceil is, y = ceil(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCelu#
- class mlprodict.npy.xop_auto_import_.OnnxCelu(*args, **kwargs)#
Version
name: Celu (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Continuously Differentiable Exponential Linear Units: Perform the linear unit element-wise on the input tensor X using formula:
max(0,x) + min(0,alpha*(exp(x/alpha)-1))
Attributes
alpha: The Alpha value in Celu formula which control the shape of the unit. The default value is 1.0. Default value is
1.0.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(float) ): Constrain input and output types to float32 tensors.
OnnxCelu_12#
- class mlprodict.npy.xop_auto_import_.OnnxCelu_12(*args, **kwargs)#
Version
name: Celu (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Continuously Differentiable Exponential Linear Units: Perform the linear unit element-wise on the input tensor X using formula:
max(0,x) + min(0,alpha*(exp(x/alpha)-1))
Attributes
alpha: The Alpha value in Celu formula which control the shape of the unit. The default value is 1.0. Default value is
1.0.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(float) ): Constrain input and output types to float32 tensors.
OnnxCenterCropPad#
- class mlprodict.npy.xop_auto_import_.OnnxCenterCropPad(*args, **kwargs)#
Version
name: CenterCropPad (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Center crop or pad an input to given dimensions.
The crop/pad dimensions can be specified for a subset of the axes. Non-specified dimensions will not be cropped or padded.
If the input dimensions are bigger than the crop shape, a centered cropping window is extracted from the input. If the input dimensions are smaller than the crop shape, the input is padded on each side equally, so that the input is centered in the output.
Attributes
axes: If provided, it specifies a subset of axes that ‘shape’ refer to. If not provided, all axes are assumed [0, 1, …, r-1], where r = rank(data). Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
Inputs
input_data (heterogeneous) - T: Input to extract the centered crop from.
shape (heterogeneous) - Tind: 1-D tensor representing the cropping window dimensions.
Outputs
output_data (heterogeneous) - T: Output data.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxCenterCropPad_18#
- class mlprodict.npy.xop_auto_import_.OnnxCenterCropPad_18(*args, **kwargs)#
Version
name: CenterCropPad (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Center crop or pad an input to given dimensions.
The crop/pad dimensions can be specified for a subset of the axes. Non-specified dimensions will not be cropped or padded.
If the input dimensions are bigger than the crop shape, a centered cropping window is extracted from the input. If the input dimensions are smaller than the crop shape, the input is padded on each side equally, so that the input is centered in the output.
Attributes
axes: If provided, it specifies a subset of axes that ‘shape’ refer to. If not provided, all axes are assumed [0, 1, …, r-1], where r = rank(data). Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
Inputs
input_data (heterogeneous) - T: Input to extract the centered crop from.
shape (heterogeneous) - Tind: 1-D tensor representing the cropping window dimensions.
Outputs
output_data (heterogeneous) - T: Output data.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxClip#
- class mlprodict.npy.xop_auto_import_.OnnxClip(*args, **kwargs)#
Version
name: Clip (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Clip operator limits the given input within an interval. The interval is specified by the inputs ‘min’ and ‘max’. They default to numeric_limits::lowest() and numeric_limits::max(), respectively.
Inputs
Between 1 and 3 inputs.
input (heterogeneous) - T: Input tensor whose elements to be clipped
min (optional, heterogeneous) - T: Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
max (optional, heterogeneous) - T: Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
Outputs
output (heterogeneous) - T: Output tensor with clipped input elements
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxClip_1#
- class mlprodict.npy.xop_auto_import_.OnnxClip_1(*args, **kwargs)#
Version
name: Clip (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Clip operator limits the given input within an interval. The interval is specified with arguments ‘min’ and ‘max’. They default to numeric_limits::lowest() and numeric_limits::max() respectively.
Attributes
consumed_inputs: legacy optimization attribute.
max: Maximum value, above which element is replaced by max
min: Minimum value, under which element is replaced by min
Inputs
input (heterogeneous) - T: Input tensor whose elements to be clipped
Outputs
output (heterogeneous) - T: Output tensor with clipped input elements
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxClip_11#
- class mlprodict.npy.xop_auto_import_.OnnxClip_11(*args, **kwargs)#
Version
name: Clip (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Clip operator limits the given input within an interval. The interval is specified by the inputs ‘min’ and ‘max’. They default to numeric_limits::lowest() and numeric_limits::max(), respectively.
Inputs
Between 1 and 3 inputs.
input (heterogeneous) - T: Input tensor whose elements to be clipped
min (optional, heterogeneous) - T: Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
max (optional, heterogeneous) - T: Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
Outputs
output (heterogeneous) - T: Output tensor with clipped input elements
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxClip_12#
- class mlprodict.npy.xop_auto_import_.OnnxClip_12(*args, **kwargs)#
Version
name: Clip (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Clip operator limits the given input within an interval. The interval is specified by the inputs ‘min’ and ‘max’. They default to numeric_limits::lowest() and numeric_limits::max(), respectively.
Inputs
Between 1 and 3 inputs.
input (heterogeneous) - T: Input tensor whose elements to be clipped
min (optional, heterogeneous) - T: Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
max (optional, heterogeneous) - T: Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
Outputs
output (heterogeneous) - T: Output tensor with clipped input elements
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxClip_13#
- class mlprodict.npy.xop_auto_import_.OnnxClip_13(*args, **kwargs)#
Version
name: Clip (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Clip operator limits the given input within an interval. The interval is specified by the inputs ‘min’ and ‘max’. They default to numeric_limits::lowest() and numeric_limits::max(), respectively.
Inputs
Between 1 and 3 inputs.
input (heterogeneous) - T: Input tensor whose elements to be clipped
min (optional, heterogeneous) - T: Minimum value, under which element is replaced by min. It must be a scalar(tensor of empty shape).
max (optional, heterogeneous) - T: Maximum value, above which element is replaced by max. It must be a scalar(tensor of empty shape).
Outputs
output (heterogeneous) - T: Output tensor with clipped input elements
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxClip_6#
- class mlprodict.npy.xop_auto_import_.OnnxClip_6(*args, **kwargs)#
Version
name: Clip (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Clip operator limits the given input within an interval. The interval is specified with arguments ‘min’ and ‘max’. They default to numeric_limits::lowest() and numeric_limits::max() respectively.
Attributes
max: Maximum value, above which element is replaced by max Default value is
3.4028234663852886e+38.min: Minimum value, under which element is replaced by min Default value is
-3.4028234663852886e+38.
Inputs
input (heterogeneous) - T: Input tensor whose elements to be clipped
Outputs
output (heterogeneous) - T: Output tensor with clipped input elements
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCol2Im#
- class mlprodict.npy.xop_auto_import_.OnnxCol2Im(*args, **kwargs)#
Version
name: Col2Im (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
The operator rearranges column blocks back into a multidimensional image
Col2Im behaves similarly to PyTorch’s fold https://pytorch.org/docs/stable/generated/torch.nn.Fold.html, but it only supports batched multi-dimensional image tensors. Another implementation in Python with N-dimension support can be found at f-dangel/unfoldNd.
- NOTE: Although specifying image_shape looks redundant because it could be calculated from
convolution formulas, it is required as input for more advanced scenarios as explained at PyTorch’s implementation (pytorch/pytorch)
Attributes
dilations: 1-dimensional tensor with dilation value along each spatial axis of the image. If not present, the dilation defaults to 1 along each spatial axis of the image.
pads: 1-dimensional tensor with padding value for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin is the number of pixels added at the beginning of axis i and xi_end is the number of pixels added at the end of axis i. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: 1-dimensional tensor with stride value along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
input (heterogeneous) - T: Input data tensor to be rearranged from column blocks back into an image. This is a 3-dimensional tensor containing [N, C * n-ary- product(block_shape), L], where N is batch dimension, C is image channel dimension and L is number of blocks.The blocks are enumerated in increasing lexicographic-order of their indices.For example, with an image-size 10*20 and block-size 9*18, there would be 2*3 blocks, enumerated in the order block(0, 0), block(0, 1), block(0, 2), block(1, 0), block(1, 1), block(1, 2).
image_shape (heterogeneous) - tensor(int64): The shape of the spatial dimensions of the image after rearranging the column blocks.This is a 1-dimensional tensor with size of at least 2, containing the value [H_img, W_img] for a 2-D image or [dim_i1, dim_i2, …, dim_iN] for a N-D image.
block_shape (heterogeneous) - tensor(int64): The shape of the block to apply on the input.This is a 1-dimensional tensor of size of at least 2, containing the value [H_block, W_block] for a 2-D image or [dim_b1, dim_b2, …, dim_bN] for a N-D block.This is the block-shape before dilation is applied to it.
Outputs
output (heterogeneous) - T: Output tensor produced by rearranging blocks into an image.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensor types.
OnnxCol2Im_18#
- class mlprodict.npy.xop_auto_import_.OnnxCol2Im_18(*args, **kwargs)#
Version
name: Col2Im (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
The operator rearranges column blocks back into a multidimensional image
Col2Im behaves similarly to PyTorch’s fold https://pytorch.org/docs/stable/generated/torch.nn.Fold.html, but it only supports batched multi-dimensional image tensors. Another implementation in Python with N-dimension support can be found at f-dangel/unfoldNd.
- NOTE: Although specifying image_shape looks redundant because it could be calculated from
convolution formulas, it is required as input for more advanced scenarios as explained at PyTorch’s implementation (pytorch/pytorch)
Attributes
dilations: 1-dimensional tensor with dilation value along each spatial axis of the image. If not present, the dilation defaults to 1 along each spatial axis of the image.
pads: 1-dimensional tensor with padding value for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin is the number of pixels added at the beginning of axis i and xi_end is the number of pixels added at the end of axis i. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: 1-dimensional tensor with stride value along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
input (heterogeneous) - T: Input data tensor to be rearranged from column blocks back into an image. This is a 3-dimensional tensor containing [N, C * n-ary- product(block_shape), L], where N is batch dimension, C is image channel dimension and L is number of blocks.The blocks are enumerated in increasing lexicographic-order of their indices.For example, with an image-size 10*20 and block-size 9*18, there would be 2*3 blocks, enumerated in the order block(0, 0), block(0, 1), block(0, 2), block(1, 0), block(1, 1), block(1, 2).
image_shape (heterogeneous) - tensor(int64): The shape of the spatial dimensions of the image after rearranging the column blocks.This is a 1-dimensional tensor with size of at least 2, containing the value [H_img, W_img] for a 2-D image or [dim_i1, dim_i2, …, dim_iN] for a N-D image.
block_shape (heterogeneous) - tensor(int64): The shape of the block to apply on the input.This is a 1-dimensional tensor of size of at least 2, containing the value [H_block, W_block] for a 2-D image or [dim_b1, dim_b2, …, dim_bN] for a N-D block.This is the block-shape before dilation is applied to it.
Outputs
output (heterogeneous) - T: Output tensor produced by rearranging blocks into an image.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensor types.
OnnxComMicrosoftAdamOptimizer#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAdamOptimizer(*args, **kwargs)#
Version
name: AdamOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
alpha: Coefficient of previous gradient in running average. Default value is
?.beta: Coefficient of previous squared gradient in running average.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn’t reduce the learning rate. Default value is
?.do_bias_correction: Compute unbiased 1st and 2nd momentums. Default value is
?.epsilon: Small scalar to avoid dividing by zero. Default value is
?.lambda: Regularization coefficient of 0.5 * lambda * ||X||_2^2. Default to 0, which means no regularization. Default value is
?.max_norm_clip: clip threshold of gradients. Default value is
?.weight_decay_mode: Modes for applying weight decay, 0 means applying decay before weight update, 1 means applying decay after weight update. Default value is
?.
Inputs
Between 6 and 10 inputs.
R (heterogeneous) - T1: The initial learning rate.
T (heterogeneous) - T2: The update count of “X”. It should be a scalar.
weights (heterogeneous) - T3: weights to optimize.
gradients (heterogeneous) - T_GRAD: gradients computed in this iteration.
moment_1 (heterogeneous) - T4: exponentially averaged historical gradients.
moment_2 (heterogeneous) - T4: exponentially averaged historical squared gradients.
mixed_precision_weights (optional, heterogeneous) - T_MIXED_PRECISION_FP: FP16 or BFloat16 weights to optimize.
loss_scale (optional, heterogeneous) - T3: loss scale for mixed precision training
global_gradient_norm (optional, heterogeneous) - T_GRAD_NORM: Global gradient norm.
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if weight tensors should be updated.
Outputs
Between 3 and 6 outputs.
new_T (heterogeneous) - T2: New update count.
new_moment_1 (heterogeneous) - T4: New averaged gradients.
new_moment_2 (heterogeneous) - T4: New averaged squared gradients.
new_weights (optional, heterogeneous) - T3: New weights.
new_gradients (optional, heterogeneous) - T_GRAD: New gradients.
new_mixed_precision_weights (optional, heterogeneous) - T_MIXED_PRECISION_FP: New FP16 or BFloat16 weights
OnnxComMicrosoftAdamOptimizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAdamOptimizer_1(*args, **kwargs)#
Version
name: AdamOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
alpha: Coefficient of previous gradient in running average. Default value is
?.beta: Coefficient of previous squared gradient in running average.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn’t reduce the learning rate. Default value is
?.do_bias_correction: Compute unbiased 1st and 2nd momentums. Default value is
?.epsilon: Small scalar to avoid dividing by zero. Default value is
?.lambda: Regularization coefficient of 0.5 * lambda * ||X||_2^2. Default to 0, which means no regularization. Default value is
?.max_norm_clip: clip threshold of gradients. Default value is
?.weight_decay_mode: Modes for applying weight decay, 0 means applying decay before weight update, 1 means applying decay after weight update. Default value is
?.
Inputs
Between 6 and 10 inputs.
R (heterogeneous) - T1: The initial learning rate.
T (heterogeneous) - T2: The update count of “X”. It should be a scalar.
weights (heterogeneous) - T3: weights to optimize.
gradients (heterogeneous) - T_GRAD: gradients computed in this iteration.
moment_1 (heterogeneous) - T4: exponentially averaged historical gradients.
moment_2 (heterogeneous) - T4: exponentially averaged historical squared gradients.
mixed_precision_weights (optional, heterogeneous) - T_MIXED_PRECISION_FP: FP16 or BFloat16 weights to optimize.
loss_scale (optional, heterogeneous) - T3: loss scale for mixed precision training
global_gradient_norm (optional, heterogeneous) - T_GRAD_NORM: Global gradient norm.
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if weight tensors should be updated.
Outputs
Between 3 and 6 outputs.
new_T (heterogeneous) - T2: New update count.
new_moment_1 (heterogeneous) - T4: New averaged gradients.
new_moment_2 (heterogeneous) - T4: New averaged squared gradients.
new_weights (optional, heterogeneous) - T3: New weights.
new_gradients (optional, heterogeneous) - T_GRAD: New gradients.
new_mixed_precision_weights (optional, heterogeneous) - T_MIXED_PRECISION_FP: New FP16 or BFloat16 weights
OnnxComMicrosoftAdamWOptimizer#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAdamWOptimizer(*args, **kwargs)#
Version
name: AdamWOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
adam_mode: Modes for applying bias correction and weight decay (default 0) 0 : Weight decay is applied before weight is updated. Computation aligned with Torch AdamW. In this mode, correct_bias should be 1 to keep aligned with PyTorch.1 : Weight decay is applied after weight is updated. Computation is aligned with Huggingface AdamW. Default value is
?.alpha: Coefficient of previously accumulated gradient in running average. Default value is
?.beta: Coefficient of previously accumulated squared-gradient in running average. Default value is
?.correct_bias: Whether or not to correct bias, enabled by default. Default value is
?.epsilon: Small scalar to avoid dividing by zero. Default value is
?.weight_decay: weight decay coefficient. Default value is
?.
Inputs
Between 6 and 7 inputs.
lr (heterogeneous) - T1: The learning rate.
step (heterogeneous) - T2: The update count of weights. It should be a scalar.
weights (heterogeneous) - S_WEIGHT: Sequence of weights to optimize.
gradients (heterogeneous) - S_GRAD: Sequence of gradients computed in this iteration.
momentums_1 (heterogeneous) - S_MOMENT: Sequence of exponentially averaged historical gradients.
momentums_2 (heterogeneous) - S_MOMENT: Sequence of exponentially averaged historical squared gradients.
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if weight updates are skipped, applicable to gradient infinity check in mixed precision training.
Outputs
Between 1 and 4 outputs.
updated_flag (heterogeneous) - T2: Whether gradient is applied or not.
updated_weights (optional, heterogeneous) - S_WEIGHT: Sequence of weights after optimize.
updated_momentums_1 (optional, heterogeneous) - S_MOMENT: Sequence of momentum_1 after optimize.
updated_momentums_2 (optional, heterogeneous) - S_MOMENT: Sequence of momentum_2 after optimize.
OnnxComMicrosoftAdamWOptimizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAdamWOptimizer_1(*args, **kwargs)#
Version
name: AdamWOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
adam_mode: Modes for applying bias correction and weight decay (default 0) 0 : Weight decay is applied before weight is updated. Computation aligned with Torch AdamW. In this mode, correct_bias should be 1 to keep aligned with PyTorch.1 : Weight decay is applied after weight is updated. Computation is aligned with Huggingface AdamW. Default value is
?.alpha: Coefficient of previously accumulated gradient in running average. Default value is
?.beta: Coefficient of previously accumulated squared-gradient in running average. Default value is
?.correct_bias: Whether or not to correct bias, enabled by default. Default value is
?.epsilon: Small scalar to avoid dividing by zero. Default value is
?.weight_decay: weight decay coefficient. Default value is
?.
Inputs
Between 6 and 7 inputs.
lr (heterogeneous) - T1: The learning rate.
step (heterogeneous) - T2: The update count of weights. It should be a scalar.
weights (heterogeneous) - S_WEIGHT: Sequence of weights to optimize.
gradients (heterogeneous) - S_GRAD: Sequence of gradients computed in this iteration.
momentums_1 (heterogeneous) - S_MOMENT: Sequence of exponentially averaged historical gradients.
momentums_2 (heterogeneous) - S_MOMENT: Sequence of exponentially averaged historical squared gradients.
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if weight updates are skipped, applicable to gradient infinity check in mixed precision training.
Outputs
Between 1 and 4 outputs.
updated_flag (heterogeneous) - T2: Whether gradient is applied or not.
updated_weights (optional, heterogeneous) - S_WEIGHT: Sequence of weights after optimize.
updated_momentums_1 (optional, heterogeneous) - S_MOMENT: Sequence of momentum_1 after optimize.
updated_momentums_2 (optional, heterogeneous) - S_MOMENT: Sequence of momentum_2 after optimize.
OnnxComMicrosoftAdasumAllReduce#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAdasumAllReduce(*args, **kwargs)#
Version
name: AdasumAllReduce (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
reduce_algo: Algorithms for Adasum. Valid values are: CpuReduction(1) or GpuHierarchicalReduction(2) Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be reduced
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: reduced tensors
OnnxComMicrosoftAdasumAllReduce_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAdasumAllReduce_1(*args, **kwargs)#
Version
name: AdasumAllReduce (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
reduce_algo: Algorithms for Adasum. Valid values are: CpuReduction(1) or GpuHierarchicalReduction(2) Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be reduced
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: reduced tensors
OnnxComMicrosoftAll#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAll(*args, **kwargs)#
Version
name: All (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Return true if all elements are true and false otherwise.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - T: output.
OnnxComMicrosoftAll_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAll_1(*args, **kwargs)#
Version
name: All (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Return true if all elements are true and false otherwise.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - T: output.
OnnxComMicrosoftAttention#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAttention(*args, **kwargs)#
Version
name: Attention (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Multi-Head Self Attention that can be either unidirectional (like GPT-2) or bidirectional (like BERT). The mask_index input is optional. Besides raw attention mask with shape (batch_size, past_sequence_length + sequence_length) or (batch_size, sequence_length, past_sequence_length + sequence_length) with value 0 for masked and 1 otherwise, we also support other two formats: When input has right-side padding, mask_index is one dimension with shape (batch_size), where value of each element is the end position, or valid length of actual sequence excluding padding. When input has left-side padding, mask_index has shape (2 * batch_size), where the values are the exclusive end positions followed by the inclusive start positions. When unidirectional is 1, and each token only attend to previous tokens. For GPT-2, both past and present state are optional. Present state could appear in output even when past state is not in input.
Attributes
num_heads (required): Number of attention heads Default value is
?.qkv_hidden_sizes: Hidden layer sizes of Q, K, V paths in Attention Default value is
?.unidirectional: Whether every token can only attend to previous tokens. Default value is 0. Default value is
?.
Inputs
Between 3 and 6 inputs.
input (heterogeneous) - T: 3D input tensor with shape (batch_size, sequence_length, input_hidden_size)
weight (heterogeneous) - T: 2D input tensor with shape (input_hidden_size, 3 * hidden_size), where hidden_size = num_heads * head_size
bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
mask_index (optional, heterogeneous) - M: Attention mask with shape (batch_size, 1, max_sequence_length, max_sequence_length), (batch_size, past_sequence_length + sequence_length)or (batch_size, sequence_length, past_sequence_length + sequence_length), or index with shape (batch_size) or (2 * batch_size).
past (optional, heterogeneous) - T: past state for key and value with shape (2, batch_size, num_heads, past_sequence_length, head_size).
extra_add (optional, heterogeneous) - T: additional add to QxK’ with shape (batch_size, num_heads, sequence_length, sequence_length).
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
present (optional, heterogeneous) - T: present state for key and value with shape (2, batch_size, num_heads, past_sequence_length + sequence_length, head_size)
OnnxComMicrosoftAttention_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAttention_1(*args, **kwargs)#
Version
name: Attention (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Multi-Head Self Attention that can be either unidirectional (like GPT-2) or bidirectional (like BERT). The mask_index input is optional. Besides raw attention mask with shape (batch_size, past_sequence_length + sequence_length) or (batch_size, sequence_length, past_sequence_length + sequence_length) with value 0 for masked and 1 otherwise, we also support other two formats: When input has right-side padding, mask_index is one dimension with shape (batch_size), where value of each element is the end position, or valid length of actual sequence excluding padding. When input has left-side padding, mask_index has shape (2 * batch_size), where the values are the exclusive end positions followed by the inclusive start positions. When unidirectional is 1, and each token only attend to previous tokens. For GPT-2, both past and present state are optional. Present state could appear in output even when past state is not in input.
Attributes
num_heads (required): Number of attention heads Default value is
?.qkv_hidden_sizes: Hidden layer sizes of Q, K, V paths in Attention Default value is
?.unidirectional: Whether every token can only attend to previous tokens. Default value is 0. Default value is
?.
Inputs
Between 3 and 6 inputs.
input (heterogeneous) - T: 3D input tensor with shape (batch_size, sequence_length, input_hidden_size)
weight (heterogeneous) - T: 2D input tensor with shape (input_hidden_size, 3 * hidden_size), where hidden_size = num_heads * head_size
bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
mask_index (optional, heterogeneous) - M: Attention mask with shape (batch_size, 1, max_sequence_length, max_sequence_length), (batch_size, past_sequence_length + sequence_length)or (batch_size, sequence_length, past_sequence_length + sequence_length), or index with shape (batch_size) or (2 * batch_size).
past (optional, heterogeneous) - T: past state for key and value with shape (2, batch_size, num_heads, past_sequence_length, head_size).
extra_add (optional, heterogeneous) - T: additional add to QxK’ with shape (batch_size, num_heads, sequence_length, sequence_length).
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
present (optional, heterogeneous) - T: present state for key and value with shape (2, batch_size, num_heads, past_sequence_length + sequence_length, head_size)
OnnxComMicrosoftAttnLSTM#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAttnLSTM(*args, **kwargs)#
Version
name: AttnLSTM (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes an one-layer RNN where its RNN Cell is an AttentionWrapper wrapped a LSTM Cell. The RNN layer contains following basic component: LSTM Cell, Bahdanau Attention Mechanism, AttentionWrapp.
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Softmax(x) - exp(x) / sum(exp(x))
- Bahdanau Attention Mechanism:
M - Memory tensor.
VALUES - masked Memory by its real sequence length.
MW - Memory layer weight.
- KEYS - Processed memory tensor by the memory layer.
KEYS = M * MW
Query - Query tensor, normally at specific time step in sequence.
QW - Query layer weight in the attention mechanism
PQ - processed query, = Query * QW
`V’ - attention vector
- ALIGN - calculated alignment based on Query and KEYS
ALIGN = softmax(reduce_sum(V * Tanh(KEYS + PQ)))
- CONTEXT - context based on ALIGN and VALUES
CONTEXT = ALIGN * VALUES
- LSTM Cell:
X - input tensor concat with attention state in the attention wrapper
i - input gate
o - output gate
f - forget gate
c - cell gate
t - time step (t-1 means previous time step)
W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
Wb[iofc] - W bias vectors for input, output, forget, and cell gates
Rb[iofc] - R bias vectors for input, output, forget, and cell gates
P[iof] - P peephole weight vector for input, output, and forget gates
WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
PB[iof] - P peephole weight vector for backward input, output, and forget gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
Ct = ft (.) Ct-1 + it (.) ct
ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo)
Ht = ot (.) h(Ct)
- AttentionWrapp Notations:
- `lstm()’ - wrapped inner cell.
Ht, Ct = lstm(concat(Xt, ATTNt-1), Ct-1)
- am() - attention mechanism the wrapper used.
CONTEXTt, ALIGNt = am(Ht, ALIGNt-1)
AW - attention layer weights, optional.
- ATTN - attention state, initial is zero. If AW provided, it is the output of the attention layer,
ATTNt = concat(Ht, CONTEXTt) * AW
- otherwise,
ATTNt = CONTEXTt
- RNN layer output:
Y - if needed is the sequence of Ht from lstm cell.
Y_h - is the last valid H from lstm cell.
Y_c - is the last valid C from lstm cell.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01. Default value is
?.activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators. Default value is
?.activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified. Default value is
?.clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified. Default value is
?.direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
?.hidden_size: Number of neurons in the hidden layer. Default value is
?.input_forget: Couple the input and forget gates if 1, default 0. Default value is
?.
Inputs
Between 3 and 14 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size]
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, 4*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 4*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size]
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
QW (optional, heterogeneous) - T: The weight tensor of the query layer in the attention mechanism. Should be of shape [num_directions, am_query_depth(hidden_size of lstm), am_attn_size]
MW (optional, heterogeneous) - T: The weight tensor of the memory layer in the attention mechanism. Should be of shape [num_directions, memory_depth, am_attn_size]
V (optional, heterogeneous) - T: The attention_v tensor in the attention mechanism. Should be of shape [num_directions, am_attn_size]
M (optional, heterogeneous) - T: The sequence of the memory (input) for attention mechanism. Should be of [batch_size, max_memory_step, memory_depth]
memory_seq_lens (optional, heterogeneous) - T1: The sequence length of the input memory for the attention mechanism. Should be of [batch_size]
AW (optional, heterogeneous) - T: The weights of attention layer in the attention wrapper. If exists, should be of shape [num_directions, memory_depth+hidden_size, aw_attn_size]. Please note that attention mechanism context depth is also memory_depth in the attention mechanism.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
OnnxComMicrosoftAttnLSTM_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftAttnLSTM_1(*args, **kwargs)#
Version
name: AttnLSTM (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes an one-layer RNN where its RNN Cell is an AttentionWrapper wrapped a LSTM Cell. The RNN layer contains following basic component: LSTM Cell, Bahdanau Attention Mechanism, AttentionWrapp.
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Softmax(x) - exp(x) / sum(exp(x))
- Bahdanau Attention Mechanism:
M - Memory tensor.
VALUES - masked Memory by its real sequence length.
MW - Memory layer weight.
- KEYS - Processed memory tensor by the memory layer.
KEYS = M * MW
Query - Query tensor, normally at specific time step in sequence.
QW - Query layer weight in the attention mechanism
PQ - processed query, = Query * QW
`V’ - attention vector
- ALIGN - calculated alignment based on Query and KEYS
ALIGN = softmax(reduce_sum(V * Tanh(KEYS + PQ)))
- CONTEXT - context based on ALIGN and VALUES
CONTEXT = ALIGN * VALUES
- LSTM Cell:
X - input tensor concat with attention state in the attention wrapper
i - input gate
o - output gate
f - forget gate
c - cell gate
t - time step (t-1 means previous time step)
W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
Wb[iofc] - W bias vectors for input, output, forget, and cell gates
Rb[iofc] - R bias vectors for input, output, forget, and cell gates
P[iof] - P peephole weight vector for input, output, and forget gates
WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
PB[iof] - P peephole weight vector for backward input, output, and forget gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
Ct = ft (.) Ct-1 + it (.) ct
ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo)
Ht = ot (.) h(Ct)
- AttentionWrapp Notations:
- `lstm()’ - wrapped inner cell.
Ht, Ct = lstm(concat(Xt, ATTNt-1), Ct-1)
- am() - attention mechanism the wrapper used.
CONTEXTt, ALIGNt = am(Ht, ALIGNt-1)
AW - attention layer weights, optional.
- ATTN - attention state, initial is zero. If AW provided, it is the output of the attention layer,
ATTNt = concat(Ht, CONTEXTt) * AW
- otherwise,
ATTNt = CONTEXTt
- RNN layer output:
Y - if needed is the sequence of Ht from lstm cell.
Y_h - is the last valid H from lstm cell.
Y_c - is the last valid C from lstm cell.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01. Default value is
?.activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators. Default value is
?.activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified. Default value is
?.clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified. Default value is
?.direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
?.hidden_size: Number of neurons in the hidden layer. Default value is
?.input_forget: Couple the input and forget gates if 1, default 0. Default value is
?.
Inputs
Between 3 and 14 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size]
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, 4*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 4*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size]
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
QW (optional, heterogeneous) - T: The weight tensor of the query layer in the attention mechanism. Should be of shape [num_directions, am_query_depth(hidden_size of lstm), am_attn_size]
MW (optional, heterogeneous) - T: The weight tensor of the memory layer in the attention mechanism. Should be of shape [num_directions, memory_depth, am_attn_size]
V (optional, heterogeneous) - T: The attention_v tensor in the attention mechanism. Should be of shape [num_directions, am_attn_size]
M (optional, heterogeneous) - T: The sequence of the memory (input) for attention mechanism. Should be of [batch_size, max_memory_step, memory_depth]
memory_seq_lens (optional, heterogeneous) - T1: The sequence length of the input memory for the attention mechanism. Should be of [batch_size]
AW (optional, heterogeneous) - T: The weights of attention layer in the attention wrapper. If exists, should be of shape [num_directions, memory_depth+hidden_size, aw_attn_size]. Please note that attention mechanism context depth is also memory_depth in the attention mechanism.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
OnnxComMicrosoftBatchNormInternal#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBatchNormInternal(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Variant of BatchNormalization with additional output for saved_mean/inv_std_dev.
Attributes
epsilon: epsilon value Default value is
?.momentum: momentum value Default value is
?.training_mode: true if training Default value is
?.
Inputs
X (heterogeneous) - T: Input tensor.
scale (heterogeneous) - T1: Scale tensor of shape (C).
B (heterogeneous) - T1: Bias tensor of shape (C).
input_mean (heterogeneous) - T2: running mean tensor of shape (C).
input_var (heterogeneous) - T2: running variance tensor of shape (C).
Outputs
Between 1 and 5 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
running_mean (optional, heterogeneous) - T2: The running mean after BN.
running_var (optional, heterogeneous) - T2: Running var after BN
saved_mean (optional, heterogeneous) - T2: Mean of the batch
saved_inv_std (optional, heterogeneous) - T2: Inverse standard deviation for the batch
OnnxComMicrosoftBatchNormInternal_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBatchNormInternal_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Variant of BatchNormalization with additional output for saved_mean/inv_std_dev.
Attributes
epsilon: epsilon value Default value is
?.momentum: momentum value Default value is
?.training_mode: true if training Default value is
?.
Inputs
X (heterogeneous) - T: Input tensor.
scale (heterogeneous) - T1: Scale tensor of shape (C).
B (heterogeneous) - T1: Bias tensor of shape (C).
input_mean (heterogeneous) - T2: running mean tensor of shape (C).
input_var (heterogeneous) - T2: running variance tensor of shape (C).
Outputs
Between 1 and 5 outputs.
Y (heterogeneous) - T: The output tensor of the same shape as X
running_mean (optional, heterogeneous) - T2: The running mean after BN.
running_var (optional, heterogeneous) - T2: Running var after BN
saved_mean (optional, heterogeneous) - T2: Mean of the batch
saved_inv_std (optional, heterogeneous) - T2: Inverse standard deviation for the batch
OnnxComMicrosoftBatchNormalizationGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBatchNormalizationGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
BatchNormalizationGrad
Attributes
epsilon (required): epsilon value Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient output from previous node
X (heterogeneous) - T: Input
scale (heterogeneous) - T1: Scale tensor
mean (heterogeneous) - T2: Mean of X
variance (heterogeneous) - T2: Variance of X
Outputs
X_grad (heterogeneous) - T: Gradient of the input
scale_grad (heterogeneous) - T1: Gradient of the scale
bias_grad (heterogeneous) - T1: Gradient of the bias
OnnxComMicrosoftBatchNormalizationGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBatchNormalizationGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
BatchNormalizationGrad
Attributes
epsilon (required): epsilon value Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient output from previous node
X (heterogeneous) - T: Input
scale (heterogeneous) - T1: Scale tensor
mean (heterogeneous) - T2: Mean of X
variance (heterogeneous) - T2: Variance of X
Outputs
X_grad (heterogeneous) - T: Gradient of the input
scale_grad (heterogeneous) - T1: Gradient of the scale
bias_grad (heterogeneous) - T1: Gradient of the bias
OnnxComMicrosoftBeamSearch#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBeamSearch(*args, **kwargs)#
Version
name: BeamSearch (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Beam Search for text generation. Supports GPT-2 decoder.
Attributes
decoder (required): Decoder subgraph to execute in a loop. Default value is
?.early_stopping: early stop or not Default value is
?.encoder_decoder_init: subgraph for initialization of encoder and decoder. It will be called once before decoder subgraph. Default value is
?.eos_token_id (required): The id of the end-of-sequence token Default value is
?.model_type: model type: 0 for GPT-2; 1 for encoder decoder like T5 Default value is
?.no_repeat_ngram_size: no repeat ngrams size Default value is
?.pad_token_id (required): The id of the padding token Default value is
?.
Inputs
Between 6 and 10 inputs.
input_ids (heterogeneous) - I: The sequence used as a prompt for the generation. Shape is (batch_size, sequence_length)
max_length (heterogeneous) - I: The maximum length of the sequence to be generated. Shape is (1)
min_length (optional, heterogeneous) - I: The minimum length below which the score of eos_token_id is set to -Inf. Shape is (1)
num_beams (heterogeneous) - I: Number of beams for beam search. 1 means no beam search. Shape is (1)
num_return_sequences (heterogeneous) - I: The number of returned sequences in the batch. Shape is (1)
temperature (heterogeneous) - T: The value used to module the next token probabilities. Accepts value > 0.0. Shape is (1)
length_penalty (optional, heterogeneous) - T: Exponential penalty to the length. Default value 1.0 means no penalty.Value > 1.0 encourages longer sequences, while values < 1.0 produces shorter sequences.Shape is (1,)
repetition_penalty (optional, heterogeneous) - T: The parameter for repetition penalty. Default value 1.0 means no penalty. Accepts value > 0.0. Shape is (1)
vocab_mask (optional, heterogeneous) - M: Mask of vocabulary. Words that masked with 0 are not allowed to be generated, and 1 is allowed. Shape is (vacab_size)
prefix_vocab_mask (optional, heterogeneous) - M: Mask of vocabulary for first step. Words that masked with 0 are not allowed to be generated, and 1 is allowed. Shape is (batch_size, vocab_size)
Outputs
Between 1 and 3 outputs.
sequences (heterogeneous) - I: Word IDs of generated sequences. Shape is (batch_size, num_return_sequences, max_sequence_length)
sequences_scores (optional, heterogeneous) - T: Final beam score of the generated sequences. Shape is (batch_size, num_return_sequences)
scores (optional, heterogeneous) - T: Processed beam scores for each vocabulary token at each generation step.Beam scores consisting of log softmax scores for each vocabulary token and sum of log softmax of previously generated tokens in this beam.Shape is (max_length - sequence_length, batch_size, num_beams, vocab_size)
OnnxComMicrosoftBeamSearch_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBeamSearch_1(*args, **kwargs)#
Version
name: BeamSearch (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Beam Search for text generation. Supports GPT-2 decoder.
Attributes
decoder (required): Decoder subgraph to execute in a loop. Default value is
?.early_stopping: early stop or not Default value is
?.encoder_decoder_init: subgraph for initialization of encoder and decoder. It will be called once before decoder subgraph. Default value is
?.eos_token_id (required): The id of the end-of-sequence token Default value is
?.model_type: model type: 0 for GPT-2; 1 for encoder decoder like T5 Default value is
?.no_repeat_ngram_size: no repeat ngrams size Default value is
?.pad_token_id (required): The id of the padding token Default value is
?.
Inputs
Between 6 and 10 inputs.
input_ids (heterogeneous) - I: The sequence used as a prompt for the generation. Shape is (batch_size, sequence_length)
max_length (heterogeneous) - I: The maximum length of the sequence to be generated. Shape is (1)
min_length (optional, heterogeneous) - I: The minimum length below which the score of eos_token_id is set to -Inf. Shape is (1)
num_beams (heterogeneous) - I: Number of beams for beam search. 1 means no beam search. Shape is (1)
num_return_sequences (heterogeneous) - I: The number of returned sequences in the batch. Shape is (1)
temperature (heterogeneous) - T: The value used to module the next token probabilities. Accepts value > 0.0. Shape is (1)
length_penalty (optional, heterogeneous) - T: Exponential penalty to the length. Default value 1.0 means no penalty.Value > 1.0 encourages longer sequences, while values < 1.0 produces shorter sequences.Shape is (1,)
repetition_penalty (optional, heterogeneous) - T: The parameter for repetition penalty. Default value 1.0 means no penalty. Accepts value > 0.0. Shape is (1)
vocab_mask (optional, heterogeneous) - M: Mask of vocabulary. Words that masked with 0 are not allowed to be generated, and 1 is allowed. Shape is (vacab_size)
prefix_vocab_mask (optional, heterogeneous) - M: Mask of vocabulary for first step. Words that masked with 0 are not allowed to be generated, and 1 is allowed. Shape is (batch_size, vocab_size)
Outputs
Between 1 and 3 outputs.
sequences (heterogeneous) - I: Word IDs of generated sequences. Shape is (batch_size, num_return_sequences, max_sequence_length)
sequences_scores (optional, heterogeneous) - T: Final beam score of the generated sequences. Shape is (batch_size, num_return_sequences)
scores (optional, heterogeneous) - T: Processed beam scores for each vocabulary token at each generation step.Beam scores consisting of log softmax scores for each vocabulary token and sum of log softmax of previously generated tokens in this beam.Shape is (max_length - sequence_length, batch_size, num_beams, vocab_size)
OnnxComMicrosoftBiasDropout#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasDropout(*args, **kwargs)#
Version
name: BiasDropout (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
output, dropout_mask = Dropout(data + bias, ratio) + residual, Intended to specialize the dropout pattern commonly found in transformer models.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one. Default value is
?.
Inputs
Between 2 and 5 inputs.
data (heterogeneous) - T: The input data as Tensor.
bias (heterogeneous) - T: The bias input, a vector with the same shape as last dim of data OR same shape with data
residual (optional, heterogeneous) - T: The residual input, must have the same shape as data
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T2: The output mask of dropout.
OnnxComMicrosoftBiasDropout_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasDropout_1(*args, **kwargs)#
Version
name: BiasDropout (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
output, dropout_mask = Dropout(data + bias, ratio) + residual, Intended to specialize the dropout pattern commonly found in transformer models.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one. Default value is
?.
Inputs
Between 2 and 5 inputs.
data (heterogeneous) - T: The input data as Tensor.
bias (heterogeneous) - T: The bias input, a vector with the same shape as last dim of data OR same shape with data
residual (optional, heterogeneous) - T: The residual input, must have the same shape as data
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T2: The output mask of dropout.
OnnxComMicrosoftBiasFastGeluGrad_dX#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasFastGeluGrad_dX(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes dX for FastGeluGrad with bias
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
B (heterogeneous) - T: The bias tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftBiasFastGeluGrad_dX_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasFastGeluGrad_dX_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes dX for FastGeluGrad with bias
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
B (heterogeneous) - T: The bias tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftBiasGelu#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasGelu(*args, **kwargs)#
Version
name: BiasGelu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Bias Gelu. It’s an extension of Gelu. It takes the sum of input A and bias input B as the input of Gelu activation.
Inputs
A (heterogeneous) - T: The normal input data.
B (heterogeneous) - T: The bias input data that is a 1D tensor.
Outputs
C (heterogeneous) - T: The output.
OnnxComMicrosoftBiasGeluGrad_dX#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasGeluGrad_dX(*args, **kwargs)#
Version
name: BiasGeluGrad_dX (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes dX for BiasGeluGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
B (heterogeneous) - T: The bias tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftBiasGeluGrad_dX_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasGeluGrad_dX_1(*args, **kwargs)#
Version
name: BiasGeluGrad_dX (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes dX for BiasGeluGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
B (heterogeneous) - T: The bias tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftBiasGelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasGelu_1(*args, **kwargs)#
Version
name: BiasGelu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Bias Gelu. It’s an extension of Gelu. It takes the sum of input A and bias input B as the input of Gelu activation.
Inputs
A (heterogeneous) - T: The normal input data.
B (heterogeneous) - T: The bias input data that is a 1D tensor.
Outputs
C (heterogeneous) - T: The output.
OnnxComMicrosoftBiasSoftmax#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasSoftmax(*args, **kwargs)#
Version
name: BiasSoftmax (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Y = softmax(scores + bias)) with simple broadcast on bias. Intended to specialize softmax(scores + additive_mask) commonly found in transformer models.
Attributes
broadcast_axis: broadcast bias across input for dimensions broadcast_axis to softmax_axis-1 Default value is
?.softmax_axis: apply softmax to elements for dimensions softmax_axis or higher Default value is
?.
Inputs
data (heterogeneous) - T: The input data as Tensor.
bias (heterogeneous) - T: The bias (or mask) as Tensor.
Outputs
output (heterogeneous) - T: The output.
OnnxComMicrosoftBiasSoftmax_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBiasSoftmax_1(*args, **kwargs)#
Version
name: BiasSoftmax (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Y = softmax(scores + bias)) with simple broadcast on bias. Intended to specialize softmax(scores + additive_mask) commonly found in transformer models.
Attributes
broadcast_axis: broadcast bias across input for dimensions broadcast_axis to softmax_axis-1 Default value is
?.softmax_axis: apply softmax to elements for dimensions softmax_axis or higher Default value is
?.
Inputs
data (heterogeneous) - T: The input data as Tensor.
bias (heterogeneous) - T: The bias (or mask) as Tensor.
Outputs
output (heterogeneous) - T: The output.
OnnxComMicrosoftBifurcationDetector#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBifurcationDetector(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Component for aggressive decoding. Find the bifurcation index of predicted tokens, between source tokens, starting from previous suffix match index, and predicted tokens. Concat predicted tokens, starting from bifurcation index, to the back of current tokens. This forms the output tokens. Detect suffix match index in source tokens, between source tokens and output tokens. Detection is based on finding the appearances of last n-gram in output tokens in source tokens. A match is considered found if source tokens contain a single matching n-gram. Return the index of the start of the n-gram in source tokens. No matching if found if src tokens contain multiple or zero matching n-grams. Return -1.
Attributes
max_ngram_size: The maximum NGram size for suffix matching. Default value is
?.min_ngram_size: The minimum NGram size for suffix matching. Default value is
?.
Inputs
Between 3 and 4 inputs.
src_tokens (heterogeneous) - T: Encoder input ids.
cur_tokens (heterogeneous) - T: Decoder input ids.
prev_suffix_match_idx (heterogeneous) - T: Previous suffix match index
pred_tokens (optional, heterogeneous) - T: Predicted token ids from aggressive decoding
Outputs
tokens (heterogeneous) - T: Decoder input ids after merging predicted tokens
suffix_match_idx (heterogeneous) - T: new suffix match index
OnnxComMicrosoftBifurcationDetector_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBifurcationDetector_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Component for aggressive decoding. Find the bifurcation index of predicted tokens, between source tokens, starting from previous suffix match index, and predicted tokens. Concat predicted tokens, starting from bifurcation index, to the back of current tokens. This forms the output tokens. Detect suffix match index in source tokens, between source tokens and output tokens. Detection is based on finding the appearances of last n-gram in output tokens in source tokens. A match is considered found if source tokens contain a single matching n-gram. Return the index of the start of the n-gram in source tokens. No matching if found if src tokens contain multiple or zero matching n-grams. Return -1.
Attributes
max_ngram_size: The maximum NGram size for suffix matching. Default value is
?.min_ngram_size: The minimum NGram size for suffix matching. Default value is
?.
Inputs
Between 3 and 4 inputs.
src_tokens (heterogeneous) - T: Encoder input ids.
cur_tokens (heterogeneous) - T: Decoder input ids.
prev_suffix_match_idx (heterogeneous) - T: Previous suffix match index
pred_tokens (optional, heterogeneous) - T: Predicted token ids from aggressive decoding
Outputs
tokens (heterogeneous) - T: Decoder input ids after merging predicted tokens
suffix_match_idx (heterogeneous) - T: new suffix match index
OnnxComMicrosoftBitmaskBiasDropout#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBitmaskBiasDropout(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
output, dropout_bitmask = Dropout(data + bias, ratio) + residual, Intended to specialize the dropout pattern commonly found in transformer models.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one. Default value is
?.
Inputs
Between 2 and 5 inputs.
data (heterogeneous) - T: The input data as Tensor.
bias (heterogeneous) - T: The bias input, a vector with the same shape as last dim of data OR same shape with data
residual (optional, heterogeneous) - T: The residual input, must have the same shape as data
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T3: The output mask of dropout.
OnnxComMicrosoftBitmaskBiasDropout_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBitmaskBiasDropout_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
output, dropout_bitmask = Dropout(data + bias, ratio) + residual, Intended to specialize the dropout pattern commonly found in transformer models.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one. Default value is
?.
Inputs
Between 2 and 5 inputs.
data (heterogeneous) - T: The input data as Tensor.
bias (heterogeneous) - T: The bias input, a vector with the same shape as last dim of data OR same shape with data
residual (optional, heterogeneous) - T: The residual input, must have the same shape as data
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T3: The output mask of dropout.
OnnxComMicrosoftBitmaskDropout#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBitmaskDropout(*args, **kwargs)#
Version
name: BitmaskDropout (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
BitmaskDropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs: output (floating-point tensor) and mask (optional Tensor<uint32>). If training_mode is true then the output Y will be a random dropout. Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, the user can simply not pass training_mode input or set it to false.
output = scale * data * mask,
where
scale = 1. / (1. - ratio).
This op functions in much the same was as Dropout-11 and Dropout-13 do, execpt that the mask is output as a bit-packed uint32 tensor, instead of a boolean tensor.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one. Default value is
?.
Inputs
Between 1 and 3 inputs.
data (heterogeneous) - T: The input data as Tensor.
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T3: The bit-packed output mask.
OnnxComMicrosoftBitmaskDropoutGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBitmaskDropoutGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
BitmaskDropoutGrad
Inputs
Between 2 and 4 inputs.
dy (heterogeneous) - T: The gradient tensor from output.
mask (heterogeneous) - T3: The mask output of the dropout.
ratio (optional, heterogeneous) - T1: Same value as the ratio input supplied to the dropout op with value in [0, 1). If this input is not specified, a default value of 0.5 is used.
training_mode (optional, heterogeneous) - T2: Same value as the training_mode input supplied to the dropout op. If this input is not specified, a default value of false is used.
Outputs
dx (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftBitmaskDropoutGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBitmaskDropoutGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
BitmaskDropoutGrad
Inputs
Between 2 and 4 inputs.
dy (heterogeneous) - T: The gradient tensor from output.
mask (heterogeneous) - T3: The mask output of the dropout.
ratio (optional, heterogeneous) - T1: Same value as the ratio input supplied to the dropout op with value in [0, 1). If this input is not specified, a default value of 0.5 is used.
training_mode (optional, heterogeneous) - T2: Same value as the training_mode input supplied to the dropout op. If this input is not specified, a default value of false is used.
Outputs
dx (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftBitmaskDropout_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBitmaskDropout_1(*args, **kwargs)#
Version
name: BitmaskDropout (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
BitmaskDropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs: output (floating-point tensor) and mask (optional Tensor<uint32>). If training_mode is true then the output Y will be a random dropout. Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, the user can simply not pass training_mode input or set it to false.
output = scale * data * mask,
where
scale = 1. / (1. - ratio).
This op functions in much the same was as Dropout-11 and Dropout-13 do, execpt that the mask is output as a bit-packed uint32 tensor, instead of a boolean tensor.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one. Default value is
?.
Inputs
Between 1 and 3 inputs.
data (heterogeneous) - T: The input data as Tensor.
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T3: The bit-packed output mask.
OnnxComMicrosoftBroadcastGradientArgs#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBroadcastGradientArgs(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Returns the reduction axes for computing gradients of s0 op s1 with broadcast.The ouput axes are deterministic from last to first. Output is an empty vector when no reduction is necessary for the corresponding input.
Inputs
a_shape (heterogeneous) - T: The 1st input shape as Tensor.
b_shape (heterogeneous) - T: The 2nd input shape as Tensor.
Outputs
Between 0 and 2 outputs.
a_axes (optional, heterogeneous) - T: The reduction axes for 1st input, last to first.
b_axes (optional, heterogeneous) - T: The reduction axes for 2nd input, last to first.
OnnxComMicrosoftBroadcastGradientArgs_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftBroadcastGradientArgs_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Returns the reduction axes for computing gradients of s0 op s1 with broadcast.The ouput axes are deterministic from last to first. Output is an empty vector when no reduction is necessary for the corresponding input.
Inputs
a_shape (heterogeneous) - T: The 1st input shape as Tensor.
b_shape (heterogeneous) - T: The 2nd input shape as Tensor.
Outputs
Between 0 and 2 outputs.
a_axes (optional, heterogeneous) - T: The reduction axes for 1st input, last to first.
b_axes (optional, heterogeneous) - T: The reduction axes for 2nd input, last to first.
OnnxComMicrosoftCDist#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftCDist(*args, **kwargs)#
Version
name: CDist (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
metric: The distance metric to use. If a string, the distance function can be “braycurtis”, “canberra”, “chebyshev”, “cityblock”, “correlation”, “cosine”, “dice”, “euclidean”, “hamming”, “jaccard”, “jensenshannon”, “kulsinski”, “mahalanobis”, “matching”, “minkowski”, “rogerstanimoto”, “russellrao”, “seuclidean”, “sokalmichener”, “sokalsneath”, “sqeuclidean”, “wminkowski”, “yule”. Default value is
?.
Inputs
A (heterogeneous) - T: 2D matrix with shape (M,N)
B (heterogeneous) - T: 2D matrix with shape (K,N)
Outputs
C (heterogeneous) - T: A 2D Matrix that represents the distance between each pair of the two collections of inputs.
OnnxComMicrosoftCDist_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftCDist_1(*args, **kwargs)#
Version
name: CDist (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
metric: The distance metric to use. If a string, the distance function can be “braycurtis”, “canberra”, “chebyshev”, “cityblock”, “correlation”, “cosine”, “dice”, “euclidean”, “hamming”, “jaccard”, “jensenshannon”, “kulsinski”, “mahalanobis”, “matching”, “minkowski”, “rogerstanimoto”, “russellrao”, “seuclidean”, “sokalmichener”, “sokalsneath”, “sqeuclidean”, “wminkowski”, “yule”. Default value is
?.
Inputs
A (heterogeneous) - T: 2D matrix with shape (M,N)
B (heterogeneous) - T: 2D matrix with shape (K,N)
Outputs
C (heterogeneous) - T: A 2D Matrix that represents the distance between each pair of the two collections of inputs.
OnnxComMicrosoftComplexMul#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftComplexMul(*args, **kwargs)#
Version
name: ComplexMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
A (heterogeneous) - T: input_0
B (heterogeneous) - T: input_1
Outputs
C (heterogeneous) - T: output tensor
OnnxComMicrosoftComplexMulConj#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftComplexMulConj(*args, **kwargs)#
Version
name: ComplexMulConj (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
A (heterogeneous) - T: input_0
B (heterogeneous) - T: input_1
Outputs
C (heterogeneous) - T: output tensor
OnnxComMicrosoftComplexMulConj_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftComplexMulConj_1(*args, **kwargs)#
Version
name: ComplexMulConj (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
A (heterogeneous) - T: input_0
B (heterogeneous) - T: input_1
Outputs
C (heterogeneous) - T: output tensor
OnnxComMicrosoftComplexMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftComplexMul_1(*args, **kwargs)#
Version
name: ComplexMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
A (heterogeneous) - T: input_0
B (heterogeneous) - T: input_1
Outputs
C (heterogeneous) - T: output tensor
OnnxComMicrosoftConcatTraining#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftConcatTraining(*args, **kwargs)#
Version
name: ConcatTraining (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Concatenate a list of tensors into a single tensor
Attributes
axis (required): Which axis to concat on Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
Between 1 and 2 outputs.
concat_result (heterogeneous) - T: Concatenated tensor
per_input_length (optional, heterogeneous) - Tint: Vector of length of each concatenated input along the ‘axis’ dimension
OnnxComMicrosoftConcatTraining_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftConcatTraining_1(*args, **kwargs)#
Version
name: ConcatTraining (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Concatenate a list of tensors into a single tensor
Attributes
axis (required): Which axis to concat on Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
Between 1 and 2 outputs.
concat_result (heterogeneous) - T: Concatenated tensor
per_input_length (optional, heterogeneous) - Tint: Vector of length of each concatenated input along the ‘axis’ dimension
OnnxComMicrosoftConvGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftConvGrad(*args, **kwargs)#
Version
name: ConvGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
W (heterogeneous) - T: Weight tensor
Outputs
Between 0 and 3 outputs.
dX (optional, heterogeneous) - T: Gradient of X
dW (optional, heterogeneous) - T: Gradient of W
dB (optional, heterogeneous) - T: Gradient of B
OnnxComMicrosoftConvGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftConvGrad_1(*args, **kwargs)#
Version
name: ConvGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
W (heterogeneous) - T: Weight tensor
Outputs
Between 0 and 3 outputs.
dX (optional, heterogeneous) - T: Gradient of X
dW (optional, heterogeneous) - T: Gradient of W
dB (optional, heterogeneous) - T: Gradient of B
OnnxComMicrosoftConvTransposeWithDynamicPads#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftConvTransposeWithDynamicPads(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.output_padding:
Default value is
?.strides:
Default value is
?.Inputs
Between 2 and 4 inputs.
X (heterogeneous) - T:
W (heterogeneous) - T:
Pads (optional, heterogeneous) - tensor(int64):
B (optional, heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftConvTransposeWithDynamicPads_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftConvTransposeWithDynamicPads_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.output_padding:
Default value is
?.strides:
Default value is
?.Inputs
Between 2 and 4 inputs.
X (heterogeneous) - T:
W (heterogeneous) - T:
Pads (optional, heterogeneous) - tensor(int64):
B (optional, heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftCropAndResize#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftCropAndResize(*args, **kwargs)#
Version
name: CropAndResize (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Extracts crops from the input image tensor and resizes them using bilinear sampling or nearest neighbor sampling (possibly with aspect ratio change) to a common output size specified by crop_height and crop_width. Returns a tensor with crops from the input image at positions defined at the bounding box locations in boxes. The cropped boxes are all resized (with bilinear or nearest neighbor interpolation) to a fixed size = [crop_height, crop_width]. The result is a 4-D tensor [num_boxes, crop_height, crop_width, depth]. The resizing is corner aligned.
Attributes
extrapolation_value: Value used for extrapolation, when applicable. Default is 0.0f. Default value is
?.mode: The pooling method. Two modes are supported: ‘bilinear’ and ‘nearest’. Default is ‘bilinear’. Default value is
?.
Inputs
X (heterogeneous) - T1: Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T1: RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[y1, x1, y2, x2], …]. The RoIs’ coordinates are normalized in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the ‘batch_indices’ input.
batch_indices (heterogeneous) - T2: 1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
crop_size (heterogeneous) - T2: 1-D tensor of 2 elements: [crop_height, crop_width]. All cropped image patches are resized to this size. Both crop_height and crop_width need to be positive.
Outputs
Y (heterogeneous) - T1: RoI pooled output, 4-D tensor of shape (num_rois, C, crop_height, crop_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
OnnxComMicrosoftCropAndResize_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftCropAndResize_1(*args, **kwargs)#
Version
name: CropAndResize (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Extracts crops from the input image tensor and resizes them using bilinear sampling or nearest neighbor sampling (possibly with aspect ratio change) to a common output size specified by crop_height and crop_width. Returns a tensor with crops from the input image at positions defined at the bounding box locations in boxes. The cropped boxes are all resized (with bilinear or nearest neighbor interpolation) to a fixed size = [crop_height, crop_width]. The result is a 4-D tensor [num_boxes, crop_height, crop_width, depth]. The resizing is corner aligned.
Attributes
extrapolation_value: Value used for extrapolation, when applicable. Default is 0.0f. Default value is
?.mode: The pooling method. Two modes are supported: ‘bilinear’ and ‘nearest’. Default is ‘bilinear’. Default value is
?.
Inputs
X (heterogeneous) - T1: Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T1: RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[y1, x1, y2, x2], …]. The RoIs’ coordinates are normalized in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the ‘batch_indices’ input.
batch_indices (heterogeneous) - T2: 1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
crop_size (heterogeneous) - T2: 1-D tensor of 2 elements: [crop_height, crop_width]. All cropped image patches are resized to this size. Both crop_height and crop_width need to be positive.
Outputs
Y (heterogeneous) - T1: RoI pooled output, 4-D tensor of shape (num_rois, C, crop_height, crop_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
OnnxComMicrosoftDecoderAttention#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDecoderAttention(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
This DecoderAttention supports self attention and cross attention, key and value cache, and key_padding_mask. The attention mask is not support at the moment. Some boolean parameters are passed by runtime input for generic purpose
Attributes
num_heads (required): Number of attention heads Default value is
?.
Inputs
query (heterogeneous) - T: 3D input tensor with shape (sequence_length, batch_size, hidden_size), hidden_size = num_heads * head_size
key (heterogeneous) - T: 3D input tensor with shape (total_sequence_length, batch_size, hidden_size)
q_weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, hidden_size)
kv_weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, 2 * hidden_size)
bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
key_padding_mask (optional, heterogeneous) - B: 2D input tensor with shape (batch_size, total_sequence_length)
key_cache (optional, heterogeneous) - T: input tensor with shape (batch_size, num_heads, sequence_length or total_sequence_length, head_size)
value_cache (optional, heterogeneous) - T: input tensor with shape (batch_size, num_heads, sequence_length or total_sequence_length, head_size)
static_kv (heterogeneous) - B: If static_kv = true, cross-attention; else self-attention
use_past (heterogeneous) - B: If use_past = true, use cache; else no cache
has_layer_state (heterogeneous) - B: If has_layer_state = true, layer_state = {} or [a,b]; else layer_state = None
has_key_padding_mask (heterogeneous) - B: has_key_padding_mask or not
Outputs
Between 1 and 3 outputs.
output (heterogeneous) - T: 3D output tensor with shape (sequence_length, batch_size, hidden_size)
new_key_cache (optional, heterogeneous) - T: output tensor with shape (batch_size, num_heads, new sequence_length, head_size)
new_value_cache (optional, heterogeneous) - T: output tensor with shape (batch_size, num_heads, new sequence_length, head_size)
OnnxComMicrosoftDecoderAttention_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDecoderAttention_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
This DecoderAttention supports self attention and cross attention, key and value cache, and key_padding_mask. The attention mask is not support at the moment. Some boolean parameters are passed by runtime input for generic purpose
Attributes
num_heads (required): Number of attention heads Default value is
?.
Inputs
query (heterogeneous) - T: 3D input tensor with shape (sequence_length, batch_size, hidden_size), hidden_size = num_heads * head_size
key (heterogeneous) - T: 3D input tensor with shape (total_sequence_length, batch_size, hidden_size)
q_weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, hidden_size)
kv_weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, 2 * hidden_size)
bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
key_padding_mask (optional, heterogeneous) - B: 2D input tensor with shape (batch_size, total_sequence_length)
key_cache (optional, heterogeneous) - T: input tensor with shape (batch_size, num_heads, sequence_length or total_sequence_length, head_size)
value_cache (optional, heterogeneous) - T: input tensor with shape (batch_size, num_heads, sequence_length or total_sequence_length, head_size)
static_kv (heterogeneous) - B: If static_kv = true, cross-attention; else self-attention
use_past (heterogeneous) - B: If use_past = true, use cache; else no cache
has_layer_state (heterogeneous) - B: If has_layer_state = true, layer_state = {} or [a,b]; else layer_state = None
has_key_padding_mask (heterogeneous) - B: has_key_padding_mask or not
Outputs
Between 1 and 3 outputs.
output (heterogeneous) - T: 3D output tensor with shape (sequence_length, batch_size, hidden_size)
new_key_cache (optional, heterogeneous) - T: output tensor with shape (batch_size, num_heads, new sequence_length, head_size)
new_value_cache (optional, heterogeneous) - T: output tensor with shape (batch_size, num_heads, new sequence_length, head_size)
OnnxComMicrosoftDequantizeLinear#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDequantizeLinear(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The linear dequantization operator. It consumes a quantized data, a scale, a zero point and computes the full precision data. The dequantization formula is y = (x - x_zero_point) * x_scale. Scale and zero point must have same shape. They must be either scalar (per tensor) or 1-D tensor (per ‘axis’).
Attributes
axis: The axis along which same quantization parameters are applied. It’s optional.If it’s not specified, it means per-tensor quantization and input ‘x_scale’ and ‘x_zero_point’ must be scalars.If it’s specified, it means per ‘axis’ quantization and input ‘x_scale’ and ‘x_zero_point’ must be 1-D tensors. Default value is
?.
Inputs
x (heterogeneous) - T1: N-D quantized Input tensor to be de-quantized.
x_scale (heterogeneous) - T2: Scale for input ‘x’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-axis quantization.If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
x_zero_point (heterogeneous) - T1: Zero point for input ‘x’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-axis quantization.If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
Outputs
y (heterogeneous) - T2: N-D full precision output tensor. It has same shape as input ‘x’.
OnnxComMicrosoftDequantizeLinear_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDequantizeLinear_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The linear dequantization operator. It consumes a quantized data, a scale, a zero point and computes the full precision data. The dequantization formula is y = (x - x_zero_point) * x_scale. Scale and zero point must have same shape. They must be either scalar (per tensor) or 1-D tensor (per ‘axis’).
Attributes
axis: The axis along which same quantization parameters are applied. It’s optional.If it’s not specified, it means per-tensor quantization and input ‘x_scale’ and ‘x_zero_point’ must be scalars.If it’s specified, it means per ‘axis’ quantization and input ‘x_scale’ and ‘x_zero_point’ must be 1-D tensors. Default value is
?.
Inputs
x (heterogeneous) - T1: N-D quantized Input tensor to be de-quantized.
x_scale (heterogeneous) - T2: Scale for input ‘x’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-axis quantization.If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
x_zero_point (heterogeneous) - T1: Zero point for input ‘x’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-axis quantization.If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
Outputs
y (heterogeneous) - T2: N-D full precision output tensor. It has same shape as input ‘x’.
OnnxComMicrosoftDivGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDivGrad(*args, **kwargs)#
Version
name: DivGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
dY (heterogeneous) - T: Gradient of output
A (heterogeneous) - T: dividend
B (heterogeneous) - T: divisor
Outputs
Between 0 and 2 outputs.
dA (optional, heterogeneous) - T: Gradient of dividend
dB (optional, heterogeneous) - T: Gradient of divisor
OnnxComMicrosoftDivGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDivGrad_1(*args, **kwargs)#
Version
name: DivGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
dY (heterogeneous) - T: Gradient of output
A (heterogeneous) - T: dividend
B (heterogeneous) - T: divisor
Outputs
Between 0 and 2 outputs.
dA (optional, heterogeneous) - T: Gradient of dividend
dB (optional, heterogeneous) - T: Gradient of divisor
OnnxComMicrosoftDropoutGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDropoutGrad(*args, **kwargs)#
Version
name: DropoutGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
DropoutGrad
Inputs
Between 2 and 4 inputs.
dy (heterogeneous) - T: The gradient tensor from output.
mask (heterogeneous) - T2: The mask output of the dropout.
ratio (optional, heterogeneous) - T1: Same value as the ratio input supplied to the dropout op with value in [0, 1). If this input is not specified, a default value of 0.5 is used.
training_mode (optional, heterogeneous) - T2: Same value as the training_mode input supplied to the dropout op. If this input is not specified, a default value of false is used.
Outputs
dx (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftDropoutGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDropoutGrad_1(*args, **kwargs)#
Version
name: DropoutGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
DropoutGrad
Inputs
Between 2 and 4 inputs.
dy (heterogeneous) - T: The gradient tensor from output.
mask (heterogeneous) - T2: The mask output of the dropout.
ratio (optional, heterogeneous) - T1: Same value as the ratio input supplied to the dropout op with value in [0, 1). If this input is not specified, a default value of 0.5 is used.
training_mode (optional, heterogeneous) - T2: Same value as the training_mode input supplied to the dropout op. If this input is not specified, a default value of false is used.
Outputs
dx (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftDynamicQuantizeLSTM#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDynamicQuantizeLSTM(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01. Default value is
?.activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators. Default value is
?.activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified. Default value is
?.clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified. Default value is
?.direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
?.hidden_size: Number of neurons in the hidden layer Default value is
?.input_forget: Couple the input and forget gates if 1. Default value is
?.
Inputs
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T2: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, input_size, 4*hidden_size].
R (heterogeneous) - T2: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, hidden_size, 4*hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
W_scale (heterogeneous) - T: W’s scale. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
W_zero_point (heterogeneous) - T2: W’s zero point. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
R_scale (heterogeneous) - T: R’s scale. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
R_zero_point (heterogeneous) - T2: R’s zero point. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
OnnxComMicrosoftDynamicQuantizeLSTM_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDynamicQuantizeLSTM_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01. Default value is
?.activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators. Default value is
?.activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified. Default value is
?.clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified. Default value is
?.direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
?.hidden_size: Number of neurons in the hidden layer Default value is
?.input_forget: Couple the input and forget gates if 1. Default value is
?.
Inputs
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T2: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, input_size, 4*hidden_size].
R (heterogeneous) - T2: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, hidden_size, 4*hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
W_scale (heterogeneous) - T: W’s scale. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
W_zero_point (heterogeneous) - T2: W’s zero point. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
R_scale (heterogeneous) - T: R’s scale. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
R_zero_point (heterogeneous) - T2: R’s zero point. Its size is [num_directions] for per-tensor/layer quantization, or [num_directions, 4*hidden_size] for per-channel quantization on the axis input_size.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
OnnxComMicrosoftDynamicQuantizeMatMul#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDynamicQuantizeMatMul(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
Between 3 and 5 inputs.
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
b_scale (heterogeneous) - T1: Scale of quantized input ‘B’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
b_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
bias (optional, heterogeneous) - T1: 1D input tensor, whose dimension is same as B’s last dimension
Outputs
Y (heterogeneous) - T1: Matrix multiply results from A * B
OnnxComMicrosoftDynamicQuantizeMatMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftDynamicQuantizeMatMul_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
Between 3 and 5 inputs.
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
b_scale (heterogeneous) - T1: Scale of quantized input ‘B’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
b_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
bias (optional, heterogeneous) - T1: 1D input tensor, whose dimension is same as B’s last dimension
Outputs
Y (heterogeneous) - T1: Matrix multiply results from A * B
OnnxComMicrosoftEmbedLayerNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftEmbedLayerNormalization(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
EmbedLayerNormalization is the fusion of embedding layer in BERT model, with optional mask processing. The embedding layer takes input_ids (word IDs) and segment_ids (sentence IDs) to look up word_embedding, position_embedding, and segment_emedding; the embeddings are added then applied layer normalization using gamma and beta tensors. The last input mask is optional. If mask is provided, mask index (that is position of first 0 in mask, or number of words) will be calculated.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
?.
Inputs
Between 7 and 9 inputs.
input_ids (heterogeneous) - T1: 2D words IDs with shape (batch_size, sequence_length)
segment_ids (optional, heterogeneous) - T1: 2D segment IDs with shape (batch_size, sequence_length)
word_embedding (heterogeneous) - T: 2D with shape (,hidden_size)
position_embedding (heterogeneous) - T: 2D with shape (, hidden_size)
segment_embedding (optional, heterogeneous) - T: 2D with shape (, hidden_size)
gamma (heterogeneous) - T: 1D gamma tensor for layer normalization with shape (hidden_size)
beta (heterogeneous) - T: 1D beta tensor for layer normalization with shape (hidden_size)
mask (optional, heterogeneous) - T1: 2D attention mask with shape (batch_size, sequence_length)
position_ids (optional, heterogeneous) - T1: 2D position ids with shape (batch_size, sequence_length)
Outputs
Between 2 and 3 outputs.
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
mask_index (heterogeneous) - T1: 1D mask_index tensor with shape (batch_size)
embedding_sum (optional, heterogeneous) - T: sum of word_embedding and position_embedding without layer normalization
OnnxComMicrosoftEmbedLayerNormalization_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftEmbedLayerNormalization_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
EmbedLayerNormalization is the fusion of embedding layer in BERT model, with optional mask processing. The embedding layer takes input_ids (word IDs) and segment_ids (sentence IDs) to look up word_embedding, position_embedding, and segment_emedding; the embeddings are added then applied layer normalization using gamma and beta tensors. The last input mask is optional. If mask is provided, mask index (that is position of first 0 in mask, or number of words) will be calculated.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
?.
Inputs
Between 7 and 9 inputs.
input_ids (heterogeneous) - T1: 2D words IDs with shape (batch_size, sequence_length)
segment_ids (optional, heterogeneous) - T1: 2D segment IDs with shape (batch_size, sequence_length)
word_embedding (heterogeneous) - T: 2D with shape (,hidden_size)
position_embedding (heterogeneous) - T: 2D with shape (, hidden_size)
segment_embedding (optional, heterogeneous) - T: 2D with shape (, hidden_size)
gamma (heterogeneous) - T: 1D gamma tensor for layer normalization with shape (hidden_size)
beta (heterogeneous) - T: 1D beta tensor for layer normalization with shape (hidden_size)
mask (optional, heterogeneous) - T1: 2D attention mask with shape (batch_size, sequence_length)
position_ids (optional, heterogeneous) - T1: 2D position ids with shape (batch_size, sequence_length)
Outputs
Between 2 and 3 outputs.
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
mask_index (heterogeneous) - T1: 1D mask_index tensor with shape (batch_size)
embedding_sum (optional, heterogeneous) - T: sum of word_embedding and position_embedding without layer normalization
OnnxComMicrosoftExpandDims#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftExpandDims(*args, **kwargs)#
Version
name: ExpandDims (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
ExpandDims echo operator.
Inputs
X (heterogeneous) - T: input
axis (heterogeneous) - tensor(int32): Specified axis to insert a dimension
Outputs
Y (heterogeneous) - T: output
OnnxComMicrosoftExpandDims_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftExpandDims_1(*args, **kwargs)#
Version
name: ExpandDims (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
ExpandDims echo operator.
Inputs
X (heterogeneous) - T: input
axis (heterogeneous) - tensor(int32): Specified axis to insert a dimension
Outputs
Y (heterogeneous) - T: output
OnnxComMicrosoftFastGelu#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFastGelu(*args, **kwargs)#
Version
name: FastGelu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
GELU (Gaussian Error Linear Unit) approximation: Y=0.5*X*(1+tanh(0.797885*X+0.035677*X*X*X)) with an optional input of bias that will be added to X before GELU.
Inputs
Between 1 and 2 inputs.
X (heterogeneous) - T: input tensor
bias (optional, heterogeneous) - T: bias tensor
Outputs
Y (heterogeneous) - T: output tensor
OnnxComMicrosoftFastGeluGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFastGeluGrad(*args, **kwargs)#
Version
name: FastGeluGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
FastGeluGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftFastGeluGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFastGeluGrad_1(*args, **kwargs)#
Version
name: FastGeluGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
FastGeluGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftFastGelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFastGelu_1(*args, **kwargs)#
Version
name: FastGelu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
GELU (Gaussian Error Linear Unit) approximation: Y=0.5*X*(1+tanh(0.797885*X+0.035677*X*X*X)) with an optional input of bias that will be added to X before GELU.
Inputs
Between 1 and 2 inputs.
X (heterogeneous) - T: input tensor
bias (optional, heterogeneous) - T: bias tensor
Outputs
Y (heterogeneous) - T: output tensor
OnnxComMicrosoftFusedConv#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFusedConv(*args, **kwargs)#
Version
name: FusedConv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The fused convolution operator schema is the same as Conv besides it includes an attribute activation.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
Between 2 and 4 inputs.
X (heterogeneous) - T:
W (heterogeneous) - T:
B (optional, heterogeneous) - T:
Z (optional, heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftFusedConv_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFusedConv_1(*args, **kwargs)#
Version
name: FusedConv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The fused convolution operator schema is the same as Conv besides it includes an attribute activation.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
Between 2 and 4 inputs.
X (heterogeneous) - T:
W (heterogeneous) - T:
B (optional, heterogeneous) - T:
Z (optional, heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftFusedGemm#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFusedGemm(*args, **kwargs)#
Version
name: FusedGemm (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The FusedGemm operator schema is the same as Gemm besides it includes attributes activation and leaky_relu_alpha.
Attributes
activation:
Default value is
?.activation_alpha:
Default value is
?.activation_beta:
Default value is
?.activation_gamma:
Default value is
?.alpha: Scalar multiplier for the product of input tensors A * B. Default value is
?.beta: Scalar multiplier for input tensor C. Default value is
?.transA: Whether A should be transposed Default value is
?.transB: Whether B should be transposed Default value is
?.
Inputs
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (heterogeneous) - T: Input tensor C. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
OnnxComMicrosoftFusedGemm_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFusedGemm_1(*args, **kwargs)#
Version
name: FusedGemm (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The FusedGemm operator schema is the same as Gemm besides it includes attributes activation and leaky_relu_alpha.
Attributes
activation:
Default value is
?.activation_alpha:
Default value is
?.activation_beta:
Default value is
?.activation_gamma:
Default value is
?.alpha: Scalar multiplier for the product of input tensors A * B. Default value is
?.beta: Scalar multiplier for input tensor C. Default value is
?.transA: Whether A should be transposed Default value is
?.transB: Whether B should be transposed Default value is
?.
Inputs
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (heterogeneous) - T: Input tensor C. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
OnnxComMicrosoftFusedMatMul#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFusedMatMul(*args, **kwargs)#
Version
name: FusedMatMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Attributes
alpha: Scalar multiplier for the product of the input tensors. Default value is
?.transA: Whether A should be transposed on the last two dimensions before doing multiplication Default value is
?.transB: Whether B should be transposed on the last two dimensions before doing multiplication Default value is
?.transBatchA: Whether A should be transposed on the 1st dimension and batch dimensions (dim-1 to dim-rank-2) before doing multiplication Default value is
?.transBatchB: Whether B should be transposed on the 1st dimension and batch dimensions (dim-1 to dim-rank-2) before doing multiplication Default value is
?.
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results
OnnxComMicrosoftFusedMatMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftFusedMatMul_1(*args, **kwargs)#
Version
name: FusedMatMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Attributes
alpha: Scalar multiplier for the product of the input tensors. Default value is
?.transA: Whether A should be transposed on the last two dimensions before doing multiplication Default value is
?.transB: Whether B should be transposed on the last two dimensions before doing multiplication Default value is
?.transBatchA: Whether A should be transposed on the 1st dimension and batch dimensions (dim-1 to dim-rank-2) before doing multiplication Default value is
?.transBatchB: Whether B should be transposed on the 1st dimension and batch dimensions (dim-1 to dim-rank-2) before doing multiplication Default value is
?.
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results
OnnxComMicrosoftGatherElementsGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherElementsGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
GatherElementsGrad
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
?.
Inputs
dY (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
shape (heterogeneous) - I: Shape of the GatherElements input data.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
dX (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
OnnxComMicrosoftGatherElementsGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherElementsGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
GatherElementsGrad
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
?.
Inputs
dY (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
shape (heterogeneous) - I: Shape of the GatherElements input data.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
dX (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
OnnxComMicrosoftGatherGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherGrad(*args, **kwargs)#
Version
name: GatherGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range in [-r, r-1] Default value is
?.
Inputs
shape (heterogeneous) - I: Shape of the Gather input X.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of any rank q.
dY (heterogeneous) - T: Gradient of output
Outputs
dX (heterogeneous) - T: Gradient of input
OnnxComMicrosoftGatherGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherGrad_1(*args, **kwargs)#
Version
name: GatherGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range in [-r, r-1] Default value is
?.
Inputs
shape (heterogeneous) - I: Shape of the Gather input X.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of any rank q.
dY (heterogeneous) - T: Gradient of output
Outputs
dX (heterogeneous) - T: Gradient of input
OnnxComMicrosoftGatherND#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherND(*args, **kwargs)#
Version
name: GatherND (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q >= 1, gather slices of data into an output tensor of rank q - 1 + r - indices[-1]. Example 1:
data = [[0,1],[2,3]] indices = [[0,0],[1,1]] output = [0,3]
- Example 2:
data = [[0,1],[2,3]] indices = [[1],[0]] output = [[2,3],[0,1]]
- Example 3:
data = [[[0,1],[2,3]],[[4,5],[6,7]]] indices = [[0,1],[1,0]] output = [[2,3],[4,5]]
- Example 4:
data = [[[0,1],[2,3]],[[4,5],[6,7]]] indices = [[[0,1]],[[1,0]]] output = [[[2,3]],[[4,5]]]
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of rank q >= 1.
Outputs
output (heterogeneous) - T: Tensor of rank q-1+r-indices[-1].
OnnxComMicrosoftGatherNDGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherNDGrad(*args, **kwargs)#
Version
name: GatherNDGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
batch_dims: The number of batch dims. The gather of indexing starts from dimension of data[batch_dims+1:] Default value is
?.
Inputs
shape (heterogeneous) - T1: The shape of source data input of GatherND.
indices (heterogeneous) - Tind: Tensor of rank q >= 1.
update (heterogeneous) - T: The gradient of the output.
Outputs
output (heterogeneous) - T: Tensor graident of the input.
OnnxComMicrosoftGatherNDGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherNDGrad_1(*args, **kwargs)#
Version
name: GatherNDGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
batch_dims: The number of batch dims. The gather of indexing starts from dimension of data[batch_dims+1:] Default value is
?.
Inputs
shape (heterogeneous) - T1: The shape of source data input of GatherND.
indices (heterogeneous) - Tind: Tensor of rank q >= 1.
update (heterogeneous) - T: The gradient of the output.
Outputs
output (heterogeneous) - T: Tensor graident of the input.
OnnxComMicrosoftGatherND_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGatherND_1(*args, **kwargs)#
Version
name: GatherND (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q >= 1, gather slices of data into an output tensor of rank q - 1 + r - indices[-1]. Example 1:
data = [[0,1],[2,3]] indices = [[0,0],[1,1]] output = [0,3]
- Example 2:
data = [[0,1],[2,3]] indices = [[1],[0]] output = [[2,3],[0,1]]
- Example 3:
data = [[[0,1],[2,3]],[[4,5],[6,7]]] indices = [[0,1],[1,0]] output = [[2,3],[4,5]]
- Example 4:
data = [[[0,1],[2,3]],[[4,5],[6,7]]] indices = [[[0,1]],[[1,0]]] output = [[[2,3]],[[4,5]]]
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of rank q >= 1.
Outputs
output (heterogeneous) - T: Tensor of rank q-1+r-indices[-1].
OnnxComMicrosoftGelu#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGelu(*args, **kwargs)#
Version
name: Gelu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Gaussian Error Linear Unit. A high-performing neural network activation function.The GELU nonlinearity is the expected transformation of a stochastic regularizer which randomly applies the identity or zero map to a neuron’s input. The GELU nonlinearity weights inputs by their magnitude, rather than gates inputs by their sign as in ReLUs.
Inputs
X (heterogeneous) - T: The input data as Tensor.
Outputs
Y (heterogeneous) - T: The output.
OnnxComMicrosoftGeluGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGeluGrad(*args, **kwargs)#
Version
name: GeluGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
GeluGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftGeluGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGeluGrad_1(*args, **kwargs)#
Version
name: GeluGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
GeluGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
X (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftGelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGelu_1(*args, **kwargs)#
Version
name: Gelu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Gaussian Error Linear Unit. A high-performing neural network activation function.The GELU nonlinearity is the expected transformation of a stochastic regularizer which randomly applies the identity or zero map to a neuron’s input. The GELU nonlinearity weights inputs by their magnitude, rather than gates inputs by their sign as in ReLUs.
Inputs
X (heterogeneous) - T: The input data as Tensor.
Outputs
Y (heterogeneous) - T: The output.
OnnxComMicrosoftGistBinarizeDecoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistBinarizeDecoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistBinarizeDecoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistBinarizeDecoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistBinarizeEncoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistBinarizeEncoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistBinarizeEncoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistBinarizeEncoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistPack16Decoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack16Decoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compressed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPack16Decoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack16Decoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compressed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPack16Encoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack16Encoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistPack16Encoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack16Encoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistPack1Decoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack1Decoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: 1 bit compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPack1Decoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack1Decoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: 1 bit compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPack1Encoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack1Encoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: 1 bit compressed output
OnnxComMicrosoftGistPack1Encoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack1Encoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: 1 bit compressed output
OnnxComMicrosoftGistPack8Decoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack8Decoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPack8Decoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack8Decoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPack8Encoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack8Encoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistPack8Encoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPack8Encoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistPackMsfp15Decoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPackMsfp15Decoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPackMsfp15Decoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPackMsfp15Decoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
X (heterogeneous) - T1: compresssed input
Outputs
Y (heterogeneous) - T: uncompressed output
OnnxComMicrosoftGistPackMsfp15Encoder#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPackMsfp15Encoder(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGistPackMsfp15Encoder_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGistPackMsfp15Encoder_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: uncompressed input
Outputs
Y (heterogeneous) - T1: compressed output
OnnxComMicrosoftGridSample#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGridSample(*args, **kwargs)#
Version
name: GridSample (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Given an input and a flow-field grid, computes the output using input values and pixel locations from grid. Currently, only spatial (4-D) inputs are supported. For input with shape (N, C, H, W) and grid with shape (N, H_out, W_out, 2), the output will have shape (N, C, H_out, W_out). For each output location output[n, :, h, w], the size-2 vector grid[n, h, w] specifies input pixel locations x and y, which are used to interpolate the output value output[n, :, h, w]. The GridSample operator is often used in doing grid generator and sampler in the [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/master/generated/torch.nn.functional.grid_sample.html#torch-nn-functional-grid-sample).
Attributes
align_corners: If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input’s corner pixels. If align_corners=0, they are instead considered as referring to the corner points of the input’s corner pixels, making the sampling more resolution agnostic. Default value is
?.mode: Three interpolation modes: bilinear (default), nearest and bicubic. Default value is
?.padding_mode: Support padding modes for outside grid values: zeros`(default), `border, reflection. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. Default value is
?.
Inputs
X (heterogeneous) - T1: 4-D tensor of shape (N, C, H, W), where N is the batch size, C is the numbers of channels, H and W are the height and width of the input data.
Grid (heterogeneous) - T1: Input offset, 4-D tensor of shape (N, H_out, W_out, 2), where H_out and W_out are the height and width of grid and output, Grid specifies the sampling pixel locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode.
Outputs
Y (heterogeneous) - T2: 4-D tensor of shape (N, C, H_out, W_out).
OnnxComMicrosoftGridSample_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGridSample_1(*args, **kwargs)#
Version
name: GridSample (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Given an input and a flow-field grid, computes the output using input values and pixel locations from grid. Currently, only spatial (4-D) inputs are supported. For input with shape (N, C, H, W) and grid with shape (N, H_out, W_out, 2), the output will have shape (N, C, H_out, W_out). For each output location output[n, :, h, w], the size-2 vector grid[n, h, w] specifies input pixel locations x and y, which are used to interpolate the output value output[n, :, h, w]. The GridSample operator is often used in doing grid generator and sampler in the [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/master/generated/torch.nn.functional.grid_sample.html#torch-nn-functional-grid-sample).
Attributes
align_corners: If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input’s corner pixels. If align_corners=0, they are instead considered as referring to the corner points of the input’s corner pixels, making the sampling more resolution agnostic. Default value is
?.mode: Three interpolation modes: bilinear (default), nearest and bicubic. Default value is
?.padding_mode: Support padding modes for outside grid values: zeros`(default), `border, reflection. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. Default value is
?.
Inputs
X (heterogeneous) - T1: 4-D tensor of shape (N, C, H, W), where N is the batch size, C is the numbers of channels, H and W are the height and width of the input data.
Grid (heterogeneous) - T1: Input offset, 4-D tensor of shape (N, H_out, W_out, 2), where H_out and W_out are the height and width of grid and output, Grid specifies the sampling pixel locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode.
Outputs
Y (heterogeneous) - T2: 4-D tensor of shape (N, C, H_out, W_out).
OnnxComMicrosoftGroup#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGroup(*args, **kwargs)#
Version
name: Group (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
if all the inputs are available, the output will be true
Inputs
Between 1 and 2147483647 inputs.
input_tensors (variadic) - T: list of dependency tensors
Outputs
done (heterogeneous) - B: all the dependency tensors are ready
OnnxComMicrosoftGroup_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftGroup_1(*args, **kwargs)#
Version
name: Group (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
if all the inputs are available, the output will be true
Inputs
Between 1 and 2147483647 inputs.
input_tensors (variadic) - T: list of dependency tensors
Outputs
done (heterogeneous) - B: all the dependency tensors are ready
OnnxComMicrosoftInPlaceAccumulator#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftInPlaceAccumulator(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
in-place accumulator for tensors
Inputs
Between 2 and 3 inputs.
old_sum (heterogeneous) - T: historical result of accumulator
value (heterogeneous) - T_GRAD: the value that will be added to the accumulator
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if tensor should be updated
Outputs
new_sum (heterogeneous) - T: updated result of accumulator
OnnxComMicrosoftInPlaceAccumulator_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftInPlaceAccumulator_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
in-place accumulator for tensors
Inputs
Between 2 and 3 inputs.
old_sum (heterogeneous) - T: historical result of accumulator
value (heterogeneous) - T_GRAD: the value that will be added to the accumulator
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if tensor should be updated
Outputs
new_sum (heterogeneous) - T: updated result of accumulator
OnnxComMicrosoftInverse#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftInverse(*args, **kwargs)#
Version
name: Inverse (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: Input tensor. Every matrix in the batch must be invertible.
Outputs
Y (heterogeneous) - T: Output tensor of the same type and shape as the input tensor.
OnnxComMicrosoftInverse_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftInverse_1(*args, **kwargs)#
Version
name: Inverse (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
X (heterogeneous) - T: Input tensor. Every matrix in the batch must be invertible.
Outputs
Y (heterogeneous) - T: Output tensor of the same type and shape as the input tensor.
OnnxComMicrosoftInvertibleLayerNormalizationGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftInvertibleLayerNormalizationGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
LayerNormalizationGrad
Attributes
axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is
?.
Inputs
Y_grad (heterogeneous) - V: The gradient tensor from output.
Y (heterogeneous) - V: Output data tensor from the forward path
scale (heterogeneous) - V: Scale tensor.
bias (heterogeneous) - V: Bias tensor.
inv_std_var (heterogeneous) - U: inverse std variance of X.
Outputs
X_grad (heterogeneous) - T: Gradient of the input.
scale_grad (heterogeneous) - V: Gradient of the scale.
bias_grad (heterogeneous) - V: Gradient of the bias.
OnnxComMicrosoftInvertibleLayerNormalizationGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftInvertibleLayerNormalizationGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
LayerNormalizationGrad
Attributes
axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is
?.
Inputs
Y_grad (heterogeneous) - V: The gradient tensor from output.
Y (heterogeneous) - V: Output data tensor from the forward path
scale (heterogeneous) - V: Scale tensor.
bias (heterogeneous) - V: Bias tensor.
inv_std_var (heterogeneous) - U: inverse std variance of X.
Outputs
X_grad (heterogeneous) - T: Gradient of the input.
scale_grad (heterogeneous) - V: Gradient of the scale.
bias_grad (heterogeneous) - V: Gradient of the bias.
OnnxComMicrosoftIrfft#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftIrfft(*args, **kwargs)#
Version
name: Irfft (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
normalized:
Default value is
?.onesided:
Default value is
?.signal_ndim (required):
Default value is
?.Inputs
X (heterogeneous) - T: input tensor
Outputs
Y (heterogeneous) - T: output tensor
OnnxComMicrosoftIrfft_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftIrfft_1(*args, **kwargs)#
Version
name: Irfft (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
normalized:
Default value is
?.onesided:
Default value is
?.signal_ndim (required):
Default value is
?.Inputs
X (heterogeneous) - T: input tensor
Outputs
Y (heterogeneous) - T: output tensor
OnnxComMicrosoftIsAllFinite#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftIsAllFinite(*args, **kwargs)#
Version
name: IsAllFinite (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
IsAllFinite
Attributes
isinf_only: If true, check only for Inf, -Inf. Default value is
?.isnan_only: If true, check only for NaN. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - V: Input tensors to check.
Outputs
output (heterogeneous) - T: The output scalar. Its value is true if all input tensors are finite. Otherwise, the output value would be false.
OnnxComMicrosoftIsAllFinite_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftIsAllFinite_1(*args, **kwargs)#
Version
name: IsAllFinite (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
IsAllFinite
Attributes
isinf_only: If true, check only for Inf, -Inf. Default value is
?.isnan_only: If true, check only for NaN. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - V: Input tensors to check.
Outputs
output (heterogeneous) - T: The output scalar. Its value is true if all input tensors are finite. Otherwise, the output value would be false.
OnnxComMicrosoftIsFinite#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftIsFinite(*args, **kwargs)#
Version
name: IsFinite (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
IsFinite
Inputs
X (heterogeneous) - T: The input tensor.
Outputs
Y (heterogeneous) - T1: The output tensor. Its shape is the same as the input.
OnnxComMicrosoftIsFinite_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftIsFinite_1(*args, **kwargs)#
Version
name: IsFinite (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
IsFinite
Inputs
X (heterogeneous) - T: The input tensor.
Outputs
Y (heterogeneous) - T1: The output tensor. Its shape is the same as the input.
OnnxComMicrosoftLambOptimizer#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLambOptimizer(*args, **kwargs)#
Version
name: LambOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
alpha: Coefficient of previous gradient in running average. Default value is
?.beta: Coefficient of previous squared gradient in running average.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn’t reduce the learning rate. Default value is
?.do_bias_correction: Compute unbiased 1st and 2nd momentums. Default value is
?.epsilon: Small scalar to avoid dividing by zero. Default value is
?.lambda: Regularization coefficient of 0.5 * lambda * ||X||_2^2. Default to 0, which means no regularization. Default value is
?.max_norm_clip: clip threshold of gradients. Default value is
?.ratio_max: Upper bound on confidence ratio. Default value is
?.ratio_min: Lower bound on confidence ratio. Default value is
?.
Inputs
Between 0 and 5125 inputs.
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if weight tensors should be updated.
loss_scale (optional, heterogeneous) - T2: Loss scale for mixed precision training.
gradient_norm (optional, heterogeneous) - T_GRAD_NORM: Norm of global gradient.
R (optional, heterogeneous) - T1: The initial learning rate.
step (optional, heterogeneous) - TInt64: One-based index of the current training iteration.
__group_0__weights (optional) - T2: weights to optimize.
__group_0__gradients (optional) - T3: gradients computed in this iteration.
__group_0__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_0__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_0__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1__weights (optional) - T2: weights to optimize.
__group_1__gradients (optional) - T3: gradients computed in this iteration.
__group_1__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_2__weights (optional) - T2: weights to optimize.
__group_2__gradients (optional) - T3: gradients computed in this iteration.
__group_2__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_2__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_2__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_3__weights (optional) - T2: weights to optimize.
__group_3__gradients (optional) - T3: gradients computed in this iteration.
__group_3__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_3__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_3__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_4__weights (optional) - T2: weights to optimize.
__group_4__gradients (optional) - T3: gradients computed in this iteration.
__group_4__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_4__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_4__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_5__weights (optional) - T2: weights to optimize.
__group_5__gradients (optional) - T3: gradients computed in this iteration.
__group_5__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_5__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_5__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_6__weights (optional) - T2: weights to optimize.
__group_6__gradients (optional) - T3: gradients computed in this iteration.
__group_6__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_6__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_6__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_7__weights (optional) - T2: weights to optimize.
__group_7__gradients (optional) - T3: gradients computed in this iteration.
__group_7__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_7__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_7__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_8__weights (optional) - T2: weights to optimize.
__group_8__gradients (optional) - T3: gradients computed in this iteration.
__group_8__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_8__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_8__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_9__weights (optional) - T2: weights to optimize.
__group_9__gradients (optional) - T3: gradients computed in this iteration.
__group_9__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_9__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_9__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_10__weights (optional) - T2: weights to optimize.
__group_10__gradients (optional) - T3: gradients computed in this iteration.
__group_10__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_10__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_10__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_11__weights (optional) - T2: weights to optimize.
__group_11__gradients (optional) - T3: gradients computed in this iteration.
__group_11__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_11__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_11__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_12__weights (optional) - T2: weights to optimize.
__group_12__gradients (optional) - T3: gradients computed in this iteration.
__group_12__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_12__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_12__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_13__weights (optional) - T2: weights to optimize.
__group_13__gradients (optional) - T3: gradients computed in this iteration.
__group_13__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_13__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_13__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_14__weights (optional) - T2: weights to optimize.
__group_14__gradients (optional) - T3: gradients computed in this iteration.
__group_14__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_14__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_14__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_15__weights (optional) - T2: weights to optimize.
__group_15__gradients (optional) - T3: gradients computed in this iteration.
__group_15__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_15__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_15__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_16__weights (optional) - T2: weights to optimize.
__group_16__gradients (optional) - T3: gradients computed in this iteration.
__group_16__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_16__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_16__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_17__weights (optional) - T2: weights to optimize.
__group_17__gradients (optional) - T3: gradients computed in this iteration.
__group_17__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_17__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_17__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_18__weights (optional) - T2: weights to optimize.
__group_18__gradients (optional) - T3: gradients computed in this iteration.
__group_18__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_18__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_18__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_19__weights (optional) - T2: weights to optimize.
__group_19__gradients (optional) - T3: gradients computed in this iteration.
__group_19__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_19__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_19__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_20__weights (optional) - T2: weights to optimize.
__group_20__gradients (optional) - T3: gradients computed in this iteration.
__group_20__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_20__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_20__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_21__weights (optional) - T2: weights to optimize.
__group_21__gradients (optional) - T3: gradients computed in this iteration.
__group_21__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_21__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_21__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_22__weights (optional) - T2: weights to optimize.
__group_22__gradients (optional) - T3: gradients computed in this iteration.
__group_22__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_22__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_22__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_23__weights (optional) - T2: weights to optimize.
__group_23__gradients (optional) - T3: gradients computed in this iteration.
__group_23__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_23__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_23__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_24__weights (optional) - T2: weights to optimize.
__group_24__gradients (optional) - T3: gradients computed in this iteration.
__group_24__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_24__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_24__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_25__weights (optional) - T2: weights to optimize.
__group_25__gradients (optional) - T3: gradients computed in this iteration.
__group_25__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_25__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_25__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_26__weights (optional) - T2: weights to optimize.
__group_26__gradients (optional) - T3: gradients computed in this iteration.
__group_26__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_26__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_26__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_27__weights (optional) - T2: weights to optimize.
__group_27__gradients (optional) - T3: gradients computed in this iteration.
__group_27__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_27__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_27__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_28__weights (optional) - T2: weights to optimize.
__group_28__gradients (optional) - T3: gradients computed in this iteration.
__group_28__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_28__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_28__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_29__weights (optional) - T2: weights to optimize.
__group_29__gradients (optional) - T3: gradients computed in this iteration.
__group_29__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_29__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_29__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_30__weights (optional) - T2: weights to optimize.
__group_30__gradients (optional) - T3: gradients computed in this iteration.
__group_30__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_30__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_30__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_31__weights (optional) - T2: weights to optimize.
__group_31__gradients (optional) - T3: gradients computed in this iteration.
__group_31__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_31__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_31__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_32__weights (optional) - T2: weights to optimize.
__group_32__gradients (optional) - T3: gradients computed in this iteration.
__group_32__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_32__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_32__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_33__weights (optional) - T2: weights to optimize.
__group_33__gradients (optional) - T3: gradients computed in this iteration.
__group_33__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_33__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_33__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_34__weights (optional) - T2: weights to optimize.
__group_34__gradients (optional) - T3: gradients computed in this iteration.
__group_34__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_34__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_34__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_35__weights (optional) - T2: weights to optimize.
__group_35__gradients (optional) - T3: gradients computed in this iteration.
__group_35__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_35__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_35__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_36__weights (optional) - T2: weights to optimize.
__group_36__gradients (optional) - T3: gradients computed in this iteration.
__group_36__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_36__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_36__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_37__weights (optional) - T2: weights to optimize.
__group_37__gradients (optional) - T3: gradients computed in this iteration.
__group_37__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_37__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_37__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_38__weights (optional) - T2: weights to optimize.
__group_38__gradients (optional) - T3: gradients computed in this iteration.
__group_38__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_38__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_38__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_39__weights (optional) - T2: weights to optimize.
__group_39__gradients (optional) - T3: gradients computed in this iteration.
__group_39__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_39__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_39__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_40__weights (optional) - T2: weights to optimize.
__group_40__gradients (optional) - T3: gradients computed in this iteration.
__group_40__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_40__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_40__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_41__weights (optional) - T2: weights to optimize.
__group_41__gradients (optional) - T3: gradients computed in this iteration.
__group_41__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_41__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_41__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_42__weights (optional) - T2: weights to optimize.
__group_42__gradients (optional) - T3: gradients computed in this iteration.
__group_42__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_42__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_42__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_43__weights (optional) - T2: weights to optimize.
__group_43__gradients (optional) - T3: gradients computed in this iteration.
__group_43__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_43__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_43__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_44__weights (optional) - T2: weights to optimize.
__group_44__gradients (optional) - T3: gradients computed in this iteration.
__group_44__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_44__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_44__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_45__weights (optional) - T2: weights to optimize.
__group_45__gradients (optional) - T3: gradients computed in this iteration.
__group_45__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_45__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_45__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_46__weights (optional) - T2: weights to optimize.
__group_46__gradients (optional) - T3: gradients computed in this iteration.
__group_46__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_46__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_46__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_47__weights (optional) - T2: weights to optimize.
__group_47__gradients (optional) - T3: gradients computed in this iteration.
__group_47__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_47__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_47__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_48__weights (optional) - T2: weights to optimize.
__group_48__gradients (optional) - T3: gradients computed in this iteration.
__group_48__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_48__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_48__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_49__weights (optional) - T2: weights to optimize.
__group_49__gradients (optional) - T3: gradients computed in this iteration.
__group_49__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_49__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_49__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_50__weights (optional) - T2: weights to optimize.
__group_50__gradients (optional) - T3: gradients computed in this iteration.
__group_50__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_50__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_50__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_51__weights (optional) - T2: weights to optimize.
__group_51__gradients (optional) - T3: gradients computed in this iteration.
__group_51__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_51__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_51__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_52__weights (optional) - T2: weights to optimize.
__group_52__gradients (optional) - T3: gradients computed in this iteration.
__group_52__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_52__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_52__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_53__weights (optional) - T2: weights to optimize.
__group_53__gradients (optional) - T3: gradients computed in this iteration.
__group_53__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_53__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_53__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_54__weights (optional) - T2: weights to optimize.
__group_54__gradients (optional) - T3: gradients computed in this iteration.
__group_54__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_54__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_54__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_55__weights (optional) - T2: weights to optimize.
__group_55__gradients (optional) - T3: gradients computed in this iteration.
__group_55__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_55__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_55__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_56__weights (optional) - T2: weights to optimize.
__group_56__gradients (optional) - T3: gradients computed in this iteration.
__group_56__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_56__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_56__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_57__weights (optional) - T2: weights to optimize.
__group_57__gradients (optional) - T3: gradients computed in this iteration.
__group_57__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_57__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_57__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_58__weights (optional) - T2: weights to optimize.
__group_58__gradients (optional) - T3: gradients computed in this iteration.
__group_58__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_58__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_58__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_59__weights (optional) - T2: weights to optimize.
__group_59__gradients (optional) - T3: gradients computed in this iteration.
__group_59__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_59__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_59__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_60__weights (optional) - T2: weights to optimize.
__group_60__gradients (optional) - T3: gradients computed in this iteration.
__group_60__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_60__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_60__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_61__weights (optional) - T2: weights to optimize.
__group_61__gradients (optional) - T3: gradients computed in this iteration.
__group_61__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_61__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_61__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_62__weights (optional) - T2: weights to optimize.
__group_62__gradients (optional) - T3: gradients computed in this iteration.
__group_62__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_62__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_62__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_63__weights (optional) - T2: weights to optimize.
__group_63__gradients (optional) - T3: gradients computed in this iteration.
__group_63__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_63__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_63__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_64__weights (optional) - T2: weights to optimize.
__group_64__gradients (optional) - T3: gradients computed in this iteration.
__group_64__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_64__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_64__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_65__weights (optional) - T2: weights to optimize.
__group_65__gradients (optional) - T3: gradients computed in this iteration.
__group_65__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_65__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_65__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_66__weights (optional) - T2: weights to optimize.
__group_66__gradients (optional) - T3: gradients computed in this iteration.
__group_66__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_66__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_66__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_67__weights (optional) - T2: weights to optimize.
__group_67__gradients (optional) - T3: gradients computed in this iteration.
__group_67__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_67__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_67__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_68__weights (optional) - T2: weights to optimize.
__group_68__gradients (optional) - T3: gradients computed in this iteration.
__group_68__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_68__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_68__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_69__weights (optional) - T2: weights to optimize.
__group_69__gradients (optional) - T3: gradients computed in this iteration.
__group_69__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_69__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_69__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_70__weights (optional) - T2: weights to optimize.
__group_70__gradients (optional) - T3: gradients computed in this iteration.
__group_70__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_70__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_70__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_71__weights (optional) - T2: weights to optimize.
__group_71__gradients (optional) - T3: gradients computed in this iteration.
__group_71__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_71__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_71__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_72__weights (optional) - T2: weights to optimize.
__group_72__gradients (optional) - T3: gradients computed in this iteration.
__group_72__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_72__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_72__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_73__weights (optional) - T2: weights to optimize.
__group_73__gradients (optional) - T3: gradients computed in this iteration.
__group_73__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_73__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_73__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_74__weights (optional) - T2: weights to optimize.
__group_74__gradients (optional) - T3: gradients computed in this iteration.
__group_74__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_74__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_74__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_75__weights (optional) - T2: weights to optimize.
__group_75__gradients (optional) - T3: gradients computed in this iteration.
__group_75__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_75__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_75__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_76__weights (optional) - T2: weights to optimize.
__group_76__gradients (optional) - T3: gradients computed in this iteration.
__group_76__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_76__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_76__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_77__weights (optional) - T2: weights to optimize.
__group_77__gradients (optional) - T3: gradients computed in this iteration.
__group_77__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_77__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_77__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_78__weights (optional) - T2: weights to optimize.
__group_78__gradients (optional) - T3: gradients computed in this iteration.
__group_78__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_78__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_78__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_79__weights (optional) - T2: weights to optimize.
__group_79__gradients (optional) - T3: gradients computed in this iteration.
__group_79__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_79__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_79__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_80__weights (optional) - T2: weights to optimize.
__group_80__gradients (optional) - T3: gradients computed in this iteration.
__group_80__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_80__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_80__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_81__weights (optional) - T2: weights to optimize.
__group_81__gradients (optional) - T3: gradients computed in this iteration.
__group_81__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_81__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_81__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_82__weights (optional) - T2: weights to optimize.
__group_82__gradients (optional) - T3: gradients computed in this iteration.
__group_82__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_82__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_82__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_83__weights (optional) - T2: weights to optimize.
__group_83__gradients (optional) - T3: gradients computed in this iteration.
__group_83__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_83__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_83__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_84__weights (optional) - T2: weights to optimize.
__group_84__gradients (optional) - T3: gradients computed in this iteration.
__group_84__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_84__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_84__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_85__weights (optional) - T2: weights to optimize.
__group_85__gradients (optional) - T3: gradients computed in this iteration.
__group_85__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_85__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_85__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_86__weights (optional) - T2: weights to optimize.
__group_86__gradients (optional) - T3: gradients computed in this iteration.
__group_86__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_86__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_86__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_87__weights (optional) - T2: weights to optimize.
__group_87__gradients (optional) - T3: gradients computed in this iteration.
__group_87__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_87__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_87__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_88__weights (optional) - T2: weights to optimize.
__group_88__gradients (optional) - T3: gradients computed in this iteration.
__group_88__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_88__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_88__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_89__weights (optional) - T2: weights to optimize.
__group_89__gradients (optional) - T3: gradients computed in this iteration.
__group_89__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_89__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_89__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_90__weights (optional) - T2: weights to optimize.
__group_90__gradients (optional) - T3: gradients computed in this iteration.
__group_90__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_90__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_90__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_91__weights (optional) - T2: weights to optimize.
__group_91__gradients (optional) - T3: gradients computed in this iteration.
__group_91__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_91__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_91__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_92__weights (optional) - T2: weights to optimize.
__group_92__gradients (optional) - T3: gradients computed in this iteration.
__group_92__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_92__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_92__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_93__weights (optional) - T2: weights to optimize.
__group_93__gradients (optional) - T3: gradients computed in this iteration.
__group_93__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_93__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_93__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_94__weights (optional) - T2: weights to optimize.
__group_94__gradients (optional) - T3: gradients computed in this iteration.
__group_94__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_94__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_94__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_95__weights (optional) - T2: weights to optimize.
__group_95__gradients (optional) - T3: gradients computed in this iteration.
__group_95__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_95__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_95__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_96__weights (optional) - T2: weights to optimize.
__group_96__gradients (optional) - T3: gradients computed in this iteration.
__group_96__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_96__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_96__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_97__weights (optional) - T2: weights to optimize.
__group_97__gradients (optional) - T3: gradients computed in this iteration.
__group_97__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_97__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_97__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_98__weights (optional) - T2: weights to optimize.
__group_98__gradients (optional) - T3: gradients computed in this iteration.
__group_98__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_98__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_98__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_99__weights (optional) - T2: weights to optimize.
__group_99__gradients (optional) - T3: gradients computed in this iteration.
__group_99__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_99__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_99__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_100__weights (optional) - T2: weights to optimize.
__group_100__gradients (optional) - T3: gradients computed in this iteration.
__group_100__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_100__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_100__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_101__weights (optional) - T2: weights to optimize.
__group_101__gradients (optional) - T3: gradients computed in this iteration.
__group_101__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_101__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_101__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_102__weights (optional) - T2: weights to optimize.
__group_102__gradients (optional) - T3: gradients computed in this iteration.
__group_102__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_102__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_102__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_103__weights (optional) - T2: weights to optimize.
__group_103__gradients (optional) - T3: gradients computed in this iteration.
__group_103__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_103__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_103__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_104__weights (optional) - T2: weights to optimize.
__group_104__gradients (optional) - T3: gradients computed in this iteration.
__group_104__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_104__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_104__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_105__weights (optional) - T2: weights to optimize.
__group_105__gradients (optional) - T3: gradients computed in this iteration.
__group_105__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_105__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_105__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_106__weights (optional) - T2: weights to optimize.
__group_106__gradients (optional) - T3: gradients computed in this iteration.
__group_106__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_106__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_106__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_107__weights (optional) - T2: weights to optimize.
__group_107__gradients (optional) - T3: gradients computed in this iteration.
__group_107__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_107__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_107__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_108__weights (optional) - T2: weights to optimize.
__group_108__gradients (optional) - T3: gradients computed in this iteration.
__group_108__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_108__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_108__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_109__weights (optional) - T2: weights to optimize.
__group_109__gradients (optional) - T3: gradients computed in this iteration.
__group_109__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_109__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_109__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_110__weights (optional) - T2: weights to optimize.
__group_110__gradients (optional) - T3: gradients computed in this iteration.
__group_110__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_110__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_110__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_111__weights (optional) - T2: weights to optimize.
__group_111__gradients (optional) - T3: gradients computed in this iteration.
__group_111__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_111__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_111__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_112__weights (optional) - T2: weights to optimize.
__group_112__gradients (optional) - T3: gradients computed in this iteration.
__group_112__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_112__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_112__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_113__weights (optional) - T2: weights to optimize.
__group_113__gradients (optional) - T3: gradients computed in this iteration.
__group_113__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_113__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_113__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_114__weights (optional) - T2: weights to optimize.
__group_114__gradients (optional) - T3: gradients computed in this iteration.
__group_114__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_114__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_114__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_115__weights (optional) - T2: weights to optimize.
__group_115__gradients (optional) - T3: gradients computed in this iteration.
__group_115__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_115__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_115__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_116__weights (optional) - T2: weights to optimize.
__group_116__gradients (optional) - T3: gradients computed in this iteration.
__group_116__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_116__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_116__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_117__weights (optional) - T2: weights to optimize.
__group_117__gradients (optional) - T3: gradients computed in this iteration.
__group_117__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_117__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_117__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_118__weights (optional) - T2: weights to optimize.
__group_118__gradients (optional) - T3: gradients computed in this iteration.
__group_118__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_118__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_118__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_119__weights (optional) - T2: weights to optimize.
__group_119__gradients (optional) - T3: gradients computed in this iteration.
__group_119__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_119__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_119__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_120__weights (optional) - T2: weights to optimize.
__group_120__gradients (optional) - T3: gradients computed in this iteration.
__group_120__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_120__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_120__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_121__weights (optional) - T2: weights to optimize.
__group_121__gradients (optional) - T3: gradients computed in this iteration.
__group_121__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_121__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_121__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_122__weights (optional) - T2: weights to optimize.
__group_122__gradients (optional) - T3: gradients computed in this iteration.
__group_122__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_122__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_122__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_123__weights (optional) - T2: weights to optimize.
__group_123__gradients (optional) - T3: gradients computed in this iteration.
__group_123__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_123__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_123__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_124__weights (optional) - T2: weights to optimize.
__group_124__gradients (optional) - T3: gradients computed in this iteration.
__group_124__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_124__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_124__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_125__weights (optional) - T2: weights to optimize.
__group_125__gradients (optional) - T3: gradients computed in this iteration.
__group_125__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_125__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_125__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_126__weights (optional) - T2: weights to optimize.
__group_126__gradients (optional) - T3: gradients computed in this iteration.
__group_126__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_126__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_126__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_127__weights (optional) - T2: weights to optimize.
__group_127__gradients (optional) - T3: gradients computed in this iteration.
__group_127__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_127__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_127__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_128__weights (optional) - T2: weights to optimize.
__group_128__gradients (optional) - T3: gradients computed in this iteration.
__group_128__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_128__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_128__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_129__weights (optional) - T2: weights to optimize.
__group_129__gradients (optional) - T3: gradients computed in this iteration.
__group_129__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_129__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_129__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_130__weights (optional) - T2: weights to optimize.
__group_130__gradients (optional) - T3: gradients computed in this iteration.
__group_130__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_130__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_130__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_131__weights (optional) - T2: weights to optimize.
__group_131__gradients (optional) - T3: gradients computed in this iteration.
__group_131__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_131__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_131__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_132__weights (optional) - T2: weights to optimize.
__group_132__gradients (optional) - T3: gradients computed in this iteration.
__group_132__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_132__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_132__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_133__weights (optional) - T2: weights to optimize.
__group_133__gradients (optional) - T3: gradients computed in this iteration.
__group_133__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_133__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_133__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_134__weights (optional) - T2: weights to optimize.
__group_134__gradients (optional) - T3: gradients computed in this iteration.
__group_134__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_134__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_134__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_135__weights (optional) - T2: weights to optimize.
__group_135__gradients (optional) - T3: gradients computed in this iteration.
__group_135__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_135__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_135__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_136__weights (optional) - T2: weights to optimize.
__group_136__gradients (optional) - T3: gradients computed in this iteration.
__group_136__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_136__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_136__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_137__weights (optional) - T2: weights to optimize.
__group_137__gradients (optional) - T3: gradients computed in this iteration.
__group_137__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_137__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_137__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_138__weights (optional) - T2: weights to optimize.
__group_138__gradients (optional) - T3: gradients computed in this iteration.
__group_138__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_138__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_138__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_139__weights (optional) - T2: weights to optimize.
__group_139__gradients (optional) - T3: gradients computed in this iteration.
__group_139__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_139__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_139__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_140__weights (optional) - T2: weights to optimize.
__group_140__gradients (optional) - T3: gradients computed in this iteration.
__group_140__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_140__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_140__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_141__weights (optional) - T2: weights to optimize.
__group_141__gradients (optional) - T3: gradients computed in this iteration.
__group_141__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_141__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_141__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_142__weights (optional) - T2: weights to optimize.
__group_142__gradients (optional) - T3: gradients computed in this iteration.
__group_142__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_142__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_142__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_143__weights (optional) - T2: weights to optimize.
__group_143__gradients (optional) - T3: gradients computed in this iteration.
__group_143__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_143__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_143__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_144__weights (optional) - T2: weights to optimize.
__group_144__gradients (optional) - T3: gradients computed in this iteration.
__group_144__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_144__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_144__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_145__weights (optional) - T2: weights to optimize.
__group_145__gradients (optional) - T3: gradients computed in this iteration.
__group_145__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_145__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_145__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_146__weights (optional) - T2: weights to optimize.
__group_146__gradients (optional) - T3: gradients computed in this iteration.
__group_146__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_146__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_146__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_147__weights (optional) - T2: weights to optimize.
__group_147__gradients (optional) - T3: gradients computed in this iteration.
__group_147__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_147__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_147__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_148__weights (optional) - T2: weights to optimize.
__group_148__gradients (optional) - T3: gradients computed in this iteration.
__group_148__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_148__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_148__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_149__weights (optional) - T2: weights to optimize.
__group_149__gradients (optional) - T3: gradients computed in this iteration.
__group_149__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_149__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_149__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_150__weights (optional) - T2: weights to optimize.
__group_150__gradients (optional) - T3: gradients computed in this iteration.
__group_150__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_150__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_150__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_151__weights (optional) - T2: weights to optimize.
__group_151__gradients (optional) - T3: gradients computed in this iteration.
__group_151__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_151__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_151__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_152__weights (optional) - T2: weights to optimize.
__group_152__gradients (optional) - T3: gradients computed in this iteration.
__group_152__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_152__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_152__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_153__weights (optional) - T2: weights to optimize.
__group_153__gradients (optional) - T3: gradients computed in this iteration.
__group_153__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_153__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_153__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_154__weights (optional) - T2: weights to optimize.
__group_154__gradients (optional) - T3: gradients computed in this iteration.
__group_154__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_154__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_154__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_155__weights (optional) - T2: weights to optimize.
__group_155__gradients (optional) - T3: gradients computed in this iteration.
__group_155__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_155__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_155__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_156__weights (optional) - T2: weights to optimize.
__group_156__gradients (optional) - T3: gradients computed in this iteration.
__group_156__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_156__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_156__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_157__weights (optional) - T2: weights to optimize.
__group_157__gradients (optional) - T3: gradients computed in this iteration.
__group_157__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_157__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_157__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_158__weights (optional) - T2: weights to optimize.
__group_158__gradients (optional) - T3: gradients computed in this iteration.
__group_158__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_158__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_158__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_159__weights (optional) - T2: weights to optimize.
__group_159__gradients (optional) - T3: gradients computed in this iteration.
__group_159__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_159__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_159__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_160__weights (optional) - T2: weights to optimize.
__group_160__gradients (optional) - T3: gradients computed in this iteration.
__group_160__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_160__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_160__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_161__weights (optional) - T2: weights to optimize.
__group_161__gradients (optional) - T3: gradients computed in this iteration.
__group_161__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_161__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_161__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_162__weights (optional) - T2: weights to optimize.
__group_162__gradients (optional) - T3: gradients computed in this iteration.
__group_162__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_162__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_162__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_163__weights (optional) - T2: weights to optimize.
__group_163__gradients (optional) - T3: gradients computed in this iteration.
__group_163__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_163__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_163__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_164__weights (optional) - T2: weights to optimize.
__group_164__gradients (optional) - T3: gradients computed in this iteration.
__group_164__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_164__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_164__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_165__weights (optional) - T2: weights to optimize.
__group_165__gradients (optional) - T3: gradients computed in this iteration.
__group_165__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_165__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_165__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_166__weights (optional) - T2: weights to optimize.
__group_166__gradients (optional) - T3: gradients computed in this iteration.
__group_166__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_166__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_166__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_167__weights (optional) - T2: weights to optimize.
__group_167__gradients (optional) - T3: gradients computed in this iteration.
__group_167__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_167__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_167__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_168__weights (optional) - T2: weights to optimize.
__group_168__gradients (optional) - T3: gradients computed in this iteration.
__group_168__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_168__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_168__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_169__weights (optional) - T2: weights to optimize.
__group_169__gradients (optional) - T3: gradients computed in this iteration.
__group_169__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_169__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_169__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_170__weights (optional) - T2: weights to optimize.
__group_170__gradients (optional) - T3: gradients computed in this iteration.
__group_170__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_170__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_170__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_171__weights (optional) - T2: weights to optimize.
__group_171__gradients (optional) - T3: gradients computed in this iteration.
__group_171__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_171__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_171__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_172__weights (optional) - T2: weights to optimize.
__group_172__gradients (optional) - T3: gradients computed in this iteration.
__group_172__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_172__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_172__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_173__weights (optional) - T2: weights to optimize.
__group_173__gradients (optional) - T3: gradients computed in this iteration.
__group_173__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_173__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_173__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_174__weights (optional) - T2: weights to optimize.
__group_174__gradients (optional) - T3: gradients computed in this iteration.
__group_174__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_174__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_174__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_175__weights (optional) - T2: weights to optimize.
__group_175__gradients (optional) - T3: gradients computed in this iteration.
__group_175__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_175__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_175__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_176__weights (optional) - T2: weights to optimize.
__group_176__gradients (optional) - T3: gradients computed in this iteration.
__group_176__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_176__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_176__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_177__weights (optional) - T2: weights to optimize.
__group_177__gradients (optional) - T3: gradients computed in this iteration.
__group_177__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_177__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_177__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_178__weights (optional) - T2: weights to optimize.
__group_178__gradients (optional) - T3: gradients computed in this iteration.
__group_178__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_178__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_178__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_179__weights (optional) - T2: weights to optimize.
__group_179__gradients (optional) - T3: gradients computed in this iteration.
__group_179__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_179__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_179__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_180__weights (optional) - T2: weights to optimize.
__group_180__gradients (optional) - T3: gradients computed in this iteration.
__group_180__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_180__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_180__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_181__weights (optional) - T2: weights to optimize.
__group_181__gradients (optional) - T3: gradients computed in this iteration.
__group_181__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_181__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_181__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_182__weights (optional) - T2: weights to optimize.
__group_182__gradients (optional) - T3: gradients computed in this iteration.
__group_182__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_182__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_182__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_183__weights (optional) - T2: weights to optimize.
__group_183__gradients (optional) - T3: gradients computed in this iteration.
__group_183__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_183__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_183__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_184__weights (optional) - T2: weights to optimize.
__group_184__gradients (optional) - T3: gradients computed in this iteration.
__group_184__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_184__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_184__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_185__weights (optional) - T2: weights to optimize.
__group_185__gradients (optional) - T3: gradients computed in this iteration.
__group_185__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_185__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_185__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_186__weights (optional) - T2: weights to optimize.
__group_186__gradients (optional) - T3: gradients computed in this iteration.
__group_186__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_186__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_186__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_187__weights (optional) - T2: weights to optimize.
__group_187__gradients (optional) - T3: gradients computed in this iteration.
__group_187__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_187__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_187__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_188__weights (optional) - T2: weights to optimize.
__group_188__gradients (optional) - T3: gradients computed in this iteration.
__group_188__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_188__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_188__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_189__weights (optional) - T2: weights to optimize.
__group_189__gradients (optional) - T3: gradients computed in this iteration.
__group_189__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_189__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_189__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_190__weights (optional) - T2: weights to optimize.
__group_190__gradients (optional) - T3: gradients computed in this iteration.
__group_190__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_190__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_190__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_191__weights (optional) - T2: weights to optimize.
__group_191__gradients (optional) - T3: gradients computed in this iteration.
__group_191__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_191__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_191__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_192__weights (optional) - T2: weights to optimize.
__group_192__gradients (optional) - T3: gradients computed in this iteration.
__group_192__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_192__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_192__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_193__weights (optional) - T2: weights to optimize.
__group_193__gradients (optional) - T3: gradients computed in this iteration.
__group_193__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_193__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_193__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_194__weights (optional) - T2: weights to optimize.
__group_194__gradients (optional) - T3: gradients computed in this iteration.
__group_194__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_194__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_194__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_195__weights (optional) - T2: weights to optimize.
__group_195__gradients (optional) - T3: gradients computed in this iteration.
__group_195__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_195__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_195__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_196__weights (optional) - T2: weights to optimize.
__group_196__gradients (optional) - T3: gradients computed in this iteration.
__group_196__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_196__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_196__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_197__weights (optional) - T2: weights to optimize.
__group_197__gradients (optional) - T3: gradients computed in this iteration.
__group_197__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_197__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_197__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_198__weights (optional) - T2: weights to optimize.
__group_198__gradients (optional) - T3: gradients computed in this iteration.
__group_198__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_198__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_198__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_199__weights (optional) - T2: weights to optimize.
__group_199__gradients (optional) - T3: gradients computed in this iteration.
__group_199__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_199__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_199__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_200__weights (optional) - T2: weights to optimize.
__group_200__gradients (optional) - T3: gradients computed in this iteration.
__group_200__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_200__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_200__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_201__weights (optional) - T2: weights to optimize.
__group_201__gradients (optional) - T3: gradients computed in this iteration.
__group_201__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_201__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_201__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_202__weights (optional) - T2: weights to optimize.
__group_202__gradients (optional) - T3: gradients computed in this iteration.
__group_202__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_202__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_202__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_203__weights (optional) - T2: weights to optimize.
__group_203__gradients (optional) - T3: gradients computed in this iteration.
__group_203__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_203__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_203__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_204__weights (optional) - T2: weights to optimize.
__group_204__gradients (optional) - T3: gradients computed in this iteration.
__group_204__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_204__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_204__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_205__weights (optional) - T2: weights to optimize.
__group_205__gradients (optional) - T3: gradients computed in this iteration.
__group_205__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_205__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_205__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_206__weights (optional) - T2: weights to optimize.
__group_206__gradients (optional) - T3: gradients computed in this iteration.
__group_206__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_206__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_206__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_207__weights (optional) - T2: weights to optimize.
__group_207__gradients (optional) - T3: gradients computed in this iteration.
__group_207__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_207__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_207__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_208__weights (optional) - T2: weights to optimize.
__group_208__gradients (optional) - T3: gradients computed in this iteration.
__group_208__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_208__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_208__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_209__weights (optional) - T2: weights to optimize.
__group_209__gradients (optional) - T3: gradients computed in this iteration.
__group_209__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_209__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_209__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_210__weights (optional) - T2: weights to optimize.
__group_210__gradients (optional) - T3: gradients computed in this iteration.
__group_210__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_210__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_210__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_211__weights (optional) - T2: weights to optimize.
__group_211__gradients (optional) - T3: gradients computed in this iteration.
__group_211__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_211__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_211__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_212__weights (optional) - T2: weights to optimize.
__group_212__gradients (optional) - T3: gradients computed in this iteration.
__group_212__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_212__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_212__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_213__weights (optional) - T2: weights to optimize.
__group_213__gradients (optional) - T3: gradients computed in this iteration.
__group_213__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_213__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_213__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_214__weights (optional) - T2: weights to optimize.
__group_214__gradients (optional) - T3: gradients computed in this iteration.
__group_214__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_214__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_214__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_215__weights (optional) - T2: weights to optimize.
__group_215__gradients (optional) - T3: gradients computed in this iteration.
__group_215__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_215__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_215__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_216__weights (optional) - T2: weights to optimize.
__group_216__gradients (optional) - T3: gradients computed in this iteration.
__group_216__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_216__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_216__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_217__weights (optional) - T2: weights to optimize.
__group_217__gradients (optional) - T3: gradients computed in this iteration.
__group_217__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_217__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_217__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_218__weights (optional) - T2: weights to optimize.
__group_218__gradients (optional) - T3: gradients computed in this iteration.
__group_218__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_218__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_218__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_219__weights (optional) - T2: weights to optimize.
__group_219__gradients (optional) - T3: gradients computed in this iteration.
__group_219__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_219__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_219__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_220__weights (optional) - T2: weights to optimize.
__group_220__gradients (optional) - T3: gradients computed in this iteration.
__group_220__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_220__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_220__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_221__weights (optional) - T2: weights to optimize.
__group_221__gradients (optional) - T3: gradients computed in this iteration.
__group_221__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_221__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_221__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_222__weights (optional) - T2: weights to optimize.
__group_222__gradients (optional) - T3: gradients computed in this iteration.
__group_222__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_222__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_222__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_223__weights (optional) - T2: weights to optimize.
__group_223__gradients (optional) - T3: gradients computed in this iteration.
__group_223__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_223__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_223__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_224__weights (optional) - T2: weights to optimize.
__group_224__gradients (optional) - T3: gradients computed in this iteration.
__group_224__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_224__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_224__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_225__weights (optional) - T2: weights to optimize.
__group_225__gradients (optional) - T3: gradients computed in this iteration.
__group_225__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_225__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_225__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_226__weights (optional) - T2: weights to optimize.
__group_226__gradients (optional) - T3: gradients computed in this iteration.
__group_226__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_226__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_226__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_227__weights (optional) - T2: weights to optimize.
__group_227__gradients (optional) - T3: gradients computed in this iteration.
__group_227__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_227__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_227__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_228__weights (optional) - T2: weights to optimize.
__group_228__gradients (optional) - T3: gradients computed in this iteration.
__group_228__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_228__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_228__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_229__weights (optional) - T2: weights to optimize.
__group_229__gradients (optional) - T3: gradients computed in this iteration.
__group_229__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_229__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_229__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_230__weights (optional) - T2: weights to optimize.
__group_230__gradients (optional) - T3: gradients computed in this iteration.
__group_230__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_230__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_230__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_231__weights (optional) - T2: weights to optimize.
__group_231__gradients (optional) - T3: gradients computed in this iteration.
__group_231__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_231__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_231__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_232__weights (optional) - T2: weights to optimize.
__group_232__gradients (optional) - T3: gradients computed in this iteration.
__group_232__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_232__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_232__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_233__weights (optional) - T2: weights to optimize.
__group_233__gradients (optional) - T3: gradients computed in this iteration.
__group_233__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_233__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_233__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_234__weights (optional) - T2: weights to optimize.
__group_234__gradients (optional) - T3: gradients computed in this iteration.
__group_234__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_234__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_234__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_235__weights (optional) - T2: weights to optimize.
__group_235__gradients (optional) - T3: gradients computed in this iteration.
__group_235__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_235__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_235__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_236__weights (optional) - T2: weights to optimize.
__group_236__gradients (optional) - T3: gradients computed in this iteration.
__group_236__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_236__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_236__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_237__weights (optional) - T2: weights to optimize.
__group_237__gradients (optional) - T3: gradients computed in this iteration.
__group_237__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_237__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_237__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_238__weights (optional) - T2: weights to optimize.
__group_238__gradients (optional) - T3: gradients computed in this iteration.
__group_238__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_238__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_238__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_239__weights (optional) - T2: weights to optimize.
__group_239__gradients (optional) - T3: gradients computed in this iteration.
__group_239__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_239__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_239__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_240__weights (optional) - T2: weights to optimize.
__group_240__gradients (optional) - T3: gradients computed in this iteration.
__group_240__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_240__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_240__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_241__weights (optional) - T2: weights to optimize.
__group_241__gradients (optional) - T3: gradients computed in this iteration.
__group_241__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_241__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_241__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_242__weights (optional) - T2: weights to optimize.
__group_242__gradients (optional) - T3: gradients computed in this iteration.
__group_242__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_242__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_242__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_243__weights (optional) - T2: weights to optimize.
__group_243__gradients (optional) - T3: gradients computed in this iteration.
__group_243__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_243__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_243__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_244__weights (optional) - T2: weights to optimize.
__group_244__gradients (optional) - T3: gradients computed in this iteration.
__group_244__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_244__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_244__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_245__weights (optional) - T2: weights to optimize.
__group_245__gradients (optional) - T3: gradients computed in this iteration.
__group_245__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_245__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_245__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_246__weights (optional) - T2: weights to optimize.
__group_246__gradients (optional) - T3: gradients computed in this iteration.
__group_246__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_246__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_246__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_247__weights (optional) - T2: weights to optimize.
__group_247__gradients (optional) - T3: gradients computed in this iteration.
__group_247__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_247__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_247__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_248__weights (optional) - T2: weights to optimize.
__group_248__gradients (optional) - T3: gradients computed in this iteration.
__group_248__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_248__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_248__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_249__weights (optional) - T2: weights to optimize.
__group_249__gradients (optional) - T3: gradients computed in this iteration.
__group_249__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_249__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_249__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_250__weights (optional) - T2: weights to optimize.
__group_250__gradients (optional) - T3: gradients computed in this iteration.
__group_250__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_250__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_250__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_251__weights (optional) - T2: weights to optimize.
__group_251__gradients (optional) - T3: gradients computed in this iteration.
__group_251__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_251__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_251__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_252__weights (optional) - T2: weights to optimize.
__group_252__gradients (optional) - T3: gradients computed in this iteration.
__group_252__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_252__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_252__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_253__weights (optional) - T2: weights to optimize.
__group_253__gradients (optional) - T3: gradients computed in this iteration.
__group_253__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_253__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_253__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_254__weights (optional) - T2: weights to optimize.
__group_254__gradients (optional) - T3: gradients computed in this iteration.
__group_254__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_254__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_254__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_255__weights (optional) - T2: weights to optimize.
__group_255__gradients (optional) - T3: gradients computed in this iteration.
__group_255__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_255__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_255__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_256__weights (optional) - T2: weights to optimize.
__group_256__gradients (optional) - T3: gradients computed in this iteration.
__group_256__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_256__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_256__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_257__weights (optional) - T2: weights to optimize.
__group_257__gradients (optional) - T3: gradients computed in this iteration.
__group_257__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_257__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_257__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_258__weights (optional) - T2: weights to optimize.
__group_258__gradients (optional) - T3: gradients computed in this iteration.
__group_258__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_258__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_258__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_259__weights (optional) - T2: weights to optimize.
__group_259__gradients (optional) - T3: gradients computed in this iteration.
__group_259__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_259__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_259__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_260__weights (optional) - T2: weights to optimize.
__group_260__gradients (optional) - T3: gradients computed in this iteration.
__group_260__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_260__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_260__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_261__weights (optional) - T2: weights to optimize.
__group_261__gradients (optional) - T3: gradients computed in this iteration.
__group_261__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_261__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_261__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_262__weights (optional) - T2: weights to optimize.
__group_262__gradients (optional) - T3: gradients computed in this iteration.
__group_262__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_262__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_262__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_263__weights (optional) - T2: weights to optimize.
__group_263__gradients (optional) - T3: gradients computed in this iteration.
__group_263__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_263__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_263__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_264__weights (optional) - T2: weights to optimize.
__group_264__gradients (optional) - T3: gradients computed in this iteration.
__group_264__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_264__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_264__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_265__weights (optional) - T2: weights to optimize.
__group_265__gradients (optional) - T3: gradients computed in this iteration.
__group_265__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_265__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_265__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_266__weights (optional) - T2: weights to optimize.
__group_266__gradients (optional) - T3: gradients computed in this iteration.
__group_266__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_266__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_266__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_267__weights (optional) - T2: weights to optimize.
__group_267__gradients (optional) - T3: gradients computed in this iteration.
__group_267__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_267__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_267__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_268__weights (optional) - T2: weights to optimize.
__group_268__gradients (optional) - T3: gradients computed in this iteration.
__group_268__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_268__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_268__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_269__weights (optional) - T2: weights to optimize.
__group_269__gradients (optional) - T3: gradients computed in this iteration.
__group_269__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_269__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_269__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_270__weights (optional) - T2: weights to optimize.
__group_270__gradients (optional) - T3: gradients computed in this iteration.
__group_270__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_270__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_270__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_271__weights (optional) - T2: weights to optimize.
__group_271__gradients (optional) - T3: gradients computed in this iteration.
__group_271__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_271__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_271__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_272__weights (optional) - T2: weights to optimize.
__group_272__gradients (optional) - T3: gradients computed in this iteration.
__group_272__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_272__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_272__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_273__weights (optional) - T2: weights to optimize.
__group_273__gradients (optional) - T3: gradients computed in this iteration.
__group_273__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_273__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_273__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_274__weights (optional) - T2: weights to optimize.
__group_274__gradients (optional) - T3: gradients computed in this iteration.
__group_274__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_274__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_274__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_275__weights (optional) - T2: weights to optimize.
__group_275__gradients (optional) - T3: gradients computed in this iteration.
__group_275__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_275__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_275__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_276__weights (optional) - T2: weights to optimize.
__group_276__gradients (optional) - T3: gradients computed in this iteration.
__group_276__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_276__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_276__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_277__weights (optional) - T2: weights to optimize.
__group_277__gradients (optional) - T3: gradients computed in this iteration.
__group_277__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_277__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_277__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_278__weights (optional) - T2: weights to optimize.
__group_278__gradients (optional) - T3: gradients computed in this iteration.
__group_278__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_278__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_278__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_279__weights (optional) - T2: weights to optimize.
__group_279__gradients (optional) - T3: gradients computed in this iteration.
__group_279__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_279__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_279__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_280__weights (optional) - T2: weights to optimize.
__group_280__gradients (optional) - T3: gradients computed in this iteration.
__group_280__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_280__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_280__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_281__weights (optional) - T2: weights to optimize.
__group_281__gradients (optional) - T3: gradients computed in this iteration.
__group_281__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_281__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_281__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_282__weights (optional) - T2: weights to optimize.
__group_282__gradients (optional) - T3: gradients computed in this iteration.
__group_282__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_282__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_282__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_283__weights (optional) - T2: weights to optimize.
__group_283__gradients (optional) - T3: gradients computed in this iteration.
__group_283__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_283__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_283__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_284__weights (optional) - T2: weights to optimize.
__group_284__gradients (optional) - T3: gradients computed in this iteration.
__group_284__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_284__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_284__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_285__weights (optional) - T2: weights to optimize.
__group_285__gradients (optional) - T3: gradients computed in this iteration.
__group_285__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_285__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_285__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_286__weights (optional) - T2: weights to optimize.
__group_286__gradients (optional) - T3: gradients computed in this iteration.
__group_286__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_286__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_286__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_287__weights (optional) - T2: weights to optimize.
__group_287__gradients (optional) - T3: gradients computed in this iteration.
__group_287__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_287__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_287__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_288__weights (optional) - T2: weights to optimize.
__group_288__gradients (optional) - T3: gradients computed in this iteration.
__group_288__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_288__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_288__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_289__weights (optional) - T2: weights to optimize.
__group_289__gradients (optional) - T3: gradients computed in this iteration.
__group_289__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_289__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_289__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_290__weights (optional) - T2: weights to optimize.
__group_290__gradients (optional) - T3: gradients computed in this iteration.
__group_290__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_290__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_290__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_291__weights (optional) - T2: weights to optimize.
__group_291__gradients (optional) - T3: gradients computed in this iteration.
__group_291__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_291__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_291__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_292__weights (optional) - T2: weights to optimize.
__group_292__gradients (optional) - T3: gradients computed in this iteration.
__group_292__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_292__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_292__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_293__weights (optional) - T2: weights to optimize.
__group_293__gradients (optional) - T3: gradients computed in this iteration.
__group_293__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_293__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_293__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_294__weights (optional) - T2: weights to optimize.
__group_294__gradients (optional) - T3: gradients computed in this iteration.
__group_294__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_294__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_294__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_295__weights (optional) - T2: weights to optimize.
__group_295__gradients (optional) - T3: gradients computed in this iteration.
__group_295__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_295__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_295__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_296__weights (optional) - T2: weights to optimize.
__group_296__gradients (optional) - T3: gradients computed in this iteration.
__group_296__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_296__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_296__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_297__weights (optional) - T2: weights to optimize.
__group_297__gradients (optional) - T3: gradients computed in this iteration.
__group_297__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_297__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_297__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_298__weights (optional) - T2: weights to optimize.
__group_298__gradients (optional) - T3: gradients computed in this iteration.
__group_298__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_298__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_298__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_299__weights (optional) - T2: weights to optimize.
__group_299__gradients (optional) - T3: gradients computed in this iteration.
__group_299__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_299__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_299__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_300__weights (optional) - T2: weights to optimize.
__group_300__gradients (optional) - T3: gradients computed in this iteration.
__group_300__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_300__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_300__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_301__weights (optional) - T2: weights to optimize.
__group_301__gradients (optional) - T3: gradients computed in this iteration.
__group_301__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_301__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_301__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_302__weights (optional) - T2: weights to optimize.
__group_302__gradients (optional) - T3: gradients computed in this iteration.
__group_302__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_302__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_302__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_303__weights (optional) - T2: weights to optimize.
__group_303__gradients (optional) - T3: gradients computed in this iteration.
__group_303__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_303__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_303__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_304__weights (optional) - T2: weights to optimize.
__group_304__gradients (optional) - T3: gradients computed in this iteration.
__group_304__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_304__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_304__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_305__weights (optional) - T2: weights to optimize.
__group_305__gradients (optional) - T3: gradients computed in this iteration.
__group_305__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_305__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_305__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_306__weights (optional) - T2: weights to optimize.
__group_306__gradients (optional) - T3: gradients computed in this iteration.
__group_306__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_306__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_306__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_307__weights (optional) - T2: weights to optimize.
__group_307__gradients (optional) - T3: gradients computed in this iteration.
__group_307__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_307__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_307__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_308__weights (optional) - T2: weights to optimize.
__group_308__gradients (optional) - T3: gradients computed in this iteration.
__group_308__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_308__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_308__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_309__weights (optional) - T2: weights to optimize.
__group_309__gradients (optional) - T3: gradients computed in this iteration.
__group_309__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_309__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_309__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_310__weights (optional) - T2: weights to optimize.
__group_310__gradients (optional) - T3: gradients computed in this iteration.
__group_310__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_310__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_310__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_311__weights (optional) - T2: weights to optimize.
__group_311__gradients (optional) - T3: gradients computed in this iteration.
__group_311__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_311__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_311__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_312__weights (optional) - T2: weights to optimize.
__group_312__gradients (optional) - T3: gradients computed in this iteration.
__group_312__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_312__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_312__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_313__weights (optional) - T2: weights to optimize.
__group_313__gradients (optional) - T3: gradients computed in this iteration.
__group_313__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_313__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_313__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_314__weights (optional) - T2: weights to optimize.
__group_314__gradients (optional) - T3: gradients computed in this iteration.
__group_314__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_314__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_314__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_315__weights (optional) - T2: weights to optimize.
__group_315__gradients (optional) - T3: gradients computed in this iteration.
__group_315__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_315__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_315__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_316__weights (optional) - T2: weights to optimize.
__group_316__gradients (optional) - T3: gradients computed in this iteration.
__group_316__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_316__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_316__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_317__weights (optional) - T2: weights to optimize.
__group_317__gradients (optional) - T3: gradients computed in this iteration.
__group_317__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_317__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_317__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_318__weights (optional) - T2: weights to optimize.
__group_318__gradients (optional) - T3: gradients computed in this iteration.
__group_318__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_318__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_318__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_319__weights (optional) - T2: weights to optimize.
__group_319__gradients (optional) - T3: gradients computed in this iteration.
__group_319__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_319__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_319__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_320__weights (optional) - T2: weights to optimize.
__group_320__gradients (optional) - T3: gradients computed in this iteration.
__group_320__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_320__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_320__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_321__weights (optional) - T2: weights to optimize.
__group_321__gradients (optional) - T3: gradients computed in this iteration.
__group_321__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_321__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_321__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_322__weights (optional) - T2: weights to optimize.
__group_322__gradients (optional) - T3: gradients computed in this iteration.
__group_322__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_322__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_322__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_323__weights (optional) - T2: weights to optimize.
__group_323__gradients (optional) - T3: gradients computed in this iteration.
__group_323__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_323__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_323__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_324__weights (optional) - T2: weights to optimize.
__group_324__gradients (optional) - T3: gradients computed in this iteration.
__group_324__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_324__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_324__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_325__weights (optional) - T2: weights to optimize.
__group_325__gradients (optional) - T3: gradients computed in this iteration.
__group_325__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_325__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_325__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_326__weights (optional) - T2: weights to optimize.
__group_326__gradients (optional) - T3: gradients computed in this iteration.
__group_326__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_326__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_326__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_327__weights (optional) - T2: weights to optimize.
__group_327__gradients (optional) - T3: gradients computed in this iteration.
__group_327__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_327__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_327__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_328__weights (optional) - T2: weights to optimize.
__group_328__gradients (optional) - T3: gradients computed in this iteration.
__group_328__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_328__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_328__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_329__weights (optional) - T2: weights to optimize.
__group_329__gradients (optional) - T3: gradients computed in this iteration.
__group_329__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_329__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_329__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_330__weights (optional) - T2: weights to optimize.
__group_330__gradients (optional) - T3: gradients computed in this iteration.
__group_330__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_330__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_330__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_331__weights (optional) - T2: weights to optimize.
__group_331__gradients (optional) - T3: gradients computed in this iteration.
__group_331__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_331__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_331__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_332__weights (optional) - T2: weights to optimize.
__group_332__gradients (optional) - T3: gradients computed in this iteration.
__group_332__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_332__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_332__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_333__weights (optional) - T2: weights to optimize.
__group_333__gradients (optional) - T3: gradients computed in this iteration.
__group_333__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_333__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_333__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_334__weights (optional) - T2: weights to optimize.
__group_334__gradients (optional) - T3: gradients computed in this iteration.
__group_334__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_334__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_334__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_335__weights (optional) - T2: weights to optimize.
__group_335__gradients (optional) - T3: gradients computed in this iteration.
__group_335__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_335__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_335__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_336__weights (optional) - T2: weights to optimize.
__group_336__gradients (optional) - T3: gradients computed in this iteration.
__group_336__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_336__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_336__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_337__weights (optional) - T2: weights to optimize.
__group_337__gradients (optional) - T3: gradients computed in this iteration.
__group_337__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_337__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_337__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_338__weights (optional) - T2: weights to optimize.
__group_338__gradients (optional) - T3: gradients computed in this iteration.
__group_338__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_338__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_338__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_339__weights (optional) - T2: weights to optimize.
__group_339__gradients (optional) - T3: gradients computed in this iteration.
__group_339__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_339__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_339__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_340__weights (optional) - T2: weights to optimize.
__group_340__gradients (optional) - T3: gradients computed in this iteration.
__group_340__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_340__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_340__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_341__weights (optional) - T2: weights to optimize.
__group_341__gradients (optional) - T3: gradients computed in this iteration.
__group_341__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_341__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_341__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_342__weights (optional) - T2: weights to optimize.
__group_342__gradients (optional) - T3: gradients computed in this iteration.
__group_342__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_342__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_342__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_343__weights (optional) - T2: weights to optimize.
__group_343__gradients (optional) - T3: gradients computed in this iteration.
__group_343__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_343__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_343__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_344__weights (optional) - T2: weights to optimize.
__group_344__gradients (optional) - T3: gradients computed in this iteration.
__group_344__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_344__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_344__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_345__weights (optional) - T2: weights to optimize.
__group_345__gradients (optional) - T3: gradients computed in this iteration.
__group_345__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_345__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_345__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_346__weights (optional) - T2: weights to optimize.
__group_346__gradients (optional) - T3: gradients computed in this iteration.
__group_346__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_346__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_346__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_347__weights (optional) - T2: weights to optimize.
__group_347__gradients (optional) - T3: gradients computed in this iteration.
__group_347__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_347__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_347__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_348__weights (optional) - T2: weights to optimize.
__group_348__gradients (optional) - T3: gradients computed in this iteration.
__group_348__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_348__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_348__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_349__weights (optional) - T2: weights to optimize.
__group_349__gradients (optional) - T3: gradients computed in this iteration.
__group_349__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_349__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_349__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_350__weights (optional) - T2: weights to optimize.
__group_350__gradients (optional) - T3: gradients computed in this iteration.
__group_350__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_350__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_350__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_351__weights (optional) - T2: weights to optimize.
__group_351__gradients (optional) - T3: gradients computed in this iteration.
__group_351__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_351__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_351__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_352__weights (optional) - T2: weights to optimize.
__group_352__gradients (optional) - T3: gradients computed in this iteration.
__group_352__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_352__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_352__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_353__weights (optional) - T2: weights to optimize.
__group_353__gradients (optional) - T3: gradients computed in this iteration.
__group_353__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_353__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_353__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_354__weights (optional) - T2: weights to optimize.
__group_354__gradients (optional) - T3: gradients computed in this iteration.
__group_354__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_354__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_354__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_355__weights (optional) - T2: weights to optimize.
__group_355__gradients (optional) - T3: gradients computed in this iteration.
__group_355__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_355__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_355__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_356__weights (optional) - T2: weights to optimize.
__group_356__gradients (optional) - T3: gradients computed in this iteration.
__group_356__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_356__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_356__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_357__weights (optional) - T2: weights to optimize.
__group_357__gradients (optional) - T3: gradients computed in this iteration.
__group_357__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_357__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_357__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_358__weights (optional) - T2: weights to optimize.
__group_358__gradients (optional) - T3: gradients computed in this iteration.
__group_358__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_358__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_358__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_359__weights (optional) - T2: weights to optimize.
__group_359__gradients (optional) - T3: gradients computed in this iteration.
__group_359__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_359__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_359__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_360__weights (optional) - T2: weights to optimize.
__group_360__gradients (optional) - T3: gradients computed in this iteration.
__group_360__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_360__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_360__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_361__weights (optional) - T2: weights to optimize.
__group_361__gradients (optional) - T3: gradients computed in this iteration.
__group_361__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_361__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_361__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_362__weights (optional) - T2: weights to optimize.
__group_362__gradients (optional) - T3: gradients computed in this iteration.
__group_362__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_362__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_362__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_363__weights (optional) - T2: weights to optimize.
__group_363__gradients (optional) - T3: gradients computed in this iteration.
__group_363__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_363__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_363__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_364__weights (optional) - T2: weights to optimize.
__group_364__gradients (optional) - T3: gradients computed in this iteration.
__group_364__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_364__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_364__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_365__weights (optional) - T2: weights to optimize.
__group_365__gradients (optional) - T3: gradients computed in this iteration.
__group_365__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_365__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_365__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_366__weights (optional) - T2: weights to optimize.
__group_366__gradients (optional) - T3: gradients computed in this iteration.
__group_366__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_366__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_366__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_367__weights (optional) - T2: weights to optimize.
__group_367__gradients (optional) - T3: gradients computed in this iteration.
__group_367__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_367__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_367__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_368__weights (optional) - T2: weights to optimize.
__group_368__gradients (optional) - T3: gradients computed in this iteration.
__group_368__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_368__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_368__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_369__weights (optional) - T2: weights to optimize.
__group_369__gradients (optional) - T3: gradients computed in this iteration.
__group_369__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_369__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_369__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_370__weights (optional) - T2: weights to optimize.
__group_370__gradients (optional) - T3: gradients computed in this iteration.
__group_370__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_370__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_370__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_371__weights (optional) - T2: weights to optimize.
__group_371__gradients (optional) - T3: gradients computed in this iteration.
__group_371__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_371__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_371__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_372__weights (optional) - T2: weights to optimize.
__group_372__gradients (optional) - T3: gradients computed in this iteration.
__group_372__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_372__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_372__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_373__weights (optional) - T2: weights to optimize.
__group_373__gradients (optional) - T3: gradients computed in this iteration.
__group_373__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_373__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_373__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_374__weights (optional) - T2: weights to optimize.
__group_374__gradients (optional) - T3: gradients computed in this iteration.
__group_374__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_374__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_374__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_375__weights (optional) - T2: weights to optimize.
__group_375__gradients (optional) - T3: gradients computed in this iteration.
__group_375__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_375__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_375__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_376__weights (optional) - T2: weights to optimize.
__group_376__gradients (optional) - T3: gradients computed in this iteration.
__group_376__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_376__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_376__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_377__weights (optional) - T2: weights to optimize.
__group_377__gradients (optional) - T3: gradients computed in this iteration.
__group_377__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_377__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_377__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_378__weights (optional) - T2: weights to optimize.
__group_378__gradients (optional) - T3: gradients computed in this iteration.
__group_378__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_378__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_378__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_379__weights (optional) - T2: weights to optimize.
__group_379__gradients (optional) - T3: gradients computed in this iteration.
__group_379__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_379__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_379__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_380__weights (optional) - T2: weights to optimize.
__group_380__gradients (optional) - T3: gradients computed in this iteration.
__group_380__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_380__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_380__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_381__weights (optional) - T2: weights to optimize.
__group_381__gradients (optional) - T3: gradients computed in this iteration.
__group_381__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_381__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_381__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_382__weights (optional) - T2: weights to optimize.
__group_382__gradients (optional) - T3: gradients computed in this iteration.
__group_382__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_382__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_382__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_383__weights (optional) - T2: weights to optimize.
__group_383__gradients (optional) - T3: gradients computed in this iteration.
__group_383__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_383__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_383__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_384__weights (optional) - T2: weights to optimize.
__group_384__gradients (optional) - T3: gradients computed in this iteration.
__group_384__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_384__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_384__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_385__weights (optional) - T2: weights to optimize.
__group_385__gradients (optional) - T3: gradients computed in this iteration.
__group_385__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_385__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_385__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_386__weights (optional) - T2: weights to optimize.
__group_386__gradients (optional) - T3: gradients computed in this iteration.
__group_386__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_386__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_386__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_387__weights (optional) - T2: weights to optimize.
__group_387__gradients (optional) - T3: gradients computed in this iteration.
__group_387__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_387__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_387__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_388__weights (optional) - T2: weights to optimize.
__group_388__gradients (optional) - T3: gradients computed in this iteration.
__group_388__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_388__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_388__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_389__weights (optional) - T2: weights to optimize.
__group_389__gradients (optional) - T3: gradients computed in this iteration.
__group_389__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_389__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_389__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_390__weights (optional) - T2: weights to optimize.
__group_390__gradients (optional) - T3: gradients computed in this iteration.
__group_390__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_390__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_390__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_391__weights (optional) - T2: weights to optimize.
__group_391__gradients (optional) - T3: gradients computed in this iteration.
__group_391__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_391__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_391__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_392__weights (optional) - T2: weights to optimize.
__group_392__gradients (optional) - T3: gradients computed in this iteration.
__group_392__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_392__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_392__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_393__weights (optional) - T2: weights to optimize.
__group_393__gradients (optional) - T3: gradients computed in this iteration.
__group_393__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_393__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_393__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_394__weights (optional) - T2: weights to optimize.
__group_394__gradients (optional) - T3: gradients computed in this iteration.
__group_394__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_394__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_394__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_395__weights (optional) - T2: weights to optimize.
__group_395__gradients (optional) - T3: gradients computed in this iteration.
__group_395__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_395__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_395__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_396__weights (optional) - T2: weights to optimize.
__group_396__gradients (optional) - T3: gradients computed in this iteration.
__group_396__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_396__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_396__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_397__weights (optional) - T2: weights to optimize.
__group_397__gradients (optional) - T3: gradients computed in this iteration.
__group_397__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_397__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_397__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_398__weights (optional) - T2: weights to optimize.
__group_398__gradients (optional) - T3: gradients computed in this iteration.
__group_398__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_398__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_398__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_399__weights (optional) - T2: weights to optimize.
__group_399__gradients (optional) - T3: gradients computed in this iteration.
__group_399__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_399__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_399__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_400__weights (optional) - T2: weights to optimize.
__group_400__gradients (optional) - T3: gradients computed in this iteration.
__group_400__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_400__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_400__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_401__weights (optional) - T2: weights to optimize.
__group_401__gradients (optional) - T3: gradients computed in this iteration.
__group_401__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_401__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_401__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_402__weights (optional) - T2: weights to optimize.
__group_402__gradients (optional) - T3: gradients computed in this iteration.
__group_402__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_402__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_402__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_403__weights (optional) - T2: weights to optimize.
__group_403__gradients (optional) - T3: gradients computed in this iteration.
__group_403__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_403__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_403__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_404__weights (optional) - T2: weights to optimize.
__group_404__gradients (optional) - T3: gradients computed in this iteration.
__group_404__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_404__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_404__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_405__weights (optional) - T2: weights to optimize.
__group_405__gradients (optional) - T3: gradients computed in this iteration.
__group_405__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_405__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_405__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_406__weights (optional) - T2: weights to optimize.
__group_406__gradients (optional) - T3: gradients computed in this iteration.
__group_406__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_406__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_406__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_407__weights (optional) - T2: weights to optimize.
__group_407__gradients (optional) - T3: gradients computed in this iteration.
__group_407__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_407__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_407__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_408__weights (optional) - T2: weights to optimize.
__group_408__gradients (optional) - T3: gradients computed in this iteration.
__group_408__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_408__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_408__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_409__weights (optional) - T2: weights to optimize.
__group_409__gradients (optional) - T3: gradients computed in this iteration.
__group_409__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_409__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_409__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_410__weights (optional) - T2: weights to optimize.
__group_410__gradients (optional) - T3: gradients computed in this iteration.
__group_410__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_410__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_410__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_411__weights (optional) - T2: weights to optimize.
__group_411__gradients (optional) - T3: gradients computed in this iteration.
__group_411__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_411__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_411__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_412__weights (optional) - T2: weights to optimize.
__group_412__gradients (optional) - T3: gradients computed in this iteration.
__group_412__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_412__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_412__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_413__weights (optional) - T2: weights to optimize.
__group_413__gradients (optional) - T3: gradients computed in this iteration.
__group_413__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_413__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_413__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_414__weights (optional) - T2: weights to optimize.
__group_414__gradients (optional) - T3: gradients computed in this iteration.
__group_414__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_414__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_414__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_415__weights (optional) - T2: weights to optimize.
__group_415__gradients (optional) - T3: gradients computed in this iteration.
__group_415__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_415__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_415__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_416__weights (optional) - T2: weights to optimize.
__group_416__gradients (optional) - T3: gradients computed in this iteration.
__group_416__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_416__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_416__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_417__weights (optional) - T2: weights to optimize.
__group_417__gradients (optional) - T3: gradients computed in this iteration.
__group_417__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_417__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_417__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_418__weights (optional) - T2: weights to optimize.
__group_418__gradients (optional) - T3: gradients computed in this iteration.
__group_418__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_418__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_418__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_419__weights (optional) - T2: weights to optimize.
__group_419__gradients (optional) - T3: gradients computed in this iteration.
__group_419__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_419__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_419__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_420__weights (optional) - T2: weights to optimize.
__group_420__gradients (optional) - T3: gradients computed in this iteration.
__group_420__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_420__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_420__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_421__weights (optional) - T2: weights to optimize.
__group_421__gradients (optional) - T3: gradients computed in this iteration.
__group_421__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_421__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_421__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_422__weights (optional) - T2: weights to optimize.
__group_422__gradients (optional) - T3: gradients computed in this iteration.
__group_422__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_422__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_422__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_423__weights (optional) - T2: weights to optimize.
__group_423__gradients (optional) - T3: gradients computed in this iteration.
__group_423__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_423__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_423__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_424__weights (optional) - T2: weights to optimize.
__group_424__gradients (optional) - T3: gradients computed in this iteration.
__group_424__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_424__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_424__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_425__weights (optional) - T2: weights to optimize.
__group_425__gradients (optional) - T3: gradients computed in this iteration.
__group_425__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_425__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_425__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_426__weights (optional) - T2: weights to optimize.
__group_426__gradients (optional) - T3: gradients computed in this iteration.
__group_426__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_426__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_426__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_427__weights (optional) - T2: weights to optimize.
__group_427__gradients (optional) - T3: gradients computed in this iteration.
__group_427__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_427__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_427__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_428__weights (optional) - T2: weights to optimize.
__group_428__gradients (optional) - T3: gradients computed in this iteration.
__group_428__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_428__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_428__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_429__weights (optional) - T2: weights to optimize.
__group_429__gradients (optional) - T3: gradients computed in this iteration.
__group_429__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_429__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_429__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_430__weights (optional) - T2: weights to optimize.
__group_430__gradients (optional) - T3: gradients computed in this iteration.
__group_430__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_430__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_430__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_431__weights (optional) - T2: weights to optimize.
__group_431__gradients (optional) - T3: gradients computed in this iteration.
__group_431__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_431__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_431__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_432__weights (optional) - T2: weights to optimize.
__group_432__gradients (optional) - T3: gradients computed in this iteration.
__group_432__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_432__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_432__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_433__weights (optional) - T2: weights to optimize.
__group_433__gradients (optional) - T3: gradients computed in this iteration.
__group_433__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_433__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_433__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_434__weights (optional) - T2: weights to optimize.
__group_434__gradients (optional) - T3: gradients computed in this iteration.
__group_434__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_434__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_434__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_435__weights (optional) - T2: weights to optimize.
__group_435__gradients (optional) - T3: gradients computed in this iteration.
__group_435__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_435__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_435__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_436__weights (optional) - T2: weights to optimize.
__group_436__gradients (optional) - T3: gradients computed in this iteration.
__group_436__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_436__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_436__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_437__weights (optional) - T2: weights to optimize.
__group_437__gradients (optional) - T3: gradients computed in this iteration.
__group_437__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_437__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_437__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_438__weights (optional) - T2: weights to optimize.
__group_438__gradients (optional) - T3: gradients computed in this iteration.
__group_438__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_438__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_438__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_439__weights (optional) - T2: weights to optimize.
__group_439__gradients (optional) - T3: gradients computed in this iteration.
__group_439__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_439__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_439__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_440__weights (optional) - T2: weights to optimize.
__group_440__gradients (optional) - T3: gradients computed in this iteration.
__group_440__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_440__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_440__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_441__weights (optional) - T2: weights to optimize.
__group_441__gradients (optional) - T3: gradients computed in this iteration.
__group_441__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_441__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_441__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_442__weights (optional) - T2: weights to optimize.
__group_442__gradients (optional) - T3: gradients computed in this iteration.
__group_442__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_442__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_442__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_443__weights (optional) - T2: weights to optimize.
__group_443__gradients (optional) - T3: gradients computed in this iteration.
__group_443__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_443__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_443__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_444__weights (optional) - T2: weights to optimize.
__group_444__gradients (optional) - T3: gradients computed in this iteration.
__group_444__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_444__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_444__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_445__weights (optional) - T2: weights to optimize.
__group_445__gradients (optional) - T3: gradients computed in this iteration.
__group_445__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_445__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_445__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_446__weights (optional) - T2: weights to optimize.
__group_446__gradients (optional) - T3: gradients computed in this iteration.
__group_446__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_446__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_446__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_447__weights (optional) - T2: weights to optimize.
__group_447__gradients (optional) - T3: gradients computed in this iteration.
__group_447__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_447__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_447__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_448__weights (optional) - T2: weights to optimize.
__group_448__gradients (optional) - T3: gradients computed in this iteration.
__group_448__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_448__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_448__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_449__weights (optional) - T2: weights to optimize.
__group_449__gradients (optional) - T3: gradients computed in this iteration.
__group_449__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_449__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_449__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_450__weights (optional) - T2: weights to optimize.
__group_450__gradients (optional) - T3: gradients computed in this iteration.
__group_450__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_450__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_450__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_451__weights (optional) - T2: weights to optimize.
__group_451__gradients (optional) - T3: gradients computed in this iteration.
__group_451__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_451__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_451__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_452__weights (optional) - T2: weights to optimize.
__group_452__gradients (optional) - T3: gradients computed in this iteration.
__group_452__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_452__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_452__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_453__weights (optional) - T2: weights to optimize.
__group_453__gradients (optional) - T3: gradients computed in this iteration.
__group_453__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_453__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_453__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_454__weights (optional) - T2: weights to optimize.
__group_454__gradients (optional) - T3: gradients computed in this iteration.
__group_454__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_454__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_454__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_455__weights (optional) - T2: weights to optimize.
__group_455__gradients (optional) - T3: gradients computed in this iteration.
__group_455__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_455__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_455__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_456__weights (optional) - T2: weights to optimize.
__group_456__gradients (optional) - T3: gradients computed in this iteration.
__group_456__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_456__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_456__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_457__weights (optional) - T2: weights to optimize.
__group_457__gradients (optional) - T3: gradients computed in this iteration.
__group_457__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_457__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_457__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_458__weights (optional) - T2: weights to optimize.
__group_458__gradients (optional) - T3: gradients computed in this iteration.
__group_458__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_458__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_458__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_459__weights (optional) - T2: weights to optimize.
__group_459__gradients (optional) - T3: gradients computed in this iteration.
__group_459__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_459__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_459__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_460__weights (optional) - T2: weights to optimize.
__group_460__gradients (optional) - T3: gradients computed in this iteration.
__group_460__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_460__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_460__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_461__weights (optional) - T2: weights to optimize.
__group_461__gradients (optional) - T3: gradients computed in this iteration.
__group_461__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_461__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_461__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_462__weights (optional) - T2: weights to optimize.
__group_462__gradients (optional) - T3: gradients computed in this iteration.
__group_462__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_462__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_462__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_463__weights (optional) - T2: weights to optimize.
__group_463__gradients (optional) - T3: gradients computed in this iteration.
__group_463__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_463__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_463__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_464__weights (optional) - T2: weights to optimize.
__group_464__gradients (optional) - T3: gradients computed in this iteration.
__group_464__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_464__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_464__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_465__weights (optional) - T2: weights to optimize.
__group_465__gradients (optional) - T3: gradients computed in this iteration.
__group_465__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_465__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_465__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_466__weights (optional) - T2: weights to optimize.
__group_466__gradients (optional) - T3: gradients computed in this iteration.
__group_466__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_466__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_466__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_467__weights (optional) - T2: weights to optimize.
__group_467__gradients (optional) - T3: gradients computed in this iteration.
__group_467__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_467__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_467__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_468__weights (optional) - T2: weights to optimize.
__group_468__gradients (optional) - T3: gradients computed in this iteration.
__group_468__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_468__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_468__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_469__weights (optional) - T2: weights to optimize.
__group_469__gradients (optional) - T3: gradients computed in this iteration.
__group_469__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_469__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_469__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_470__weights (optional) - T2: weights to optimize.
__group_470__gradients (optional) - T3: gradients computed in this iteration.
__group_470__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_470__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_470__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_471__weights (optional) - T2: weights to optimize.
__group_471__gradients (optional) - T3: gradients computed in this iteration.
__group_471__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_471__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_471__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_472__weights (optional) - T2: weights to optimize.
__group_472__gradients (optional) - T3: gradients computed in this iteration.
__group_472__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_472__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_472__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_473__weights (optional) - T2: weights to optimize.
__group_473__gradients (optional) - T3: gradients computed in this iteration.
__group_473__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_473__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_473__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_474__weights (optional) - T2: weights to optimize.
__group_474__gradients (optional) - T3: gradients computed in this iteration.
__group_474__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_474__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_474__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_475__weights (optional) - T2: weights to optimize.
__group_475__gradients (optional) - T3: gradients computed in this iteration.
__group_475__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_475__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_475__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_476__weights (optional) - T2: weights to optimize.
__group_476__gradients (optional) - T3: gradients computed in this iteration.
__group_476__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_476__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_476__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_477__weights (optional) - T2: weights to optimize.
__group_477__gradients (optional) - T3: gradients computed in this iteration.
__group_477__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_477__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_477__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_478__weights (optional) - T2: weights to optimize.
__group_478__gradients (optional) - T3: gradients computed in this iteration.
__group_478__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_478__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_478__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_479__weights (optional) - T2: weights to optimize.
__group_479__gradients (optional) - T3: gradients computed in this iteration.
__group_479__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_479__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_479__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_480__weights (optional) - T2: weights to optimize.
__group_480__gradients (optional) - T3: gradients computed in this iteration.
__group_480__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_480__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_480__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_481__weights (optional) - T2: weights to optimize.
__group_481__gradients (optional) - T3: gradients computed in this iteration.
__group_481__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_481__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_481__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_482__weights (optional) - T2: weights to optimize.
__group_482__gradients (optional) - T3: gradients computed in this iteration.
__group_482__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_482__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_482__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_483__weights (optional) - T2: weights to optimize.
__group_483__gradients (optional) - T3: gradients computed in this iteration.
__group_483__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_483__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_483__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_484__weights (optional) - T2: weights to optimize.
__group_484__gradients (optional) - T3: gradients computed in this iteration.
__group_484__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_484__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_484__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_485__weights (optional) - T2: weights to optimize.
__group_485__gradients (optional) - T3: gradients computed in this iteration.
__group_485__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_485__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_485__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_486__weights (optional) - T2: weights to optimize.
__group_486__gradients (optional) - T3: gradients computed in this iteration.
__group_486__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_486__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_486__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_487__weights (optional) - T2: weights to optimize.
__group_487__gradients (optional) - T3: gradients computed in this iteration.
__group_487__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_487__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_487__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_488__weights (optional) - T2: weights to optimize.
__group_488__gradients (optional) - T3: gradients computed in this iteration.
__group_488__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_488__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_488__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_489__weights (optional) - T2: weights to optimize.
__group_489__gradients (optional) - T3: gradients computed in this iteration.
__group_489__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_489__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_489__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_490__weights (optional) - T2: weights to optimize.
__group_490__gradients (optional) - T3: gradients computed in this iteration.
__group_490__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_490__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_490__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_491__weights (optional) - T2: weights to optimize.
__group_491__gradients (optional) - T3: gradients computed in this iteration.
__group_491__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_491__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_491__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_492__weights (optional) - T2: weights to optimize.
__group_492__gradients (optional) - T3: gradients computed in this iteration.
__group_492__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_492__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_492__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_493__weights (optional) - T2: weights to optimize.
__group_493__gradients (optional) - T3: gradients computed in this iteration.
__group_493__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_493__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_493__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_494__weights (optional) - T2: weights to optimize.
__group_494__gradients (optional) - T3: gradients computed in this iteration.
__group_494__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_494__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_494__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_495__weights (optional) - T2: weights to optimize.
__group_495__gradients (optional) - T3: gradients computed in this iteration.
__group_495__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_495__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_495__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_496__weights (optional) - T2: weights to optimize.
__group_496__gradients (optional) - T3: gradients computed in this iteration.
__group_496__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_496__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_496__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_497__weights (optional) - T2: weights to optimize.
__group_497__gradients (optional) - T3: gradients computed in this iteration.
__group_497__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_497__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_497__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_498__weights (optional) - T2: weights to optimize.
__group_498__gradients (optional) - T3: gradients computed in this iteration.
__group_498__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_498__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_498__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_499__weights (optional) - T2: weights to optimize.
__group_499__gradients (optional) - T3: gradients computed in this iteration.
__group_499__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_499__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_499__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_500__weights (optional) - T2: weights to optimize.
__group_500__gradients (optional) - T3: gradients computed in this iteration.
__group_500__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_500__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_500__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_501__weights (optional) - T2: weights to optimize.
__group_501__gradients (optional) - T3: gradients computed in this iteration.
__group_501__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_501__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_501__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_502__weights (optional) - T2: weights to optimize.
__group_502__gradients (optional) - T3: gradients computed in this iteration.
__group_502__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_502__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_502__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_503__weights (optional) - T2: weights to optimize.
__group_503__gradients (optional) - T3: gradients computed in this iteration.
__group_503__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_503__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_503__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_504__weights (optional) - T2: weights to optimize.
__group_504__gradients (optional) - T3: gradients computed in this iteration.
__group_504__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_504__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_504__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_505__weights (optional) - T2: weights to optimize.
__group_505__gradients (optional) - T3: gradients computed in this iteration.
__group_505__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_505__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_505__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_506__weights (optional) - T2: weights to optimize.
__group_506__gradients (optional) - T3: gradients computed in this iteration.
__group_506__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_506__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_506__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_507__weights (optional) - T2: weights to optimize.
__group_507__gradients (optional) - T3: gradients computed in this iteration.
__group_507__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_507__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_507__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_508__weights (optional) - T2: weights to optimize.
__group_508__gradients (optional) - T3: gradients computed in this iteration.
__group_508__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_508__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_508__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_509__weights (optional) - T2: weights to optimize.
__group_509__gradients (optional) - T3: gradients computed in this iteration.
__group_509__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_509__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_509__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_510__weights (optional) - T2: weights to optimize.
__group_510__gradients (optional) - T3: gradients computed in this iteration.
__group_510__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_510__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_510__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_511__weights (optional) - T2: weights to optimize.
__group_511__gradients (optional) - T3: gradients computed in this iteration.
__group_511__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_511__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_511__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_512__weights (optional) - T2: weights to optimize.
__group_512__gradients (optional) - T3: gradients computed in this iteration.
__group_512__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_512__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_512__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_513__weights (optional) - T2: weights to optimize.
__group_513__gradients (optional) - T3: gradients computed in this iteration.
__group_513__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_513__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_513__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_514__weights (optional) - T2: weights to optimize.
__group_514__gradients (optional) - T3: gradients computed in this iteration.
__group_514__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_514__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_514__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_515__weights (optional) - T2: weights to optimize.
__group_515__gradients (optional) - T3: gradients computed in this iteration.
__group_515__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_515__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_515__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_516__weights (optional) - T2: weights to optimize.
__group_516__gradients (optional) - T3: gradients computed in this iteration.
__group_516__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_516__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_516__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_517__weights (optional) - T2: weights to optimize.
__group_517__gradients (optional) - T3: gradients computed in this iteration.
__group_517__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_517__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_517__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_518__weights (optional) - T2: weights to optimize.
__group_518__gradients (optional) - T3: gradients computed in this iteration.
__group_518__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_518__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_518__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_519__weights (optional) - T2: weights to optimize.
__group_519__gradients (optional) - T3: gradients computed in this iteration.
__group_519__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_519__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_519__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_520__weights (optional) - T2: weights to optimize.
__group_520__gradients (optional) - T3: gradients computed in this iteration.
__group_520__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_520__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_520__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_521__weights (optional) - T2: weights to optimize.
__group_521__gradients (optional) - T3: gradients computed in this iteration.
__group_521__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_521__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_521__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_522__weights (optional) - T2: weights to optimize.
__group_522__gradients (optional) - T3: gradients computed in this iteration.
__group_522__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_522__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_522__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_523__weights (optional) - T2: weights to optimize.
__group_523__gradients (optional) - T3: gradients computed in this iteration.
__group_523__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_523__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_523__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_524__weights (optional) - T2: weights to optimize.
__group_524__gradients (optional) - T3: gradients computed in this iteration.
__group_524__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_524__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_524__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_525__weights (optional) - T2: weights to optimize.
__group_525__gradients (optional) - T3: gradients computed in this iteration.
__group_525__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_525__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_525__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_526__weights (optional) - T2: weights to optimize.
__group_526__gradients (optional) - T3: gradients computed in this iteration.
__group_526__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_526__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_526__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_527__weights (optional) - T2: weights to optimize.
__group_527__gradients (optional) - T3: gradients computed in this iteration.
__group_527__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_527__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_527__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_528__weights (optional) - T2: weights to optimize.
__group_528__gradients (optional) - T3: gradients computed in this iteration.
__group_528__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_528__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_528__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_529__weights (optional) - T2: weights to optimize.
__group_529__gradients (optional) - T3: gradients computed in this iteration.
__group_529__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_529__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_529__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_530__weights (optional) - T2: weights to optimize.
__group_530__gradients (optional) - T3: gradients computed in this iteration.
__group_530__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_530__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_530__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_531__weights (optional) - T2: weights to optimize.
__group_531__gradients (optional) - T3: gradients computed in this iteration.
__group_531__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_531__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_531__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_532__weights (optional) - T2: weights to optimize.
__group_532__gradients (optional) - T3: gradients computed in this iteration.
__group_532__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_532__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_532__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_533__weights (optional) - T2: weights to optimize.
__group_533__gradients (optional) - T3: gradients computed in this iteration.
__group_533__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_533__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_533__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_534__weights (optional) - T2: weights to optimize.
__group_534__gradients (optional) - T3: gradients computed in this iteration.
__group_534__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_534__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_534__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_535__weights (optional) - T2: weights to optimize.
__group_535__gradients (optional) - T3: gradients computed in this iteration.
__group_535__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_535__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_535__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_536__weights (optional) - T2: weights to optimize.
__group_536__gradients (optional) - T3: gradients computed in this iteration.
__group_536__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_536__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_536__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_537__weights (optional) - T2: weights to optimize.
__group_537__gradients (optional) - T3: gradients computed in this iteration.
__group_537__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_537__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_537__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_538__weights (optional) - T2: weights to optimize.
__group_538__gradients (optional) - T3: gradients computed in this iteration.
__group_538__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_538__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_538__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_539__weights (optional) - T2: weights to optimize.
__group_539__gradients (optional) - T3: gradients computed in this iteration.
__group_539__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_539__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_539__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_540__weights (optional) - T2: weights to optimize.
__group_540__gradients (optional) - T3: gradients computed in this iteration.
__group_540__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_540__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_540__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_541__weights (optional) - T2: weights to optimize.
__group_541__gradients (optional) - T3: gradients computed in this iteration.
__group_541__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_541__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_541__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_542__weights (optional) - T2: weights to optimize.
__group_542__gradients (optional) - T3: gradients computed in this iteration.
__group_542__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_542__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_542__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_543__weights (optional) - T2: weights to optimize.
__group_543__gradients (optional) - T3: gradients computed in this iteration.
__group_543__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_543__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_543__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_544__weights (optional) - T2: weights to optimize.
__group_544__gradients (optional) - T3: gradients computed in this iteration.
__group_544__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_544__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_544__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_545__weights (optional) - T2: weights to optimize.
__group_545__gradients (optional) - T3: gradients computed in this iteration.
__group_545__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_545__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_545__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_546__weights (optional) - T2: weights to optimize.
__group_546__gradients (optional) - T3: gradients computed in this iteration.
__group_546__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_546__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_546__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_547__weights (optional) - T2: weights to optimize.
__group_547__gradients (optional) - T3: gradients computed in this iteration.
__group_547__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_547__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_547__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_548__weights (optional) - T2: weights to optimize.
__group_548__gradients (optional) - T3: gradients computed in this iteration.
__group_548__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_548__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_548__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_549__weights (optional) - T2: weights to optimize.
__group_549__gradients (optional) - T3: gradients computed in this iteration.
__group_549__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_549__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_549__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_550__weights (optional) - T2: weights to optimize.
__group_550__gradients (optional) - T3: gradients computed in this iteration.
__group_550__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_550__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_550__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_551__weights (optional) - T2: weights to optimize.
__group_551__gradients (optional) - T3: gradients computed in this iteration.
__group_551__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_551__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_551__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_552__weights (optional) - T2: weights to optimize.
__group_552__gradients (optional) - T3: gradients computed in this iteration.
__group_552__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_552__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_552__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_553__weights (optional) - T2: weights to optimize.
__group_553__gradients (optional) - T3: gradients computed in this iteration.
__group_553__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_553__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_553__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_554__weights (optional) - T2: weights to optimize.
__group_554__gradients (optional) - T3: gradients computed in this iteration.
__group_554__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_554__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_554__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_555__weights (optional) - T2: weights to optimize.
__group_555__gradients (optional) - T3: gradients computed in this iteration.
__group_555__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_555__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_555__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_556__weights (optional) - T2: weights to optimize.
__group_556__gradients (optional) - T3: gradients computed in this iteration.
__group_556__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_556__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_556__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_557__weights (optional) - T2: weights to optimize.
__group_557__gradients (optional) - T3: gradients computed in this iteration.
__group_557__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_557__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_557__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_558__weights (optional) - T2: weights to optimize.
__group_558__gradients (optional) - T3: gradients computed in this iteration.
__group_558__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_558__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_558__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_559__weights (optional) - T2: weights to optimize.
__group_559__gradients (optional) - T3: gradients computed in this iteration.
__group_559__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_559__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_559__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_560__weights (optional) - T2: weights to optimize.
__group_560__gradients (optional) - T3: gradients computed in this iteration.
__group_560__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_560__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_560__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_561__weights (optional) - T2: weights to optimize.
__group_561__gradients (optional) - T3: gradients computed in this iteration.
__group_561__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_561__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_561__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_562__weights (optional) - T2: weights to optimize.
__group_562__gradients (optional) - T3: gradients computed in this iteration.
__group_562__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_562__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_562__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_563__weights (optional) - T2: weights to optimize.
__group_563__gradients (optional) - T3: gradients computed in this iteration.
__group_563__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_563__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_563__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_564__weights (optional) - T2: weights to optimize.
__group_564__gradients (optional) - T3: gradients computed in this iteration.
__group_564__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_564__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_564__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_565__weights (optional) - T2: weights to optimize.
__group_565__gradients (optional) - T3: gradients computed in this iteration.
__group_565__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_565__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_565__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_566__weights (optional) - T2: weights to optimize.
__group_566__gradients (optional) - T3: gradients computed in this iteration.
__group_566__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_566__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_566__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_567__weights (optional) - T2: weights to optimize.
__group_567__gradients (optional) - T3: gradients computed in this iteration.
__group_567__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_567__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_567__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_568__weights (optional) - T2: weights to optimize.
__group_568__gradients (optional) - T3: gradients computed in this iteration.
__group_568__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_568__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_568__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_569__weights (optional) - T2: weights to optimize.
__group_569__gradients (optional) - T3: gradients computed in this iteration.
__group_569__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_569__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_569__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_570__weights (optional) - T2: weights to optimize.
__group_570__gradients (optional) - T3: gradients computed in this iteration.
__group_570__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_570__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_570__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_571__weights (optional) - T2: weights to optimize.
__group_571__gradients (optional) - T3: gradients computed in this iteration.
__group_571__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_571__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_571__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_572__weights (optional) - T2: weights to optimize.
__group_572__gradients (optional) - T3: gradients computed in this iteration.
__group_572__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_572__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_572__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_573__weights (optional) - T2: weights to optimize.
__group_573__gradients (optional) - T3: gradients computed in this iteration.
__group_573__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_573__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_573__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_574__weights (optional) - T2: weights to optimize.
__group_574__gradients (optional) - T3: gradients computed in this iteration.
__group_574__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_574__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_574__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_575__weights (optional) - T2: weights to optimize.
__group_575__gradients (optional) - T3: gradients computed in this iteration.
__group_575__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_575__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_575__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_576__weights (optional) - T2: weights to optimize.
__group_576__gradients (optional) - T3: gradients computed in this iteration.
__group_576__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_576__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_576__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_577__weights (optional) - T2: weights to optimize.
__group_577__gradients (optional) - T3: gradients computed in this iteration.
__group_577__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_577__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_577__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_578__weights (optional) - T2: weights to optimize.
__group_578__gradients (optional) - T3: gradients computed in this iteration.
__group_578__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_578__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_578__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_579__weights (optional) - T2: weights to optimize.
__group_579__gradients (optional) - T3: gradients computed in this iteration.
__group_579__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_579__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_579__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_580__weights (optional) - T2: weights to optimize.
__group_580__gradients (optional) - T3: gradients computed in this iteration.
__group_580__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_580__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_580__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_581__weights (optional) - T2: weights to optimize.
__group_581__gradients (optional) - T3: gradients computed in this iteration.
__group_581__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_581__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_581__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_582__weights (optional) - T2: weights to optimize.
__group_582__gradients (optional) - T3: gradients computed in this iteration.
__group_582__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_582__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_582__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_583__weights (optional) - T2: weights to optimize.
__group_583__gradients (optional) - T3: gradients computed in this iteration.
__group_583__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_583__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_583__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_584__weights (optional) - T2: weights to optimize.
__group_584__gradients (optional) - T3: gradients computed in this iteration.
__group_584__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_584__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_584__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_585__weights (optional) - T2: weights to optimize.
__group_585__gradients (optional) - T3: gradients computed in this iteration.
__group_585__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_585__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_585__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_586__weights (optional) - T2: weights to optimize.
__group_586__gradients (optional) - T3: gradients computed in this iteration.
__group_586__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_586__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_586__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_587__weights (optional) - T2: weights to optimize.
__group_587__gradients (optional) - T3: gradients computed in this iteration.
__group_587__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_587__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_587__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_588__weights (optional) - T2: weights to optimize.
__group_588__gradients (optional) - T3: gradients computed in this iteration.
__group_588__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_588__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_588__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_589__weights (optional) - T2: weights to optimize.
__group_589__gradients (optional) - T3: gradients computed in this iteration.
__group_589__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_589__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_589__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_590__weights (optional) - T2: weights to optimize.
__group_590__gradients (optional) - T3: gradients computed in this iteration.
__group_590__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_590__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_590__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_591__weights (optional) - T2: weights to optimize.
__group_591__gradients (optional) - T3: gradients computed in this iteration.
__group_591__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_591__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_591__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_592__weights (optional) - T2: weights to optimize.
__group_592__gradients (optional) - T3: gradients computed in this iteration.
__group_592__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_592__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_592__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_593__weights (optional) - T2: weights to optimize.
__group_593__gradients (optional) - T3: gradients computed in this iteration.
__group_593__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_593__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_593__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_594__weights (optional) - T2: weights to optimize.
__group_594__gradients (optional) - T3: gradients computed in this iteration.
__group_594__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_594__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_594__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_595__weights (optional) - T2: weights to optimize.
__group_595__gradients (optional) - T3: gradients computed in this iteration.
__group_595__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_595__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_595__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_596__weights (optional) - T2: weights to optimize.
__group_596__gradients (optional) - T3: gradients computed in this iteration.
__group_596__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_596__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_596__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_597__weights (optional) - T2: weights to optimize.
__group_597__gradients (optional) - T3: gradients computed in this iteration.
__group_597__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_597__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_597__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_598__weights (optional) - T2: weights to optimize.
__group_598__gradients (optional) - T3: gradients computed in this iteration.
__group_598__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_598__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_598__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_599__weights (optional) - T2: weights to optimize.
__group_599__gradients (optional) - T3: gradients computed in this iteration.
__group_599__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_599__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_599__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_600__weights (optional) - T2: weights to optimize.
__group_600__gradients (optional) - T3: gradients computed in this iteration.
__group_600__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_600__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_600__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_601__weights (optional) - T2: weights to optimize.
__group_601__gradients (optional) - T3: gradients computed in this iteration.
__group_601__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_601__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_601__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_602__weights (optional) - T2: weights to optimize.
__group_602__gradients (optional) - T3: gradients computed in this iteration.
__group_602__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_602__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_602__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_603__weights (optional) - T2: weights to optimize.
__group_603__gradients (optional) - T3: gradients computed in this iteration.
__group_603__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_603__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_603__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_604__weights (optional) - T2: weights to optimize.
__group_604__gradients (optional) - T3: gradients computed in this iteration.
__group_604__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_604__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_604__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_605__weights (optional) - T2: weights to optimize.
__group_605__gradients (optional) - T3: gradients computed in this iteration.
__group_605__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_605__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_605__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_606__weights (optional) - T2: weights to optimize.
__group_606__gradients (optional) - T3: gradients computed in this iteration.
__group_606__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_606__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_606__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_607__weights (optional) - T2: weights to optimize.
__group_607__gradients (optional) - T3: gradients computed in this iteration.
__group_607__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_607__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_607__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_608__weights (optional) - T2: weights to optimize.
__group_608__gradients (optional) - T3: gradients computed in this iteration.
__group_608__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_608__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_608__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_609__weights (optional) - T2: weights to optimize.
__group_609__gradients (optional) - T3: gradients computed in this iteration.
__group_609__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_609__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_609__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_610__weights (optional) - T2: weights to optimize.
__group_610__gradients (optional) - T3: gradients computed in this iteration.
__group_610__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_610__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_610__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_611__weights (optional) - T2: weights to optimize.
__group_611__gradients (optional) - T3: gradients computed in this iteration.
__group_611__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_611__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_611__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_612__weights (optional) - T2: weights to optimize.
__group_612__gradients (optional) - T3: gradients computed in this iteration.
__group_612__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_612__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_612__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_613__weights (optional) - T2: weights to optimize.
__group_613__gradients (optional) - T3: gradients computed in this iteration.
__group_613__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_613__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_613__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_614__weights (optional) - T2: weights to optimize.
__group_614__gradients (optional) - T3: gradients computed in this iteration.
__group_614__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_614__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_614__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_615__weights (optional) - T2: weights to optimize.
__group_615__gradients (optional) - T3: gradients computed in this iteration.
__group_615__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_615__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_615__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_616__weights (optional) - T2: weights to optimize.
__group_616__gradients (optional) - T3: gradients computed in this iteration.
__group_616__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_616__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_616__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_617__weights (optional) - T2: weights to optimize.
__group_617__gradients (optional) - T3: gradients computed in this iteration.
__group_617__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_617__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_617__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_618__weights (optional) - T2: weights to optimize.
__group_618__gradients (optional) - T3: gradients computed in this iteration.
__group_618__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_618__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_618__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_619__weights (optional) - T2: weights to optimize.
__group_619__gradients (optional) - T3: gradients computed in this iteration.
__group_619__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_619__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_619__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_620__weights (optional) - T2: weights to optimize.
__group_620__gradients (optional) - T3: gradients computed in this iteration.
__group_620__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_620__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_620__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_621__weights (optional) - T2: weights to optimize.
__group_621__gradients (optional) - T3: gradients computed in this iteration.
__group_621__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_621__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_621__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_622__weights (optional) - T2: weights to optimize.
__group_622__gradients (optional) - T3: gradients computed in this iteration.
__group_622__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_622__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_622__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_623__weights (optional) - T2: weights to optimize.
__group_623__gradients (optional) - T3: gradients computed in this iteration.
__group_623__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_623__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_623__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_624__weights (optional) - T2: weights to optimize.
__group_624__gradients (optional) - T3: gradients computed in this iteration.
__group_624__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_624__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_624__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_625__weights (optional) - T2: weights to optimize.
__group_625__gradients (optional) - T3: gradients computed in this iteration.
__group_625__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_625__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_625__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_626__weights (optional) - T2: weights to optimize.
__group_626__gradients (optional) - T3: gradients computed in this iteration.
__group_626__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_626__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_626__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_627__weights (optional) - T2: weights to optimize.
__group_627__gradients (optional) - T3: gradients computed in this iteration.
__group_627__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_627__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_627__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_628__weights (optional) - T2: weights to optimize.
__group_628__gradients (optional) - T3: gradients computed in this iteration.
__group_628__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_628__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_628__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_629__weights (optional) - T2: weights to optimize.
__group_629__gradients (optional) - T3: gradients computed in this iteration.
__group_629__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_629__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_629__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_630__weights (optional) - T2: weights to optimize.
__group_630__gradients (optional) - T3: gradients computed in this iteration.
__group_630__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_630__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_630__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_631__weights (optional) - T2: weights to optimize.
__group_631__gradients (optional) - T3: gradients computed in this iteration.
__group_631__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_631__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_631__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_632__weights (optional) - T2: weights to optimize.
__group_632__gradients (optional) - T3: gradients computed in this iteration.
__group_632__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_632__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_632__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_633__weights (optional) - T2: weights to optimize.
__group_633__gradients (optional) - T3: gradients computed in this iteration.
__group_633__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_633__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_633__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_634__weights (optional) - T2: weights to optimize.
__group_634__gradients (optional) - T3: gradients computed in this iteration.
__group_634__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_634__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_634__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_635__weights (optional) - T2: weights to optimize.
__group_635__gradients (optional) - T3: gradients computed in this iteration.
__group_635__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_635__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_635__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_636__weights (optional) - T2: weights to optimize.
__group_636__gradients (optional) - T3: gradients computed in this iteration.
__group_636__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_636__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_636__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_637__weights (optional) - T2: weights to optimize.
__group_637__gradients (optional) - T3: gradients computed in this iteration.
__group_637__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_637__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_637__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_638__weights (optional) - T2: weights to optimize.
__group_638__gradients (optional) - T3: gradients computed in this iteration.
__group_638__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_638__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_638__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_639__weights (optional) - T2: weights to optimize.
__group_639__gradients (optional) - T3: gradients computed in this iteration.
__group_639__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_639__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_639__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_640__weights (optional) - T2: weights to optimize.
__group_640__gradients (optional) - T3: gradients computed in this iteration.
__group_640__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_640__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_640__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_641__weights (optional) - T2: weights to optimize.
__group_641__gradients (optional) - T3: gradients computed in this iteration.
__group_641__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_641__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_641__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_642__weights (optional) - T2: weights to optimize.
__group_642__gradients (optional) - T3: gradients computed in this iteration.
__group_642__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_642__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_642__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_643__weights (optional) - T2: weights to optimize.
__group_643__gradients (optional) - T3: gradients computed in this iteration.
__group_643__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_643__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_643__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_644__weights (optional) - T2: weights to optimize.
__group_644__gradients (optional) - T3: gradients computed in this iteration.
__group_644__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_644__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_644__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_645__weights (optional) - T2: weights to optimize.
__group_645__gradients (optional) - T3: gradients computed in this iteration.
__group_645__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_645__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_645__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_646__weights (optional) - T2: weights to optimize.
__group_646__gradients (optional) - T3: gradients computed in this iteration.
__group_646__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_646__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_646__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_647__weights (optional) - T2: weights to optimize.
__group_647__gradients (optional) - T3: gradients computed in this iteration.
__group_647__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_647__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_647__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_648__weights (optional) - T2: weights to optimize.
__group_648__gradients (optional) - T3: gradients computed in this iteration.
__group_648__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_648__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_648__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_649__weights (optional) - T2: weights to optimize.
__group_649__gradients (optional) - T3: gradients computed in this iteration.
__group_649__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_649__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_649__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_650__weights (optional) - T2: weights to optimize.
__group_650__gradients (optional) - T3: gradients computed in this iteration.
__group_650__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_650__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_650__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_651__weights (optional) - T2: weights to optimize.
__group_651__gradients (optional) - T3: gradients computed in this iteration.
__group_651__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_651__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_651__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_652__weights (optional) - T2: weights to optimize.
__group_652__gradients (optional) - T3: gradients computed in this iteration.
__group_652__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_652__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_652__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_653__weights (optional) - T2: weights to optimize.
__group_653__gradients (optional) - T3: gradients computed in this iteration.
__group_653__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_653__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_653__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_654__weights (optional) - T2: weights to optimize.
__group_654__gradients (optional) - T3: gradients computed in this iteration.
__group_654__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_654__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_654__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_655__weights (optional) - T2: weights to optimize.
__group_655__gradients (optional) - T3: gradients computed in this iteration.
__group_655__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_655__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_655__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_656__weights (optional) - T2: weights to optimize.
__group_656__gradients (optional) - T3: gradients computed in this iteration.
__group_656__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_656__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_656__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_657__weights (optional) - T2: weights to optimize.
__group_657__gradients (optional) - T3: gradients computed in this iteration.
__group_657__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_657__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_657__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_658__weights (optional) - T2: weights to optimize.
__group_658__gradients (optional) - T3: gradients computed in this iteration.
__group_658__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_658__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_658__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_659__weights (optional) - T2: weights to optimize.
__group_659__gradients (optional) - T3: gradients computed in this iteration.
__group_659__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_659__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_659__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_660__weights (optional) - T2: weights to optimize.
__group_660__gradients (optional) - T3: gradients computed in this iteration.
__group_660__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_660__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_660__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_661__weights (optional) - T2: weights to optimize.
__group_661__gradients (optional) - T3: gradients computed in this iteration.
__group_661__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_661__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_661__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_662__weights (optional) - T2: weights to optimize.
__group_662__gradients (optional) - T3: gradients computed in this iteration.
__group_662__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_662__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_662__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_663__weights (optional) - T2: weights to optimize.
__group_663__gradients (optional) - T3: gradients computed in this iteration.
__group_663__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_663__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_663__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_664__weights (optional) - T2: weights to optimize.
__group_664__gradients (optional) - T3: gradients computed in this iteration.
__group_664__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_664__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_664__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_665__weights (optional) - T2: weights to optimize.
__group_665__gradients (optional) - T3: gradients computed in this iteration.
__group_665__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_665__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_665__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_666__weights (optional) - T2: weights to optimize.
__group_666__gradients (optional) - T3: gradients computed in this iteration.
__group_666__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_666__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_666__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_667__weights (optional) - T2: weights to optimize.
__group_667__gradients (optional) - T3: gradients computed in this iteration.
__group_667__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_667__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_667__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_668__weights (optional) - T2: weights to optimize.
__group_668__gradients (optional) - T3: gradients computed in this iteration.
__group_668__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_668__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_668__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_669__weights (optional) - T2: weights to optimize.
__group_669__gradients (optional) - T3: gradients computed in this iteration.
__group_669__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_669__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_669__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_670__weights (optional) - T2: weights to optimize.
__group_670__gradients (optional) - T3: gradients computed in this iteration.
__group_670__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_670__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_670__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_671__weights (optional) - T2: weights to optimize.
__group_671__gradients (optional) - T3: gradients computed in this iteration.
__group_671__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_671__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_671__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_672__weights (optional) - T2: weights to optimize.
__group_672__gradients (optional) - T3: gradients computed in this iteration.
__group_672__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_672__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_672__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_673__weights (optional) - T2: weights to optimize.
__group_673__gradients (optional) - T3: gradients computed in this iteration.
__group_673__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_673__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_673__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_674__weights (optional) - T2: weights to optimize.
__group_674__gradients (optional) - T3: gradients computed in this iteration.
__group_674__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_674__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_674__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_675__weights (optional) - T2: weights to optimize.
__group_675__gradients (optional) - T3: gradients computed in this iteration.
__group_675__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_675__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_675__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_676__weights (optional) - T2: weights to optimize.
__group_676__gradients (optional) - T3: gradients computed in this iteration.
__group_676__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_676__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_676__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_677__weights (optional) - T2: weights to optimize.
__group_677__gradients (optional) - T3: gradients computed in this iteration.
__group_677__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_677__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_677__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_678__weights (optional) - T2: weights to optimize.
__group_678__gradients (optional) - T3: gradients computed in this iteration.
__group_678__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_678__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_678__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_679__weights (optional) - T2: weights to optimize.
__group_679__gradients (optional) - T3: gradients computed in this iteration.
__group_679__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_679__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_679__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_680__weights (optional) - T2: weights to optimize.
__group_680__gradients (optional) - T3: gradients computed in this iteration.
__group_680__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_680__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_680__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_681__weights (optional) - T2: weights to optimize.
__group_681__gradients (optional) - T3: gradients computed in this iteration.
__group_681__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_681__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_681__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_682__weights (optional) - T2: weights to optimize.
__group_682__gradients (optional) - T3: gradients computed in this iteration.
__group_682__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_682__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_682__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_683__weights (optional) - T2: weights to optimize.
__group_683__gradients (optional) - T3: gradients computed in this iteration.
__group_683__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_683__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_683__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_684__weights (optional) - T2: weights to optimize.
__group_684__gradients (optional) - T3: gradients computed in this iteration.
__group_684__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_684__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_684__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_685__weights (optional) - T2: weights to optimize.
__group_685__gradients (optional) - T3: gradients computed in this iteration.
__group_685__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_685__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_685__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_686__weights (optional) - T2: weights to optimize.
__group_686__gradients (optional) - T3: gradients computed in this iteration.
__group_686__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_686__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_686__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_687__weights (optional) - T2: weights to optimize.
__group_687__gradients (optional) - T3: gradients computed in this iteration.
__group_687__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_687__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_687__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_688__weights (optional) - T2: weights to optimize.
__group_688__gradients (optional) - T3: gradients computed in this iteration.
__group_688__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_688__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_688__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_689__weights (optional) - T2: weights to optimize.
__group_689__gradients (optional) - T3: gradients computed in this iteration.
__group_689__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_689__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_689__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_690__weights (optional) - T2: weights to optimize.
__group_690__gradients (optional) - T3: gradients computed in this iteration.
__group_690__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_690__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_690__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_691__weights (optional) - T2: weights to optimize.
__group_691__gradients (optional) - T3: gradients computed in this iteration.
__group_691__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_691__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_691__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_692__weights (optional) - T2: weights to optimize.
__group_692__gradients (optional) - T3: gradients computed in this iteration.
__group_692__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_692__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_692__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_693__weights (optional) - T2: weights to optimize.
__group_693__gradients (optional) - T3: gradients computed in this iteration.
__group_693__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_693__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_693__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_694__weights (optional) - T2: weights to optimize.
__group_694__gradients (optional) - T3: gradients computed in this iteration.
__group_694__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_694__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_694__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_695__weights (optional) - T2: weights to optimize.
__group_695__gradients (optional) - T3: gradients computed in this iteration.
__group_695__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_695__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_695__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_696__weights (optional) - T2: weights to optimize.
__group_696__gradients (optional) - T3: gradients computed in this iteration.
__group_696__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_696__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_696__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_697__weights (optional) - T2: weights to optimize.
__group_697__gradients (optional) - T3: gradients computed in this iteration.
__group_697__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_697__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_697__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_698__weights (optional) - T2: weights to optimize.
__group_698__gradients (optional) - T3: gradients computed in this iteration.
__group_698__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_698__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_698__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_699__weights (optional) - T2: weights to optimize.
__group_699__gradients (optional) - T3: gradients computed in this iteration.
__group_699__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_699__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_699__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_700__weights (optional) - T2: weights to optimize.
__group_700__gradients (optional) - T3: gradients computed in this iteration.
__group_700__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_700__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_700__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_701__weights (optional) - T2: weights to optimize.
__group_701__gradients (optional) - T3: gradients computed in this iteration.
__group_701__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_701__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_701__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_702__weights (optional) - T2: weights to optimize.
__group_702__gradients (optional) - T3: gradients computed in this iteration.
__group_702__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_702__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_702__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_703__weights (optional) - T2: weights to optimize.
__group_703__gradients (optional) - T3: gradients computed in this iteration.
__group_703__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_703__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_703__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_704__weights (optional) - T2: weights to optimize.
__group_704__gradients (optional) - T3: gradients computed in this iteration.
__group_704__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_704__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_704__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_705__weights (optional) - T2: weights to optimize.
__group_705__gradients (optional) - T3: gradients computed in this iteration.
__group_705__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_705__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_705__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_706__weights (optional) - T2: weights to optimize.
__group_706__gradients (optional) - T3: gradients computed in this iteration.
__group_706__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_706__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_706__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_707__weights (optional) - T2: weights to optimize.
__group_707__gradients (optional) - T3: gradients computed in this iteration.
__group_707__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_707__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_707__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_708__weights (optional) - T2: weights to optimize.
__group_708__gradients (optional) - T3: gradients computed in this iteration.
__group_708__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_708__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_708__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_709__weights (optional) - T2: weights to optimize.
__group_709__gradients (optional) - T3: gradients computed in this iteration.
__group_709__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_709__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_709__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_710__weights (optional) - T2: weights to optimize.
__group_710__gradients (optional) - T3: gradients computed in this iteration.
__group_710__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_710__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_710__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_711__weights (optional) - T2: weights to optimize.
__group_711__gradients (optional) - T3: gradients computed in this iteration.
__group_711__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_711__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_711__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_712__weights (optional) - T2: weights to optimize.
__group_712__gradients (optional) - T3: gradients computed in this iteration.
__group_712__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_712__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_712__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_713__weights (optional) - T2: weights to optimize.
__group_713__gradients (optional) - T3: gradients computed in this iteration.
__group_713__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_713__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_713__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_714__weights (optional) - T2: weights to optimize.
__group_714__gradients (optional) - T3: gradients computed in this iteration.
__group_714__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_714__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_714__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_715__weights (optional) - T2: weights to optimize.
__group_715__gradients (optional) - T3: gradients computed in this iteration.
__group_715__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_715__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_715__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_716__weights (optional) - T2: weights to optimize.
__group_716__gradients (optional) - T3: gradients computed in this iteration.
__group_716__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_716__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_716__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_717__weights (optional) - T2: weights to optimize.
__group_717__gradients (optional) - T3: gradients computed in this iteration.
__group_717__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_717__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_717__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_718__weights (optional) - T2: weights to optimize.
__group_718__gradients (optional) - T3: gradients computed in this iteration.
__group_718__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_718__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_718__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_719__weights (optional) - T2: weights to optimize.
__group_719__gradients (optional) - T3: gradients computed in this iteration.
__group_719__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_719__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_719__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_720__weights (optional) - T2: weights to optimize.
__group_720__gradients (optional) - T3: gradients computed in this iteration.
__group_720__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_720__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_720__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_721__weights (optional) - T2: weights to optimize.
__group_721__gradients (optional) - T3: gradients computed in this iteration.
__group_721__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_721__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_721__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_722__weights (optional) - T2: weights to optimize.
__group_722__gradients (optional) - T3: gradients computed in this iteration.
__group_722__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_722__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_722__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_723__weights (optional) - T2: weights to optimize.
__group_723__gradients (optional) - T3: gradients computed in this iteration.
__group_723__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_723__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_723__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_724__weights (optional) - T2: weights to optimize.
__group_724__gradients (optional) - T3: gradients computed in this iteration.
__group_724__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_724__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_724__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_725__weights (optional) - T2: weights to optimize.
__group_725__gradients (optional) - T3: gradients computed in this iteration.
__group_725__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_725__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_725__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_726__weights (optional) - T2: weights to optimize.
__group_726__gradients (optional) - T3: gradients computed in this iteration.
__group_726__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_726__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_726__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_727__weights (optional) - T2: weights to optimize.
__group_727__gradients (optional) - T3: gradients computed in this iteration.
__group_727__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_727__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_727__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_728__weights (optional) - T2: weights to optimize.
__group_728__gradients (optional) - T3: gradients computed in this iteration.
__group_728__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_728__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_728__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_729__weights (optional) - T2: weights to optimize.
__group_729__gradients (optional) - T3: gradients computed in this iteration.
__group_729__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_729__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_729__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_730__weights (optional) - T2: weights to optimize.
__group_730__gradients (optional) - T3: gradients computed in this iteration.
__group_730__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_730__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_730__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_731__weights (optional) - T2: weights to optimize.
__group_731__gradients (optional) - T3: gradients computed in this iteration.
__group_731__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_731__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_731__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_732__weights (optional) - T2: weights to optimize.
__group_732__gradients (optional) - T3: gradients computed in this iteration.
__group_732__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_732__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_732__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_733__weights (optional) - T2: weights to optimize.
__group_733__gradients (optional) - T3: gradients computed in this iteration.
__group_733__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_733__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_733__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_734__weights (optional) - T2: weights to optimize.
__group_734__gradients (optional) - T3: gradients computed in this iteration.
__group_734__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_734__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_734__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_735__weights (optional) - T2: weights to optimize.
__group_735__gradients (optional) - T3: gradients computed in this iteration.
__group_735__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_735__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_735__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_736__weights (optional) - T2: weights to optimize.
__group_736__gradients (optional) - T3: gradients computed in this iteration.
__group_736__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_736__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_736__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_737__weights (optional) - T2: weights to optimize.
__group_737__gradients (optional) - T3: gradients computed in this iteration.
__group_737__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_737__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_737__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_738__weights (optional) - T2: weights to optimize.
__group_738__gradients (optional) - T3: gradients computed in this iteration.
__group_738__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_738__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_738__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_739__weights (optional) - T2: weights to optimize.
__group_739__gradients (optional) - T3: gradients computed in this iteration.
__group_739__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_739__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_739__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_740__weights (optional) - T2: weights to optimize.
__group_740__gradients (optional) - T3: gradients computed in this iteration.
__group_740__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_740__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_740__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_741__weights (optional) - T2: weights to optimize.
__group_741__gradients (optional) - T3: gradients computed in this iteration.
__group_741__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_741__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_741__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_742__weights (optional) - T2: weights to optimize.
__group_742__gradients (optional) - T3: gradients computed in this iteration.
__group_742__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_742__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_742__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_743__weights (optional) - T2: weights to optimize.
__group_743__gradients (optional) - T3: gradients computed in this iteration.
__group_743__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_743__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_743__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_744__weights (optional) - T2: weights to optimize.
__group_744__gradients (optional) - T3: gradients computed in this iteration.
__group_744__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_744__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_744__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_745__weights (optional) - T2: weights to optimize.
__group_745__gradients (optional) - T3: gradients computed in this iteration.
__group_745__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_745__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_745__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_746__weights (optional) - T2: weights to optimize.
__group_746__gradients (optional) - T3: gradients computed in this iteration.
__group_746__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_746__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_746__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_747__weights (optional) - T2: weights to optimize.
__group_747__gradients (optional) - T3: gradients computed in this iteration.
__group_747__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_747__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_747__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_748__weights (optional) - T2: weights to optimize.
__group_748__gradients (optional) - T3: gradients computed in this iteration.
__group_748__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_748__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_748__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_749__weights (optional) - T2: weights to optimize.
__group_749__gradients (optional) - T3: gradients computed in this iteration.
__group_749__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_749__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_749__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_750__weights (optional) - T2: weights to optimize.
__group_750__gradients (optional) - T3: gradients computed in this iteration.
__group_750__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_750__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_750__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_751__weights (optional) - T2: weights to optimize.
__group_751__gradients (optional) - T3: gradients computed in this iteration.
__group_751__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_751__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_751__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_752__weights (optional) - T2: weights to optimize.
__group_752__gradients (optional) - T3: gradients computed in this iteration.
__group_752__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_752__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_752__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_753__weights (optional) - T2: weights to optimize.
__group_753__gradients (optional) - T3: gradients computed in this iteration.
__group_753__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_753__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_753__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_754__weights (optional) - T2: weights to optimize.
__group_754__gradients (optional) - T3: gradients computed in this iteration.
__group_754__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_754__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_754__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_755__weights (optional) - T2: weights to optimize.
__group_755__gradients (optional) - T3: gradients computed in this iteration.
__group_755__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_755__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_755__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_756__weights (optional) - T2: weights to optimize.
__group_756__gradients (optional) - T3: gradients computed in this iteration.
__group_756__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_756__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_756__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_757__weights (optional) - T2: weights to optimize.
__group_757__gradients (optional) - T3: gradients computed in this iteration.
__group_757__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_757__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_757__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_758__weights (optional) - T2: weights to optimize.
__group_758__gradients (optional) - T3: gradients computed in this iteration.
__group_758__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_758__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_758__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_759__weights (optional) - T2: weights to optimize.
__group_759__gradients (optional) - T3: gradients computed in this iteration.
__group_759__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_759__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_759__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_760__weights (optional) - T2: weights to optimize.
__group_760__gradients (optional) - T3: gradients computed in this iteration.
__group_760__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_760__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_760__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_761__weights (optional) - T2: weights to optimize.
__group_761__gradients (optional) - T3: gradients computed in this iteration.
__group_761__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_761__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_761__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_762__weights (optional) - T2: weights to optimize.
__group_762__gradients (optional) - T3: gradients computed in this iteration.
__group_762__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_762__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_762__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_763__weights (optional) - T2: weights to optimize.
__group_763__gradients (optional) - T3: gradients computed in this iteration.
__group_763__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_763__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_763__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_764__weights (optional) - T2: weights to optimize.
__group_764__gradients (optional) - T3: gradients computed in this iteration.
__group_764__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_764__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_764__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_765__weights (optional) - T2: weights to optimize.
__group_765__gradients (optional) - T3: gradients computed in this iteration.
__group_765__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_765__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_765__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_766__weights (optional) - T2: weights to optimize.
__group_766__gradients (optional) - T3: gradients computed in this iteration.
__group_766__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_766__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_766__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_767__weights (optional) - T2: weights to optimize.
__group_767__gradients (optional) - T3: gradients computed in this iteration.
__group_767__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_767__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_767__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_768__weights (optional) - T2: weights to optimize.
__group_768__gradients (optional) - T3: gradients computed in this iteration.
__group_768__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_768__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_768__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_769__weights (optional) - T2: weights to optimize.
__group_769__gradients (optional) - T3: gradients computed in this iteration.
__group_769__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_769__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_769__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_770__weights (optional) - T2: weights to optimize.
__group_770__gradients (optional) - T3: gradients computed in this iteration.
__group_770__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_770__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_770__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_771__weights (optional) - T2: weights to optimize.
__group_771__gradients (optional) - T3: gradients computed in this iteration.
__group_771__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_771__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_771__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_772__weights (optional) - T2: weights to optimize.
__group_772__gradients (optional) - T3: gradients computed in this iteration.
__group_772__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_772__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_772__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_773__weights (optional) - T2: weights to optimize.
__group_773__gradients (optional) - T3: gradients computed in this iteration.
__group_773__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_773__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_773__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_774__weights (optional) - T2: weights to optimize.
__group_774__gradients (optional) - T3: gradients computed in this iteration.
__group_774__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_774__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_774__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_775__weights (optional) - T2: weights to optimize.
__group_775__gradients (optional) - T3: gradients computed in this iteration.
__group_775__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_775__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_775__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_776__weights (optional) - T2: weights to optimize.
__group_776__gradients (optional) - T3: gradients computed in this iteration.
__group_776__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_776__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_776__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_777__weights (optional) - T2: weights to optimize.
__group_777__gradients (optional) - T3: gradients computed in this iteration.
__group_777__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_777__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_777__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_778__weights (optional) - T2: weights to optimize.
__group_778__gradients (optional) - T3: gradients computed in this iteration.
__group_778__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_778__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_778__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_779__weights (optional) - T2: weights to optimize.
__group_779__gradients (optional) - T3: gradients computed in this iteration.
__group_779__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_779__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_779__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_780__weights (optional) - T2: weights to optimize.
__group_780__gradients (optional) - T3: gradients computed in this iteration.
__group_780__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_780__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_780__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_781__weights (optional) - T2: weights to optimize.
__group_781__gradients (optional) - T3: gradients computed in this iteration.
__group_781__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_781__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_781__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_782__weights (optional) - T2: weights to optimize.
__group_782__gradients (optional) - T3: gradients computed in this iteration.
__group_782__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_782__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_782__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_783__weights (optional) - T2: weights to optimize.
__group_783__gradients (optional) - T3: gradients computed in this iteration.
__group_783__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_783__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_783__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_784__weights (optional) - T2: weights to optimize.
__group_784__gradients (optional) - T3: gradients computed in this iteration.
__group_784__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_784__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_784__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_785__weights (optional) - T2: weights to optimize.
__group_785__gradients (optional) - T3: gradients computed in this iteration.
__group_785__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_785__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_785__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_786__weights (optional) - T2: weights to optimize.
__group_786__gradients (optional) - T3: gradients computed in this iteration.
__group_786__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_786__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_786__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_787__weights (optional) - T2: weights to optimize.
__group_787__gradients (optional) - T3: gradients computed in this iteration.
__group_787__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_787__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_787__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_788__weights (optional) - T2: weights to optimize.
__group_788__gradients (optional) - T3: gradients computed in this iteration.
__group_788__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_788__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_788__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_789__weights (optional) - T2: weights to optimize.
__group_789__gradients (optional) - T3: gradients computed in this iteration.
__group_789__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_789__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_789__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_790__weights (optional) - T2: weights to optimize.
__group_790__gradients (optional) - T3: gradients computed in this iteration.
__group_790__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_790__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_790__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_791__weights (optional) - T2: weights to optimize.
__group_791__gradients (optional) - T3: gradients computed in this iteration.
__group_791__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_791__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_791__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_792__weights (optional) - T2: weights to optimize.
__group_792__gradients (optional) - T3: gradients computed in this iteration.
__group_792__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_792__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_792__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_793__weights (optional) - T2: weights to optimize.
__group_793__gradients (optional) - T3: gradients computed in this iteration.
__group_793__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_793__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_793__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_794__weights (optional) - T2: weights to optimize.
__group_794__gradients (optional) - T3: gradients computed in this iteration.
__group_794__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_794__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_794__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_795__weights (optional) - T2: weights to optimize.
__group_795__gradients (optional) - T3: gradients computed in this iteration.
__group_795__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_795__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_795__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_796__weights (optional) - T2: weights to optimize.
__group_796__gradients (optional) - T3: gradients computed in this iteration.
__group_796__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_796__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_796__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_797__weights (optional) - T2: weights to optimize.
__group_797__gradients (optional) - T3: gradients computed in this iteration.
__group_797__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_797__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_797__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_798__weights (optional) - T2: weights to optimize.
__group_798__gradients (optional) - T3: gradients computed in this iteration.
__group_798__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_798__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_798__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_799__weights (optional) - T2: weights to optimize.
__group_799__gradients (optional) - T3: gradients computed in this iteration.
__group_799__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_799__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_799__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_800__weights (optional) - T2: weights to optimize.
__group_800__gradients (optional) - T3: gradients computed in this iteration.
__group_800__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_800__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_800__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_801__weights (optional) - T2: weights to optimize.
__group_801__gradients (optional) - T3: gradients computed in this iteration.
__group_801__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_801__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_801__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_802__weights (optional) - T2: weights to optimize.
__group_802__gradients (optional) - T3: gradients computed in this iteration.
__group_802__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_802__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_802__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_803__weights (optional) - T2: weights to optimize.
__group_803__gradients (optional) - T3: gradients computed in this iteration.
__group_803__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_803__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_803__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_804__weights (optional) - T2: weights to optimize.
__group_804__gradients (optional) - T3: gradients computed in this iteration.
__group_804__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_804__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_804__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_805__weights (optional) - T2: weights to optimize.
__group_805__gradients (optional) - T3: gradients computed in this iteration.
__group_805__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_805__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_805__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_806__weights (optional) - T2: weights to optimize.
__group_806__gradients (optional) - T3: gradients computed in this iteration.
__group_806__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_806__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_806__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_807__weights (optional) - T2: weights to optimize.
__group_807__gradients (optional) - T3: gradients computed in this iteration.
__group_807__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_807__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_807__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_808__weights (optional) - T2: weights to optimize.
__group_808__gradients (optional) - T3: gradients computed in this iteration.
__group_808__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_808__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_808__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_809__weights (optional) - T2: weights to optimize.
__group_809__gradients (optional) - T3: gradients computed in this iteration.
__group_809__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_809__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_809__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_810__weights (optional) - T2: weights to optimize.
__group_810__gradients (optional) - T3: gradients computed in this iteration.
__group_810__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_810__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_810__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_811__weights (optional) - T2: weights to optimize.
__group_811__gradients (optional) - T3: gradients computed in this iteration.
__group_811__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_811__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_811__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_812__weights (optional) - T2: weights to optimize.
__group_812__gradients (optional) - T3: gradients computed in this iteration.
__group_812__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_812__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_812__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_813__weights (optional) - T2: weights to optimize.
__group_813__gradients (optional) - T3: gradients computed in this iteration.
__group_813__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_813__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_813__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_814__weights (optional) - T2: weights to optimize.
__group_814__gradients (optional) - T3: gradients computed in this iteration.
__group_814__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_814__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_814__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_815__weights (optional) - T2: weights to optimize.
__group_815__gradients (optional) - T3: gradients computed in this iteration.
__group_815__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_815__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_815__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_816__weights (optional) - T2: weights to optimize.
__group_816__gradients (optional) - T3: gradients computed in this iteration.
__group_816__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_816__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_816__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_817__weights (optional) - T2: weights to optimize.
__group_817__gradients (optional) - T3: gradients computed in this iteration.
__group_817__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_817__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_817__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_818__weights (optional) - T2: weights to optimize.
__group_818__gradients (optional) - T3: gradients computed in this iteration.
__group_818__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_818__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_818__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_819__weights (optional) - T2: weights to optimize.
__group_819__gradients (optional) - T3: gradients computed in this iteration.
__group_819__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_819__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_819__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_820__weights (optional) - T2: weights to optimize.
__group_820__gradients (optional) - T3: gradients computed in this iteration.
__group_820__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_820__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_820__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_821__weights (optional) - T2: weights to optimize.
__group_821__gradients (optional) - T3: gradients computed in this iteration.
__group_821__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_821__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_821__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_822__weights (optional) - T2: weights to optimize.
__group_822__gradients (optional) - T3: gradients computed in this iteration.
__group_822__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_822__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_822__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_823__weights (optional) - T2: weights to optimize.
__group_823__gradients (optional) - T3: gradients computed in this iteration.
__group_823__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_823__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_823__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_824__weights (optional) - T2: weights to optimize.
__group_824__gradients (optional) - T3: gradients computed in this iteration.
__group_824__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_824__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_824__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_825__weights (optional) - T2: weights to optimize.
__group_825__gradients (optional) - T3: gradients computed in this iteration.
__group_825__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_825__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_825__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_826__weights (optional) - T2: weights to optimize.
__group_826__gradients (optional) - T3: gradients computed in this iteration.
__group_826__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_826__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_826__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_827__weights (optional) - T2: weights to optimize.
__group_827__gradients (optional) - T3: gradients computed in this iteration.
__group_827__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_827__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_827__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_828__weights (optional) - T2: weights to optimize.
__group_828__gradients (optional) - T3: gradients computed in this iteration.
__group_828__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_828__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_828__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_829__weights (optional) - T2: weights to optimize.
__group_829__gradients (optional) - T3: gradients computed in this iteration.
__group_829__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_829__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_829__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_830__weights (optional) - T2: weights to optimize.
__group_830__gradients (optional) - T3: gradients computed in this iteration.
__group_830__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_830__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_830__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_831__weights (optional) - T2: weights to optimize.
__group_831__gradients (optional) - T3: gradients computed in this iteration.
__group_831__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_831__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_831__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_832__weights (optional) - T2: weights to optimize.
__group_832__gradients (optional) - T3: gradients computed in this iteration.
__group_832__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_832__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_832__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_833__weights (optional) - T2: weights to optimize.
__group_833__gradients (optional) - T3: gradients computed in this iteration.
__group_833__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_833__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_833__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_834__weights (optional) - T2: weights to optimize.
__group_834__gradients (optional) - T3: gradients computed in this iteration.
__group_834__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_834__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_834__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_835__weights (optional) - T2: weights to optimize.
__group_835__gradients (optional) - T3: gradients computed in this iteration.
__group_835__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_835__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_835__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_836__weights (optional) - T2: weights to optimize.
__group_836__gradients (optional) - T3: gradients computed in this iteration.
__group_836__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_836__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_836__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_837__weights (optional) - T2: weights to optimize.
__group_837__gradients (optional) - T3: gradients computed in this iteration.
__group_837__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_837__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_837__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_838__weights (optional) - T2: weights to optimize.
__group_838__gradients (optional) - T3: gradients computed in this iteration.
__group_838__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_838__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_838__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_839__weights (optional) - T2: weights to optimize.
__group_839__gradients (optional) - T3: gradients computed in this iteration.
__group_839__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_839__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_839__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_840__weights (optional) - T2: weights to optimize.
__group_840__gradients (optional) - T3: gradients computed in this iteration.
__group_840__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_840__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_840__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_841__weights (optional) - T2: weights to optimize.
__group_841__gradients (optional) - T3: gradients computed in this iteration.
__group_841__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_841__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_841__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_842__weights (optional) - T2: weights to optimize.
__group_842__gradients (optional) - T3: gradients computed in this iteration.
__group_842__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_842__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_842__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_843__weights (optional) - T2: weights to optimize.
__group_843__gradients (optional) - T3: gradients computed in this iteration.
__group_843__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_843__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_843__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_844__weights (optional) - T2: weights to optimize.
__group_844__gradients (optional) - T3: gradients computed in this iteration.
__group_844__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_844__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_844__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_845__weights (optional) - T2: weights to optimize.
__group_845__gradients (optional) - T3: gradients computed in this iteration.
__group_845__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_845__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_845__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_846__weights (optional) - T2: weights to optimize.
__group_846__gradients (optional) - T3: gradients computed in this iteration.
__group_846__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_846__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_846__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_847__weights (optional) - T2: weights to optimize.
__group_847__gradients (optional) - T3: gradients computed in this iteration.
__group_847__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_847__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_847__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_848__weights (optional) - T2: weights to optimize.
__group_848__gradients (optional) - T3: gradients computed in this iteration.
__group_848__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_848__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_848__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_849__weights (optional) - T2: weights to optimize.
__group_849__gradients (optional) - T3: gradients computed in this iteration.
__group_849__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_849__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_849__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_850__weights (optional) - T2: weights to optimize.
__group_850__gradients (optional) - T3: gradients computed in this iteration.
__group_850__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_850__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_850__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_851__weights (optional) - T2: weights to optimize.
__group_851__gradients (optional) - T3: gradients computed in this iteration.
__group_851__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_851__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_851__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_852__weights (optional) - T2: weights to optimize.
__group_852__gradients (optional) - T3: gradients computed in this iteration.
__group_852__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_852__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_852__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_853__weights (optional) - T2: weights to optimize.
__group_853__gradients (optional) - T3: gradients computed in this iteration.
__group_853__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_853__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_853__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_854__weights (optional) - T2: weights to optimize.
__group_854__gradients (optional) - T3: gradients computed in this iteration.
__group_854__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_854__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_854__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_855__weights (optional) - T2: weights to optimize.
__group_855__gradients (optional) - T3: gradients computed in this iteration.
__group_855__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_855__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_855__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_856__weights (optional) - T2: weights to optimize.
__group_856__gradients (optional) - T3: gradients computed in this iteration.
__group_856__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_856__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_856__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_857__weights (optional) - T2: weights to optimize.
__group_857__gradients (optional) - T3: gradients computed in this iteration.
__group_857__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_857__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_857__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_858__weights (optional) - T2: weights to optimize.
__group_858__gradients (optional) - T3: gradients computed in this iteration.
__group_858__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_858__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_858__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_859__weights (optional) - T2: weights to optimize.
__group_859__gradients (optional) - T3: gradients computed in this iteration.
__group_859__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_859__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_859__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_860__weights (optional) - T2: weights to optimize.
__group_860__gradients (optional) - T3: gradients computed in this iteration.
__group_860__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_860__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_860__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_861__weights (optional) - T2: weights to optimize.
__group_861__gradients (optional) - T3: gradients computed in this iteration.
__group_861__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_861__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_861__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_862__weights (optional) - T2: weights to optimize.
__group_862__gradients (optional) - T3: gradients computed in this iteration.
__group_862__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_862__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_862__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_863__weights (optional) - T2: weights to optimize.
__group_863__gradients (optional) - T3: gradients computed in this iteration.
__group_863__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_863__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_863__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_864__weights (optional) - T2: weights to optimize.
__group_864__gradients (optional) - T3: gradients computed in this iteration.
__group_864__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_864__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_864__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_865__weights (optional) - T2: weights to optimize.
__group_865__gradients (optional) - T3: gradients computed in this iteration.
__group_865__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_865__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_865__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_866__weights (optional) - T2: weights to optimize.
__group_866__gradients (optional) - T3: gradients computed in this iteration.
__group_866__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_866__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_866__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_867__weights (optional) - T2: weights to optimize.
__group_867__gradients (optional) - T3: gradients computed in this iteration.
__group_867__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_867__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_867__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_868__weights (optional) - T2: weights to optimize.
__group_868__gradients (optional) - T3: gradients computed in this iteration.
__group_868__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_868__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_868__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_869__weights (optional) - T2: weights to optimize.
__group_869__gradients (optional) - T3: gradients computed in this iteration.
__group_869__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_869__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_869__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_870__weights (optional) - T2: weights to optimize.
__group_870__gradients (optional) - T3: gradients computed in this iteration.
__group_870__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_870__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_870__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_871__weights (optional) - T2: weights to optimize.
__group_871__gradients (optional) - T3: gradients computed in this iteration.
__group_871__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_871__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_871__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_872__weights (optional) - T2: weights to optimize.
__group_872__gradients (optional) - T3: gradients computed in this iteration.
__group_872__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_872__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_872__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_873__weights (optional) - T2: weights to optimize.
__group_873__gradients (optional) - T3: gradients computed in this iteration.
__group_873__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_873__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_873__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_874__weights (optional) - T2: weights to optimize.
__group_874__gradients (optional) - T3: gradients computed in this iteration.
__group_874__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_874__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_874__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_875__weights (optional) - T2: weights to optimize.
__group_875__gradients (optional) - T3: gradients computed in this iteration.
__group_875__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_875__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_875__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_876__weights (optional) - T2: weights to optimize.
__group_876__gradients (optional) - T3: gradients computed in this iteration.
__group_876__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_876__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_876__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_877__weights (optional) - T2: weights to optimize.
__group_877__gradients (optional) - T3: gradients computed in this iteration.
__group_877__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_877__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_877__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_878__weights (optional) - T2: weights to optimize.
__group_878__gradients (optional) - T3: gradients computed in this iteration.
__group_878__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_878__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_878__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_879__weights (optional) - T2: weights to optimize.
__group_879__gradients (optional) - T3: gradients computed in this iteration.
__group_879__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_879__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_879__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_880__weights (optional) - T2: weights to optimize.
__group_880__gradients (optional) - T3: gradients computed in this iteration.
__group_880__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_880__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_880__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_881__weights (optional) - T2: weights to optimize.
__group_881__gradients (optional) - T3: gradients computed in this iteration.
__group_881__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_881__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_881__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_882__weights (optional) - T2: weights to optimize.
__group_882__gradients (optional) - T3: gradients computed in this iteration.
__group_882__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_882__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_882__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_883__weights (optional) - T2: weights to optimize.
__group_883__gradients (optional) - T3: gradients computed in this iteration.
__group_883__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_883__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_883__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_884__weights (optional) - T2: weights to optimize.
__group_884__gradients (optional) - T3: gradients computed in this iteration.
__group_884__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_884__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_884__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_885__weights (optional) - T2: weights to optimize.
__group_885__gradients (optional) - T3: gradients computed in this iteration.
__group_885__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_885__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_885__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_886__weights (optional) - T2: weights to optimize.
__group_886__gradients (optional) - T3: gradients computed in this iteration.
__group_886__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_886__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_886__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_887__weights (optional) - T2: weights to optimize.
__group_887__gradients (optional) - T3: gradients computed in this iteration.
__group_887__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_887__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_887__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_888__weights (optional) - T2: weights to optimize.
__group_888__gradients (optional) - T3: gradients computed in this iteration.
__group_888__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_888__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_888__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_889__weights (optional) - T2: weights to optimize.
__group_889__gradients (optional) - T3: gradients computed in this iteration.
__group_889__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_889__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_889__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_890__weights (optional) - T2: weights to optimize.
__group_890__gradients (optional) - T3: gradients computed in this iteration.
__group_890__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_890__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_890__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_891__weights (optional) - T2: weights to optimize.
__group_891__gradients (optional) - T3: gradients computed in this iteration.
__group_891__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_891__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_891__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_892__weights (optional) - T2: weights to optimize.
__group_892__gradients (optional) - T3: gradients computed in this iteration.
__group_892__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_892__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_892__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_893__weights (optional) - T2: weights to optimize.
__group_893__gradients (optional) - T3: gradients computed in this iteration.
__group_893__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_893__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_893__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_894__weights (optional) - T2: weights to optimize.
__group_894__gradients (optional) - T3: gradients computed in this iteration.
__group_894__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_894__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_894__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_895__weights (optional) - T2: weights to optimize.
__group_895__gradients (optional) - T3: gradients computed in this iteration.
__group_895__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_895__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_895__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_896__weights (optional) - T2: weights to optimize.
__group_896__gradients (optional) - T3: gradients computed in this iteration.
__group_896__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_896__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_896__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_897__weights (optional) - T2: weights to optimize.
__group_897__gradients (optional) - T3: gradients computed in this iteration.
__group_897__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_897__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_897__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_898__weights (optional) - T2: weights to optimize.
__group_898__gradients (optional) - T3: gradients computed in this iteration.
__group_898__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_898__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_898__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_899__weights (optional) - T2: weights to optimize.
__group_899__gradients (optional) - T3: gradients computed in this iteration.
__group_899__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_899__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_899__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_900__weights (optional) - T2: weights to optimize.
__group_900__gradients (optional) - T3: gradients computed in this iteration.
__group_900__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_900__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_900__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_901__weights (optional) - T2: weights to optimize.
__group_901__gradients (optional) - T3: gradients computed in this iteration.
__group_901__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_901__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_901__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_902__weights (optional) - T2: weights to optimize.
__group_902__gradients (optional) - T3: gradients computed in this iteration.
__group_902__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_902__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_902__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_903__weights (optional) - T2: weights to optimize.
__group_903__gradients (optional) - T3: gradients computed in this iteration.
__group_903__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_903__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_903__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_904__weights (optional) - T2: weights to optimize.
__group_904__gradients (optional) - T3: gradients computed in this iteration.
__group_904__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_904__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_904__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_905__weights (optional) - T2: weights to optimize.
__group_905__gradients (optional) - T3: gradients computed in this iteration.
__group_905__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_905__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_905__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_906__weights (optional) - T2: weights to optimize.
__group_906__gradients (optional) - T3: gradients computed in this iteration.
__group_906__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_906__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_906__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_907__weights (optional) - T2: weights to optimize.
__group_907__gradients (optional) - T3: gradients computed in this iteration.
__group_907__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_907__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_907__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_908__weights (optional) - T2: weights to optimize.
__group_908__gradients (optional) - T3: gradients computed in this iteration.
__group_908__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_908__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_908__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_909__weights (optional) - T2: weights to optimize.
__group_909__gradients (optional) - T3: gradients computed in this iteration.
__group_909__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_909__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_909__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_910__weights (optional) - T2: weights to optimize.
__group_910__gradients (optional) - T3: gradients computed in this iteration.
__group_910__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_910__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_910__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_911__weights (optional) - T2: weights to optimize.
__group_911__gradients (optional) - T3: gradients computed in this iteration.
__group_911__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_911__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_911__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_912__weights (optional) - T2: weights to optimize.
__group_912__gradients (optional) - T3: gradients computed in this iteration.
__group_912__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_912__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_912__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_913__weights (optional) - T2: weights to optimize.
__group_913__gradients (optional) - T3: gradients computed in this iteration.
__group_913__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_913__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_913__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_914__weights (optional) - T2: weights to optimize.
__group_914__gradients (optional) - T3: gradients computed in this iteration.
__group_914__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_914__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_914__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_915__weights (optional) - T2: weights to optimize.
__group_915__gradients (optional) - T3: gradients computed in this iteration.
__group_915__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_915__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_915__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_916__weights (optional) - T2: weights to optimize.
__group_916__gradients (optional) - T3: gradients computed in this iteration.
__group_916__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_916__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_916__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_917__weights (optional) - T2: weights to optimize.
__group_917__gradients (optional) - T3: gradients computed in this iteration.
__group_917__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_917__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_917__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_918__weights (optional) - T2: weights to optimize.
__group_918__gradients (optional) - T3: gradients computed in this iteration.
__group_918__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_918__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_918__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_919__weights (optional) - T2: weights to optimize.
__group_919__gradients (optional) - T3: gradients computed in this iteration.
__group_919__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_919__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_919__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_920__weights (optional) - T2: weights to optimize.
__group_920__gradients (optional) - T3: gradients computed in this iteration.
__group_920__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_920__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_920__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_921__weights (optional) - T2: weights to optimize.
__group_921__gradients (optional) - T3: gradients computed in this iteration.
__group_921__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_921__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_921__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_922__weights (optional) - T2: weights to optimize.
__group_922__gradients (optional) - T3: gradients computed in this iteration.
__group_922__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_922__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_922__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_923__weights (optional) - T2: weights to optimize.
__group_923__gradients (optional) - T3: gradients computed in this iteration.
__group_923__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_923__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_923__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_924__weights (optional) - T2: weights to optimize.
__group_924__gradients (optional) - T3: gradients computed in this iteration.
__group_924__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_924__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_924__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_925__weights (optional) - T2: weights to optimize.
__group_925__gradients (optional) - T3: gradients computed in this iteration.
__group_925__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_925__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_925__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_926__weights (optional) - T2: weights to optimize.
__group_926__gradients (optional) - T3: gradients computed in this iteration.
__group_926__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_926__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_926__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_927__weights (optional) - T2: weights to optimize.
__group_927__gradients (optional) - T3: gradients computed in this iteration.
__group_927__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_927__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_927__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_928__weights (optional) - T2: weights to optimize.
__group_928__gradients (optional) - T3: gradients computed in this iteration.
__group_928__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_928__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_928__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_929__weights (optional) - T2: weights to optimize.
__group_929__gradients (optional) - T3: gradients computed in this iteration.
__group_929__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_929__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_929__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_930__weights (optional) - T2: weights to optimize.
__group_930__gradients (optional) - T3: gradients computed in this iteration.
__group_930__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_930__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_930__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_931__weights (optional) - T2: weights to optimize.
__group_931__gradients (optional) - T3: gradients computed in this iteration.
__group_931__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_931__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_931__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_932__weights (optional) - T2: weights to optimize.
__group_932__gradients (optional) - T3: gradients computed in this iteration.
__group_932__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_932__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_932__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_933__weights (optional) - T2: weights to optimize.
__group_933__gradients (optional) - T3: gradients computed in this iteration.
__group_933__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_933__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_933__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_934__weights (optional) - T2: weights to optimize.
__group_934__gradients (optional) - T3: gradients computed in this iteration.
__group_934__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_934__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_934__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_935__weights (optional) - T2: weights to optimize.
__group_935__gradients (optional) - T3: gradients computed in this iteration.
__group_935__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_935__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_935__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_936__weights (optional) - T2: weights to optimize.
__group_936__gradients (optional) - T3: gradients computed in this iteration.
__group_936__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_936__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_936__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_937__weights (optional) - T2: weights to optimize.
__group_937__gradients (optional) - T3: gradients computed in this iteration.
__group_937__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_937__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_937__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_938__weights (optional) - T2: weights to optimize.
__group_938__gradients (optional) - T3: gradients computed in this iteration.
__group_938__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_938__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_938__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_939__weights (optional) - T2: weights to optimize.
__group_939__gradients (optional) - T3: gradients computed in this iteration.
__group_939__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_939__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_939__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_940__weights (optional) - T2: weights to optimize.
__group_940__gradients (optional) - T3: gradients computed in this iteration.
__group_940__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_940__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_940__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_941__weights (optional) - T2: weights to optimize.
__group_941__gradients (optional) - T3: gradients computed in this iteration.
__group_941__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_941__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_941__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_942__weights (optional) - T2: weights to optimize.
__group_942__gradients (optional) - T3: gradients computed in this iteration.
__group_942__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_942__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_942__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_943__weights (optional) - T2: weights to optimize.
__group_943__gradients (optional) - T3: gradients computed in this iteration.
__group_943__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_943__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_943__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_944__weights (optional) - T2: weights to optimize.
__group_944__gradients (optional) - T3: gradients computed in this iteration.
__group_944__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_944__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_944__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_945__weights (optional) - T2: weights to optimize.
__group_945__gradients (optional) - T3: gradients computed in this iteration.
__group_945__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_945__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_945__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_946__weights (optional) - T2: weights to optimize.
__group_946__gradients (optional) - T3: gradients computed in this iteration.
__group_946__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_946__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_946__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_947__weights (optional) - T2: weights to optimize.
__group_947__gradients (optional) - T3: gradients computed in this iteration.
__group_947__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_947__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_947__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_948__weights (optional) - T2: weights to optimize.
__group_948__gradients (optional) - T3: gradients computed in this iteration.
__group_948__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_948__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_948__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_949__weights (optional) - T2: weights to optimize.
__group_949__gradients (optional) - T3: gradients computed in this iteration.
__group_949__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_949__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_949__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_950__weights (optional) - T2: weights to optimize.
__group_950__gradients (optional) - T3: gradients computed in this iteration.
__group_950__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_950__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_950__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_951__weights (optional) - T2: weights to optimize.
__group_951__gradients (optional) - T3: gradients computed in this iteration.
__group_951__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_951__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_951__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_952__weights (optional) - T2: weights to optimize.
__group_952__gradients (optional) - T3: gradients computed in this iteration.
__group_952__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_952__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_952__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_953__weights (optional) - T2: weights to optimize.
__group_953__gradients (optional) - T3: gradients computed in this iteration.
__group_953__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_953__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_953__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_954__weights (optional) - T2: weights to optimize.
__group_954__gradients (optional) - T3: gradients computed in this iteration.
__group_954__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_954__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_954__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_955__weights (optional) - T2: weights to optimize.
__group_955__gradients (optional) - T3: gradients computed in this iteration.
__group_955__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_955__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_955__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_956__weights (optional) - T2: weights to optimize.
__group_956__gradients (optional) - T3: gradients computed in this iteration.
__group_956__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_956__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_956__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_957__weights (optional) - T2: weights to optimize.
__group_957__gradients (optional) - T3: gradients computed in this iteration.
__group_957__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_957__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_957__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_958__weights (optional) - T2: weights to optimize.
__group_958__gradients (optional) - T3: gradients computed in this iteration.
__group_958__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_958__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_958__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_959__weights (optional) - T2: weights to optimize.
__group_959__gradients (optional) - T3: gradients computed in this iteration.
__group_959__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_959__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_959__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_960__weights (optional) - T2: weights to optimize.
__group_960__gradients (optional) - T3: gradients computed in this iteration.
__group_960__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_960__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_960__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_961__weights (optional) - T2: weights to optimize.
__group_961__gradients (optional) - T3: gradients computed in this iteration.
__group_961__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_961__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_961__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_962__weights (optional) - T2: weights to optimize.
__group_962__gradients (optional) - T3: gradients computed in this iteration.
__group_962__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_962__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_962__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_963__weights (optional) - T2: weights to optimize.
__group_963__gradients (optional) - T3: gradients computed in this iteration.
__group_963__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_963__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_963__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_964__weights (optional) - T2: weights to optimize.
__group_964__gradients (optional) - T3: gradients computed in this iteration.
__group_964__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_964__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_964__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_965__weights (optional) - T2: weights to optimize.
__group_965__gradients (optional) - T3: gradients computed in this iteration.
__group_965__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_965__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_965__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_966__weights (optional) - T2: weights to optimize.
__group_966__gradients (optional) - T3: gradients computed in this iteration.
__group_966__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_966__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_966__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_967__weights (optional) - T2: weights to optimize.
__group_967__gradients (optional) - T3: gradients computed in this iteration.
__group_967__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_967__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_967__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_968__weights (optional) - T2: weights to optimize.
__group_968__gradients (optional) - T3: gradients computed in this iteration.
__group_968__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_968__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_968__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_969__weights (optional) - T2: weights to optimize.
__group_969__gradients (optional) - T3: gradients computed in this iteration.
__group_969__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_969__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_969__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_970__weights (optional) - T2: weights to optimize.
__group_970__gradients (optional) - T3: gradients computed in this iteration.
__group_970__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_970__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_970__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_971__weights (optional) - T2: weights to optimize.
__group_971__gradients (optional) - T3: gradients computed in this iteration.
__group_971__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_971__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_971__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_972__weights (optional) - T2: weights to optimize.
__group_972__gradients (optional) - T3: gradients computed in this iteration.
__group_972__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_972__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_972__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_973__weights (optional) - T2: weights to optimize.
__group_973__gradients (optional) - T3: gradients computed in this iteration.
__group_973__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_973__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_973__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_974__weights (optional) - T2: weights to optimize.
__group_974__gradients (optional) - T3: gradients computed in this iteration.
__group_974__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_974__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_974__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_975__weights (optional) - T2: weights to optimize.
__group_975__gradients (optional) - T3: gradients computed in this iteration.
__group_975__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_975__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_975__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_976__weights (optional) - T2: weights to optimize.
__group_976__gradients (optional) - T3: gradients computed in this iteration.
__group_976__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_976__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_976__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_977__weights (optional) - T2: weights to optimize.
__group_977__gradients (optional) - T3: gradients computed in this iteration.
__group_977__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_977__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_977__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_978__weights (optional) - T2: weights to optimize.
__group_978__gradients (optional) - T3: gradients computed in this iteration.
__group_978__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_978__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_978__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_979__weights (optional) - T2: weights to optimize.
__group_979__gradients (optional) - T3: gradients computed in this iteration.
__group_979__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_979__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_979__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_980__weights (optional) - T2: weights to optimize.
__group_980__gradients (optional) - T3: gradients computed in this iteration.
__group_980__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_980__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_980__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_981__weights (optional) - T2: weights to optimize.
__group_981__gradients (optional) - T3: gradients computed in this iteration.
__group_981__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_981__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_981__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_982__weights (optional) - T2: weights to optimize.
__group_982__gradients (optional) - T3: gradients computed in this iteration.
__group_982__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_982__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_982__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_983__weights (optional) - T2: weights to optimize.
__group_983__gradients (optional) - T3: gradients computed in this iteration.
__group_983__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_983__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_983__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_984__weights (optional) - T2: weights to optimize.
__group_984__gradients (optional) - T3: gradients computed in this iteration.
__group_984__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_984__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_984__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_985__weights (optional) - T2: weights to optimize.
__group_985__gradients (optional) - T3: gradients computed in this iteration.
__group_985__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_985__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_985__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_986__weights (optional) - T2: weights to optimize.
__group_986__gradients (optional) - T3: gradients computed in this iteration.
__group_986__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_986__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_986__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_987__weights (optional) - T2: weights to optimize.
__group_987__gradients (optional) - T3: gradients computed in this iteration.
__group_987__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_987__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_987__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_988__weights (optional) - T2: weights to optimize.
__group_988__gradients (optional) - T3: gradients computed in this iteration.
__group_988__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_988__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_988__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_989__weights (optional) - T2: weights to optimize.
__group_989__gradients (optional) - T3: gradients computed in this iteration.
__group_989__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_989__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_989__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_990__weights (optional) - T2: weights to optimize.
__group_990__gradients (optional) - T3: gradients computed in this iteration.
__group_990__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_990__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_990__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_991__weights (optional) - T2: weights to optimize.
__group_991__gradients (optional) - T3: gradients computed in this iteration.
__group_991__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_991__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_991__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_992__weights (optional) - T2: weights to optimize.
__group_992__gradients (optional) - T3: gradients computed in this iteration.
__group_992__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_992__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_992__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_993__weights (optional) - T2: weights to optimize.
__group_993__gradients (optional) - T3: gradients computed in this iteration.
__group_993__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_993__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_993__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_994__weights (optional) - T2: weights to optimize.
__group_994__gradients (optional) - T3: gradients computed in this iteration.
__group_994__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_994__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_994__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_995__weights (optional) - T2: weights to optimize.
__group_995__gradients (optional) - T3: gradients computed in this iteration.
__group_995__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_995__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_995__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_996__weights (optional) - T2: weights to optimize.
__group_996__gradients (optional) - T3: gradients computed in this iteration.
__group_996__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_996__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_996__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_997__weights (optional) - T2: weights to optimize.
__group_997__gradients (optional) - T3: gradients computed in this iteration.
__group_997__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_997__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_997__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_998__weights (optional) - T2: weights to optimize.
__group_998__gradients (optional) - T3: gradients computed in this iteration.
__group_998__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_998__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_998__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_999__weights (optional) - T2: weights to optimize.
__group_999__gradients (optional) - T3: gradients computed in this iteration.
__group_999__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_999__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_999__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1000__weights (optional) - T2: weights to optimize.
__group_1000__gradients (optional) - T3: gradients computed in this iteration.
__group_1000__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1000__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1000__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1001__weights (optional) - T2: weights to optimize.
__group_1001__gradients (optional) - T3: gradients computed in this iteration.
__group_1001__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1001__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1001__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1002__weights (optional) - T2: weights to optimize.
__group_1002__gradients (optional) - T3: gradients computed in this iteration.
__group_1002__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1002__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1002__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1003__weights (optional) - T2: weights to optimize.
__group_1003__gradients (optional) - T3: gradients computed in this iteration.
__group_1003__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1003__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1003__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1004__weights (optional) - T2: weights to optimize.
__group_1004__gradients (optional) - T3: gradients computed in this iteration.
__group_1004__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1004__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1004__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1005__weights (optional) - T2: weights to optimize.
__group_1005__gradients (optional) - T3: gradients computed in this iteration.
__group_1005__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1005__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1005__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1006__weights (optional) - T2: weights to optimize.
__group_1006__gradients (optional) - T3: gradients computed in this iteration.
__group_1006__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1006__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1006__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1007__weights (optional) - T2: weights to optimize.
__group_1007__gradients (optional) - T3: gradients computed in this iteration.
__group_1007__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1007__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1007__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1008__weights (optional) - T2: weights to optimize.
__group_1008__gradients (optional) - T3: gradients computed in this iteration.
__group_1008__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1008__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1008__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1009__weights (optional) - T2: weights to optimize.
__group_1009__gradients (optional) - T3: gradients computed in this iteration.
__group_1009__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1009__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1009__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1010__weights (optional) - T2: weights to optimize.
__group_1010__gradients (optional) - T3: gradients computed in this iteration.
__group_1010__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1010__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1010__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1011__weights (optional) - T2: weights to optimize.
__group_1011__gradients (optional) - T3: gradients computed in this iteration.
__group_1011__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1011__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1011__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1012__weights (optional) - T2: weights to optimize.
__group_1012__gradients (optional) - T3: gradients computed in this iteration.
__group_1012__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1012__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1012__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1013__weights (optional) - T2: weights to optimize.
__group_1013__gradients (optional) - T3: gradients computed in this iteration.
__group_1013__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1013__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1013__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1014__weights (optional) - T2: weights to optimize.
__group_1014__gradients (optional) - T3: gradients computed in this iteration.
__group_1014__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1014__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1014__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1015__weights (optional) - T2: weights to optimize.
__group_1015__gradients (optional) - T3: gradients computed in this iteration.
__group_1015__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1015__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1015__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1016__weights (optional) - T2: weights to optimize.
__group_1016__gradients (optional) - T3: gradients computed in this iteration.
__group_1016__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1016__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1016__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1017__weights (optional) - T2: weights to optimize.
__group_1017__gradients (optional) - T3: gradients computed in this iteration.
__group_1017__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1017__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1017__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1018__weights (optional) - T2: weights to optimize.
__group_1018__gradients (optional) - T3: gradients computed in this iteration.
__group_1018__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1018__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1018__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1019__weights (optional) - T2: weights to optimize.
__group_1019__gradients (optional) - T3: gradients computed in this iteration.
__group_1019__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1019__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1019__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1020__weights (optional) - T2: weights to optimize.
__group_1020__gradients (optional) - T3: gradients computed in this iteration.
__group_1020__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1020__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1020__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1021__weights (optional) - T2: weights to optimize.
__group_1021__gradients (optional) - T3: gradients computed in this iteration.
__group_1021__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1021__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1021__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1022__weights (optional) - T2: weights to optimize.
__group_1022__gradients (optional) - T3: gradients computed in this iteration.
__group_1022__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1022__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1022__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1023__weights (optional) - T2: weights to optimize.
__group_1023__gradients (optional) - T3: gradients computed in this iteration.
__group_1023__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1023__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1023__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
Outputs
Between 0 and 5121 outputs.
new_step (optional, heterogeneous) - TInt64: One-based index of the next training iteration.
__group_0__new_weights (optional) - T2: New weights
__group_0__new_gradients (optional) - T3: New gradients
__group_0__new_moment_1 (optional) - T4: New averaged gradients
__group_0__new_moment_2 (optional) - T4: New averaged squared gradients
__group_0__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1__new_weights (optional) - T2: New weights
__group_1__new_gradients (optional) - T3: New gradients
__group_1__new_moment_1 (optional) - T4: New averaged gradients
__group_1__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_2__new_weights (optional) - T2: New weights
__group_2__new_gradients (optional) - T3: New gradients
__group_2__new_moment_1 (optional) - T4: New averaged gradients
__group_2__new_moment_2 (optional) - T4: New averaged squared gradients
__group_2__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_3__new_weights (optional) - T2: New weights
__group_3__new_gradients (optional) - T3: New gradients
__group_3__new_moment_1 (optional) - T4: New averaged gradients
__group_3__new_moment_2 (optional) - T4: New averaged squared gradients
__group_3__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_4__new_weights (optional) - T2: New weights
__group_4__new_gradients (optional) - T3: New gradients
__group_4__new_moment_1 (optional) - T4: New averaged gradients
__group_4__new_moment_2 (optional) - T4: New averaged squared gradients
__group_4__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_5__new_weights (optional) - T2: New weights
__group_5__new_gradients (optional) - T3: New gradients
__group_5__new_moment_1 (optional) - T4: New averaged gradients
__group_5__new_moment_2 (optional) - T4: New averaged squared gradients
__group_5__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_6__new_weights (optional) - T2: New weights
__group_6__new_gradients (optional) - T3: New gradients
__group_6__new_moment_1 (optional) - T4: New averaged gradients
__group_6__new_moment_2 (optional) - T4: New averaged squared gradients
__group_6__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_7__new_weights (optional) - T2: New weights
__group_7__new_gradients (optional) - T3: New gradients
__group_7__new_moment_1 (optional) - T4: New averaged gradients
__group_7__new_moment_2 (optional) - T4: New averaged squared gradients
__group_7__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_8__new_weights (optional) - T2: New weights
__group_8__new_gradients (optional) - T3: New gradients
__group_8__new_moment_1 (optional) - T4: New averaged gradients
__group_8__new_moment_2 (optional) - T4: New averaged squared gradients
__group_8__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_9__new_weights (optional) - T2: New weights
__group_9__new_gradients (optional) - T3: New gradients
__group_9__new_moment_1 (optional) - T4: New averaged gradients
__group_9__new_moment_2 (optional) - T4: New averaged squared gradients
__group_9__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_10__new_weights (optional) - T2: New weights
__group_10__new_gradients (optional) - T3: New gradients
__group_10__new_moment_1 (optional) - T4: New averaged gradients
__group_10__new_moment_2 (optional) - T4: New averaged squared gradients
__group_10__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_11__new_weights (optional) - T2: New weights
__group_11__new_gradients (optional) - T3: New gradients
__group_11__new_moment_1 (optional) - T4: New averaged gradients
__group_11__new_moment_2 (optional) - T4: New averaged squared gradients
__group_11__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_12__new_weights (optional) - T2: New weights
__group_12__new_gradients (optional) - T3: New gradients
__group_12__new_moment_1 (optional) - T4: New averaged gradients
__group_12__new_moment_2 (optional) - T4: New averaged squared gradients
__group_12__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_13__new_weights (optional) - T2: New weights
__group_13__new_gradients (optional) - T3: New gradients
__group_13__new_moment_1 (optional) - T4: New averaged gradients
__group_13__new_moment_2 (optional) - T4: New averaged squared gradients
__group_13__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_14__new_weights (optional) - T2: New weights
__group_14__new_gradients (optional) - T3: New gradients
__group_14__new_moment_1 (optional) - T4: New averaged gradients
__group_14__new_moment_2 (optional) - T4: New averaged squared gradients
__group_14__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_15__new_weights (optional) - T2: New weights
__group_15__new_gradients (optional) - T3: New gradients
__group_15__new_moment_1 (optional) - T4: New averaged gradients
__group_15__new_moment_2 (optional) - T4: New averaged squared gradients
__group_15__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_16__new_weights (optional) - T2: New weights
__group_16__new_gradients (optional) - T3: New gradients
__group_16__new_moment_1 (optional) - T4: New averaged gradients
__group_16__new_moment_2 (optional) - T4: New averaged squared gradients
__group_16__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_17__new_weights (optional) - T2: New weights
__group_17__new_gradients (optional) - T3: New gradients
__group_17__new_moment_1 (optional) - T4: New averaged gradients
__group_17__new_moment_2 (optional) - T4: New averaged squared gradients
__group_17__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_18__new_weights (optional) - T2: New weights
__group_18__new_gradients (optional) - T3: New gradients
__group_18__new_moment_1 (optional) - T4: New averaged gradients
__group_18__new_moment_2 (optional) - T4: New averaged squared gradients
__group_18__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_19__new_weights (optional) - T2: New weights
__group_19__new_gradients (optional) - T3: New gradients
__group_19__new_moment_1 (optional) - T4: New averaged gradients
__group_19__new_moment_2 (optional) - T4: New averaged squared gradients
__group_19__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_20__new_weights (optional) - T2: New weights
__group_20__new_gradients (optional) - T3: New gradients
__group_20__new_moment_1 (optional) - T4: New averaged gradients
__group_20__new_moment_2 (optional) - T4: New averaged squared gradients
__group_20__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_21__new_weights (optional) - T2: New weights
__group_21__new_gradients (optional) - T3: New gradients
__group_21__new_moment_1 (optional) - T4: New averaged gradients
__group_21__new_moment_2 (optional) - T4: New averaged squared gradients
__group_21__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_22__new_weights (optional) - T2: New weights
__group_22__new_gradients (optional) - T3: New gradients
__group_22__new_moment_1 (optional) - T4: New averaged gradients
__group_22__new_moment_2 (optional) - T4: New averaged squared gradients
__group_22__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_23__new_weights (optional) - T2: New weights
__group_23__new_gradients (optional) - T3: New gradients
__group_23__new_moment_1 (optional) - T4: New averaged gradients
__group_23__new_moment_2 (optional) - T4: New averaged squared gradients
__group_23__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_24__new_weights (optional) - T2: New weights
__group_24__new_gradients (optional) - T3: New gradients
__group_24__new_moment_1 (optional) - T4: New averaged gradients
__group_24__new_moment_2 (optional) - T4: New averaged squared gradients
__group_24__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_25__new_weights (optional) - T2: New weights
__group_25__new_gradients (optional) - T3: New gradients
__group_25__new_moment_1 (optional) - T4: New averaged gradients
__group_25__new_moment_2 (optional) - T4: New averaged squared gradients
__group_25__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_26__new_weights (optional) - T2: New weights
__group_26__new_gradients (optional) - T3: New gradients
__group_26__new_moment_1 (optional) - T4: New averaged gradients
__group_26__new_moment_2 (optional) - T4: New averaged squared gradients
__group_26__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_27__new_weights (optional) - T2: New weights
__group_27__new_gradients (optional) - T3: New gradients
__group_27__new_moment_1 (optional) - T4: New averaged gradients
__group_27__new_moment_2 (optional) - T4: New averaged squared gradients
__group_27__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_28__new_weights (optional) - T2: New weights
__group_28__new_gradients (optional) - T3: New gradients
__group_28__new_moment_1 (optional) - T4: New averaged gradients
__group_28__new_moment_2 (optional) - T4: New averaged squared gradients
__group_28__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_29__new_weights (optional) - T2: New weights
__group_29__new_gradients (optional) - T3: New gradients
__group_29__new_moment_1 (optional) - T4: New averaged gradients
__group_29__new_moment_2 (optional) - T4: New averaged squared gradients
__group_29__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_30__new_weights (optional) - T2: New weights
__group_30__new_gradients (optional) - T3: New gradients
__group_30__new_moment_1 (optional) - T4: New averaged gradients
__group_30__new_moment_2 (optional) - T4: New averaged squared gradients
__group_30__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_31__new_weights (optional) - T2: New weights
__group_31__new_gradients (optional) - T3: New gradients
__group_31__new_moment_1 (optional) - T4: New averaged gradients
__group_31__new_moment_2 (optional) - T4: New averaged squared gradients
__group_31__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_32__new_weights (optional) - T2: New weights
__group_32__new_gradients (optional) - T3: New gradients
__group_32__new_moment_1 (optional) - T4: New averaged gradients
__group_32__new_moment_2 (optional) - T4: New averaged squared gradients
__group_32__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_33__new_weights (optional) - T2: New weights
__group_33__new_gradients (optional) - T3: New gradients
__group_33__new_moment_1 (optional) - T4: New averaged gradients
__group_33__new_moment_2 (optional) - T4: New averaged squared gradients
__group_33__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_34__new_weights (optional) - T2: New weights
__group_34__new_gradients (optional) - T3: New gradients
__group_34__new_moment_1 (optional) - T4: New averaged gradients
__group_34__new_moment_2 (optional) - T4: New averaged squared gradients
__group_34__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_35__new_weights (optional) - T2: New weights
__group_35__new_gradients (optional) - T3: New gradients
__group_35__new_moment_1 (optional) - T4: New averaged gradients
__group_35__new_moment_2 (optional) - T4: New averaged squared gradients
__group_35__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_36__new_weights (optional) - T2: New weights
__group_36__new_gradients (optional) - T3: New gradients
__group_36__new_moment_1 (optional) - T4: New averaged gradients
__group_36__new_moment_2 (optional) - T4: New averaged squared gradients
__group_36__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_37__new_weights (optional) - T2: New weights
__group_37__new_gradients (optional) - T3: New gradients
__group_37__new_moment_1 (optional) - T4: New averaged gradients
__group_37__new_moment_2 (optional) - T4: New averaged squared gradients
__group_37__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_38__new_weights (optional) - T2: New weights
__group_38__new_gradients (optional) - T3: New gradients
__group_38__new_moment_1 (optional) - T4: New averaged gradients
__group_38__new_moment_2 (optional) - T4: New averaged squared gradients
__group_38__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_39__new_weights (optional) - T2: New weights
__group_39__new_gradients (optional) - T3: New gradients
__group_39__new_moment_1 (optional) - T4: New averaged gradients
__group_39__new_moment_2 (optional) - T4: New averaged squared gradients
__group_39__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_40__new_weights (optional) - T2: New weights
__group_40__new_gradients (optional) - T3: New gradients
__group_40__new_moment_1 (optional) - T4: New averaged gradients
__group_40__new_moment_2 (optional) - T4: New averaged squared gradients
__group_40__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_41__new_weights (optional) - T2: New weights
__group_41__new_gradients (optional) - T3: New gradients
__group_41__new_moment_1 (optional) - T4: New averaged gradients
__group_41__new_moment_2 (optional) - T4: New averaged squared gradients
__group_41__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_42__new_weights (optional) - T2: New weights
__group_42__new_gradients (optional) - T3: New gradients
__group_42__new_moment_1 (optional) - T4: New averaged gradients
__group_42__new_moment_2 (optional) - T4: New averaged squared gradients
__group_42__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_43__new_weights (optional) - T2: New weights
__group_43__new_gradients (optional) - T3: New gradients
__group_43__new_moment_1 (optional) - T4: New averaged gradients
__group_43__new_moment_2 (optional) - T4: New averaged squared gradients
__group_43__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_44__new_weights (optional) - T2: New weights
__group_44__new_gradients (optional) - T3: New gradients
__group_44__new_moment_1 (optional) - T4: New averaged gradients
__group_44__new_moment_2 (optional) - T4: New averaged squared gradients
__group_44__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_45__new_weights (optional) - T2: New weights
__group_45__new_gradients (optional) - T3: New gradients
__group_45__new_moment_1 (optional) - T4: New averaged gradients
__group_45__new_moment_2 (optional) - T4: New averaged squared gradients
__group_45__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_46__new_weights (optional) - T2: New weights
__group_46__new_gradients (optional) - T3: New gradients
__group_46__new_moment_1 (optional) - T4: New averaged gradients
__group_46__new_moment_2 (optional) - T4: New averaged squared gradients
__group_46__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_47__new_weights (optional) - T2: New weights
__group_47__new_gradients (optional) - T3: New gradients
__group_47__new_moment_1 (optional) - T4: New averaged gradients
__group_47__new_moment_2 (optional) - T4: New averaged squared gradients
__group_47__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_48__new_weights (optional) - T2: New weights
__group_48__new_gradients (optional) - T3: New gradients
__group_48__new_moment_1 (optional) - T4: New averaged gradients
__group_48__new_moment_2 (optional) - T4: New averaged squared gradients
__group_48__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_49__new_weights (optional) - T2: New weights
__group_49__new_gradients (optional) - T3: New gradients
__group_49__new_moment_1 (optional) - T4: New averaged gradients
__group_49__new_moment_2 (optional) - T4: New averaged squared gradients
__group_49__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_50__new_weights (optional) - T2: New weights
__group_50__new_gradients (optional) - T3: New gradients
__group_50__new_moment_1 (optional) - T4: New averaged gradients
__group_50__new_moment_2 (optional) - T4: New averaged squared gradients
__group_50__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_51__new_weights (optional) - T2: New weights
__group_51__new_gradients (optional) - T3: New gradients
__group_51__new_moment_1 (optional) - T4: New averaged gradients
__group_51__new_moment_2 (optional) - T4: New averaged squared gradients
__group_51__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_52__new_weights (optional) - T2: New weights
__group_52__new_gradients (optional) - T3: New gradients
__group_52__new_moment_1 (optional) - T4: New averaged gradients
__group_52__new_moment_2 (optional) - T4: New averaged squared gradients
__group_52__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_53__new_weights (optional) - T2: New weights
__group_53__new_gradients (optional) - T3: New gradients
__group_53__new_moment_1 (optional) - T4: New averaged gradients
__group_53__new_moment_2 (optional) - T4: New averaged squared gradients
__group_53__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_54__new_weights (optional) - T2: New weights
__group_54__new_gradients (optional) - T3: New gradients
__group_54__new_moment_1 (optional) - T4: New averaged gradients
__group_54__new_moment_2 (optional) - T4: New averaged squared gradients
__group_54__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_55__new_weights (optional) - T2: New weights
__group_55__new_gradients (optional) - T3: New gradients
__group_55__new_moment_1 (optional) - T4: New averaged gradients
__group_55__new_moment_2 (optional) - T4: New averaged squared gradients
__group_55__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_56__new_weights (optional) - T2: New weights
__group_56__new_gradients (optional) - T3: New gradients
__group_56__new_moment_1 (optional) - T4: New averaged gradients
__group_56__new_moment_2 (optional) - T4: New averaged squared gradients
__group_56__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_57__new_weights (optional) - T2: New weights
__group_57__new_gradients (optional) - T3: New gradients
__group_57__new_moment_1 (optional) - T4: New averaged gradients
__group_57__new_moment_2 (optional) - T4: New averaged squared gradients
__group_57__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_58__new_weights (optional) - T2: New weights
__group_58__new_gradients (optional) - T3: New gradients
__group_58__new_moment_1 (optional) - T4: New averaged gradients
__group_58__new_moment_2 (optional) - T4: New averaged squared gradients
__group_58__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_59__new_weights (optional) - T2: New weights
__group_59__new_gradients (optional) - T3: New gradients
__group_59__new_moment_1 (optional) - T4: New averaged gradients
__group_59__new_moment_2 (optional) - T4: New averaged squared gradients
__group_59__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_60__new_weights (optional) - T2: New weights
__group_60__new_gradients (optional) - T3: New gradients
__group_60__new_moment_1 (optional) - T4: New averaged gradients
__group_60__new_moment_2 (optional) - T4: New averaged squared gradients
__group_60__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_61__new_weights (optional) - T2: New weights
__group_61__new_gradients (optional) - T3: New gradients
__group_61__new_moment_1 (optional) - T4: New averaged gradients
__group_61__new_moment_2 (optional) - T4: New averaged squared gradients
__group_61__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_62__new_weights (optional) - T2: New weights
__group_62__new_gradients (optional) - T3: New gradients
__group_62__new_moment_1 (optional) - T4: New averaged gradients
__group_62__new_moment_2 (optional) - T4: New averaged squared gradients
__group_62__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_63__new_weights (optional) - T2: New weights
__group_63__new_gradients (optional) - T3: New gradients
__group_63__new_moment_1 (optional) - T4: New averaged gradients
__group_63__new_moment_2 (optional) - T4: New averaged squared gradients
__group_63__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_64__new_weights (optional) - T2: New weights
__group_64__new_gradients (optional) - T3: New gradients
__group_64__new_moment_1 (optional) - T4: New averaged gradients
__group_64__new_moment_2 (optional) - T4: New averaged squared gradients
__group_64__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_65__new_weights (optional) - T2: New weights
__group_65__new_gradients (optional) - T3: New gradients
__group_65__new_moment_1 (optional) - T4: New averaged gradients
__group_65__new_moment_2 (optional) - T4: New averaged squared gradients
__group_65__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_66__new_weights (optional) - T2: New weights
__group_66__new_gradients (optional) - T3: New gradients
__group_66__new_moment_1 (optional) - T4: New averaged gradients
__group_66__new_moment_2 (optional) - T4: New averaged squared gradients
__group_66__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_67__new_weights (optional) - T2: New weights
__group_67__new_gradients (optional) - T3: New gradients
__group_67__new_moment_1 (optional) - T4: New averaged gradients
__group_67__new_moment_2 (optional) - T4: New averaged squared gradients
__group_67__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_68__new_weights (optional) - T2: New weights
__group_68__new_gradients (optional) - T3: New gradients
__group_68__new_moment_1 (optional) - T4: New averaged gradients
__group_68__new_moment_2 (optional) - T4: New averaged squared gradients
__group_68__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_69__new_weights (optional) - T2: New weights
__group_69__new_gradients (optional) - T3: New gradients
__group_69__new_moment_1 (optional) - T4: New averaged gradients
__group_69__new_moment_2 (optional) - T4: New averaged squared gradients
__group_69__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_70__new_weights (optional) - T2: New weights
__group_70__new_gradients (optional) - T3: New gradients
__group_70__new_moment_1 (optional) - T4: New averaged gradients
__group_70__new_moment_2 (optional) - T4: New averaged squared gradients
__group_70__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_71__new_weights (optional) - T2: New weights
__group_71__new_gradients (optional) - T3: New gradients
__group_71__new_moment_1 (optional) - T4: New averaged gradients
__group_71__new_moment_2 (optional) - T4: New averaged squared gradients
__group_71__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_72__new_weights (optional) - T2: New weights
__group_72__new_gradients (optional) - T3: New gradients
__group_72__new_moment_1 (optional) - T4: New averaged gradients
__group_72__new_moment_2 (optional) - T4: New averaged squared gradients
__group_72__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_73__new_weights (optional) - T2: New weights
__group_73__new_gradients (optional) - T3: New gradients
__group_73__new_moment_1 (optional) - T4: New averaged gradients
__group_73__new_moment_2 (optional) - T4: New averaged squared gradients
__group_73__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_74__new_weights (optional) - T2: New weights
__group_74__new_gradients (optional) - T3: New gradients
__group_74__new_moment_1 (optional) - T4: New averaged gradients
__group_74__new_moment_2 (optional) - T4: New averaged squared gradients
__group_74__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_75__new_weights (optional) - T2: New weights
__group_75__new_gradients (optional) - T3: New gradients
__group_75__new_moment_1 (optional) - T4: New averaged gradients
__group_75__new_moment_2 (optional) - T4: New averaged squared gradients
__group_75__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_76__new_weights (optional) - T2: New weights
__group_76__new_gradients (optional) - T3: New gradients
__group_76__new_moment_1 (optional) - T4: New averaged gradients
__group_76__new_moment_2 (optional) - T4: New averaged squared gradients
__group_76__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_77__new_weights (optional) - T2: New weights
__group_77__new_gradients (optional) - T3: New gradients
__group_77__new_moment_1 (optional) - T4: New averaged gradients
__group_77__new_moment_2 (optional) - T4: New averaged squared gradients
__group_77__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_78__new_weights (optional) - T2: New weights
__group_78__new_gradients (optional) - T3: New gradients
__group_78__new_moment_1 (optional) - T4: New averaged gradients
__group_78__new_moment_2 (optional) - T4: New averaged squared gradients
__group_78__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_79__new_weights (optional) - T2: New weights
__group_79__new_gradients (optional) - T3: New gradients
__group_79__new_moment_1 (optional) - T4: New averaged gradients
__group_79__new_moment_2 (optional) - T4: New averaged squared gradients
__group_79__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_80__new_weights (optional) - T2: New weights
__group_80__new_gradients (optional) - T3: New gradients
__group_80__new_moment_1 (optional) - T4: New averaged gradients
__group_80__new_moment_2 (optional) - T4: New averaged squared gradients
__group_80__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_81__new_weights (optional) - T2: New weights
__group_81__new_gradients (optional) - T3: New gradients
__group_81__new_moment_1 (optional) - T4: New averaged gradients
__group_81__new_moment_2 (optional) - T4: New averaged squared gradients
__group_81__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_82__new_weights (optional) - T2: New weights
__group_82__new_gradients (optional) - T3: New gradients
__group_82__new_moment_1 (optional) - T4: New averaged gradients
__group_82__new_moment_2 (optional) - T4: New averaged squared gradients
__group_82__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_83__new_weights (optional) - T2: New weights
__group_83__new_gradients (optional) - T3: New gradients
__group_83__new_moment_1 (optional) - T4: New averaged gradients
__group_83__new_moment_2 (optional) - T4: New averaged squared gradients
__group_83__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_84__new_weights (optional) - T2: New weights
__group_84__new_gradients (optional) - T3: New gradients
__group_84__new_moment_1 (optional) - T4: New averaged gradients
__group_84__new_moment_2 (optional) - T4: New averaged squared gradients
__group_84__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_85__new_weights (optional) - T2: New weights
__group_85__new_gradients (optional) - T3: New gradients
__group_85__new_moment_1 (optional) - T4: New averaged gradients
__group_85__new_moment_2 (optional) - T4: New averaged squared gradients
__group_85__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_86__new_weights (optional) - T2: New weights
__group_86__new_gradients (optional) - T3: New gradients
__group_86__new_moment_1 (optional) - T4: New averaged gradients
__group_86__new_moment_2 (optional) - T4: New averaged squared gradients
__group_86__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_87__new_weights (optional) - T2: New weights
__group_87__new_gradients (optional) - T3: New gradients
__group_87__new_moment_1 (optional) - T4: New averaged gradients
__group_87__new_moment_2 (optional) - T4: New averaged squared gradients
__group_87__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_88__new_weights (optional) - T2: New weights
__group_88__new_gradients (optional) - T3: New gradients
__group_88__new_moment_1 (optional) - T4: New averaged gradients
__group_88__new_moment_2 (optional) - T4: New averaged squared gradients
__group_88__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_89__new_weights (optional) - T2: New weights
__group_89__new_gradients (optional) - T3: New gradients
__group_89__new_moment_1 (optional) - T4: New averaged gradients
__group_89__new_moment_2 (optional) - T4: New averaged squared gradients
__group_89__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_90__new_weights (optional) - T2: New weights
__group_90__new_gradients (optional) - T3: New gradients
__group_90__new_moment_1 (optional) - T4: New averaged gradients
__group_90__new_moment_2 (optional) - T4: New averaged squared gradients
__group_90__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_91__new_weights (optional) - T2: New weights
__group_91__new_gradients (optional) - T3: New gradients
__group_91__new_moment_1 (optional) - T4: New averaged gradients
__group_91__new_moment_2 (optional) - T4: New averaged squared gradients
__group_91__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_92__new_weights (optional) - T2: New weights
__group_92__new_gradients (optional) - T3: New gradients
__group_92__new_moment_1 (optional) - T4: New averaged gradients
__group_92__new_moment_2 (optional) - T4: New averaged squared gradients
__group_92__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_93__new_weights (optional) - T2: New weights
__group_93__new_gradients (optional) - T3: New gradients
__group_93__new_moment_1 (optional) - T4: New averaged gradients
__group_93__new_moment_2 (optional) - T4: New averaged squared gradients
__group_93__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_94__new_weights (optional) - T2: New weights
__group_94__new_gradients (optional) - T3: New gradients
__group_94__new_moment_1 (optional) - T4: New averaged gradients
__group_94__new_moment_2 (optional) - T4: New averaged squared gradients
__group_94__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_95__new_weights (optional) - T2: New weights
__group_95__new_gradients (optional) - T3: New gradients
__group_95__new_moment_1 (optional) - T4: New averaged gradients
__group_95__new_moment_2 (optional) - T4: New averaged squared gradients
__group_95__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_96__new_weights (optional) - T2: New weights
__group_96__new_gradients (optional) - T3: New gradients
__group_96__new_moment_1 (optional) - T4: New averaged gradients
__group_96__new_moment_2 (optional) - T4: New averaged squared gradients
__group_96__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_97__new_weights (optional) - T2: New weights
__group_97__new_gradients (optional) - T3: New gradients
__group_97__new_moment_1 (optional) - T4: New averaged gradients
__group_97__new_moment_2 (optional) - T4: New averaged squared gradients
__group_97__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_98__new_weights (optional) - T2: New weights
__group_98__new_gradients (optional) - T3: New gradients
__group_98__new_moment_1 (optional) - T4: New averaged gradients
__group_98__new_moment_2 (optional) - T4: New averaged squared gradients
__group_98__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_99__new_weights (optional) - T2: New weights
__group_99__new_gradients (optional) - T3: New gradients
__group_99__new_moment_1 (optional) - T4: New averaged gradients
__group_99__new_moment_2 (optional) - T4: New averaged squared gradients
__group_99__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_100__new_weights (optional) - T2: New weights
__group_100__new_gradients (optional) - T3: New gradients
__group_100__new_moment_1 (optional) - T4: New averaged gradients
__group_100__new_moment_2 (optional) - T4: New averaged squared gradients
__group_100__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_101__new_weights (optional) - T2: New weights
__group_101__new_gradients (optional) - T3: New gradients
__group_101__new_moment_1 (optional) - T4: New averaged gradients
__group_101__new_moment_2 (optional) - T4: New averaged squared gradients
__group_101__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_102__new_weights (optional) - T2: New weights
__group_102__new_gradients (optional) - T3: New gradients
__group_102__new_moment_1 (optional) - T4: New averaged gradients
__group_102__new_moment_2 (optional) - T4: New averaged squared gradients
__group_102__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_103__new_weights (optional) - T2: New weights
__group_103__new_gradients (optional) - T3: New gradients
__group_103__new_moment_1 (optional) - T4: New averaged gradients
__group_103__new_moment_2 (optional) - T4: New averaged squared gradients
__group_103__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_104__new_weights (optional) - T2: New weights
__group_104__new_gradients (optional) - T3: New gradients
__group_104__new_moment_1 (optional) - T4: New averaged gradients
__group_104__new_moment_2 (optional) - T4: New averaged squared gradients
__group_104__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_105__new_weights (optional) - T2: New weights
__group_105__new_gradients (optional) - T3: New gradients
__group_105__new_moment_1 (optional) - T4: New averaged gradients
__group_105__new_moment_2 (optional) - T4: New averaged squared gradients
__group_105__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_106__new_weights (optional) - T2: New weights
__group_106__new_gradients (optional) - T3: New gradients
__group_106__new_moment_1 (optional) - T4: New averaged gradients
__group_106__new_moment_2 (optional) - T4: New averaged squared gradients
__group_106__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_107__new_weights (optional) - T2: New weights
__group_107__new_gradients (optional) - T3: New gradients
__group_107__new_moment_1 (optional) - T4: New averaged gradients
__group_107__new_moment_2 (optional) - T4: New averaged squared gradients
__group_107__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_108__new_weights (optional) - T2: New weights
__group_108__new_gradients (optional) - T3: New gradients
__group_108__new_moment_1 (optional) - T4: New averaged gradients
__group_108__new_moment_2 (optional) - T4: New averaged squared gradients
__group_108__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_109__new_weights (optional) - T2: New weights
__group_109__new_gradients (optional) - T3: New gradients
__group_109__new_moment_1 (optional) - T4: New averaged gradients
__group_109__new_moment_2 (optional) - T4: New averaged squared gradients
__group_109__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_110__new_weights (optional) - T2: New weights
__group_110__new_gradients (optional) - T3: New gradients
__group_110__new_moment_1 (optional) - T4: New averaged gradients
__group_110__new_moment_2 (optional) - T4: New averaged squared gradients
__group_110__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_111__new_weights (optional) - T2: New weights
__group_111__new_gradients (optional) - T3: New gradients
__group_111__new_moment_1 (optional) - T4: New averaged gradients
__group_111__new_moment_2 (optional) - T4: New averaged squared gradients
__group_111__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_112__new_weights (optional) - T2: New weights
__group_112__new_gradients (optional) - T3: New gradients
__group_112__new_moment_1 (optional) - T4: New averaged gradients
__group_112__new_moment_2 (optional) - T4: New averaged squared gradients
__group_112__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_113__new_weights (optional) - T2: New weights
__group_113__new_gradients (optional) - T3: New gradients
__group_113__new_moment_1 (optional) - T4: New averaged gradients
__group_113__new_moment_2 (optional) - T4: New averaged squared gradients
__group_113__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_114__new_weights (optional) - T2: New weights
__group_114__new_gradients (optional) - T3: New gradients
__group_114__new_moment_1 (optional) - T4: New averaged gradients
__group_114__new_moment_2 (optional) - T4: New averaged squared gradients
__group_114__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_115__new_weights (optional) - T2: New weights
__group_115__new_gradients (optional) - T3: New gradients
__group_115__new_moment_1 (optional) - T4: New averaged gradients
__group_115__new_moment_2 (optional) - T4: New averaged squared gradients
__group_115__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_116__new_weights (optional) - T2: New weights
__group_116__new_gradients (optional) - T3: New gradients
__group_116__new_moment_1 (optional) - T4: New averaged gradients
__group_116__new_moment_2 (optional) - T4: New averaged squared gradients
__group_116__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_117__new_weights (optional) - T2: New weights
__group_117__new_gradients (optional) - T3: New gradients
__group_117__new_moment_1 (optional) - T4: New averaged gradients
__group_117__new_moment_2 (optional) - T4: New averaged squared gradients
__group_117__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_118__new_weights (optional) - T2: New weights
__group_118__new_gradients (optional) - T3: New gradients
__group_118__new_moment_1 (optional) - T4: New averaged gradients
__group_118__new_moment_2 (optional) - T4: New averaged squared gradients
__group_118__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_119__new_weights (optional) - T2: New weights
__group_119__new_gradients (optional) - T3: New gradients
__group_119__new_moment_1 (optional) - T4: New averaged gradients
__group_119__new_moment_2 (optional) - T4: New averaged squared gradients
__group_119__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_120__new_weights (optional) - T2: New weights
__group_120__new_gradients (optional) - T3: New gradients
__group_120__new_moment_1 (optional) - T4: New averaged gradients
__group_120__new_moment_2 (optional) - T4: New averaged squared gradients
__group_120__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_121__new_weights (optional) - T2: New weights
__group_121__new_gradients (optional) - T3: New gradients
__group_121__new_moment_1 (optional) - T4: New averaged gradients
__group_121__new_moment_2 (optional) - T4: New averaged squared gradients
__group_121__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_122__new_weights (optional) - T2: New weights
__group_122__new_gradients (optional) - T3: New gradients
__group_122__new_moment_1 (optional) - T4: New averaged gradients
__group_122__new_moment_2 (optional) - T4: New averaged squared gradients
__group_122__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_123__new_weights (optional) - T2: New weights
__group_123__new_gradients (optional) - T3: New gradients
__group_123__new_moment_1 (optional) - T4: New averaged gradients
__group_123__new_moment_2 (optional) - T4: New averaged squared gradients
__group_123__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_124__new_weights (optional) - T2: New weights
__group_124__new_gradients (optional) - T3: New gradients
__group_124__new_moment_1 (optional) - T4: New averaged gradients
__group_124__new_moment_2 (optional) - T4: New averaged squared gradients
__group_124__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_125__new_weights (optional) - T2: New weights
__group_125__new_gradients (optional) - T3: New gradients
__group_125__new_moment_1 (optional) - T4: New averaged gradients
__group_125__new_moment_2 (optional) - T4: New averaged squared gradients
__group_125__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_126__new_weights (optional) - T2: New weights
__group_126__new_gradients (optional) - T3: New gradients
__group_126__new_moment_1 (optional) - T4: New averaged gradients
__group_126__new_moment_2 (optional) - T4: New averaged squared gradients
__group_126__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_127__new_weights (optional) - T2: New weights
__group_127__new_gradients (optional) - T3: New gradients
__group_127__new_moment_1 (optional) - T4: New averaged gradients
__group_127__new_moment_2 (optional) - T4: New averaged squared gradients
__group_127__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_128__new_weights (optional) - T2: New weights
__group_128__new_gradients (optional) - T3: New gradients
__group_128__new_moment_1 (optional) - T4: New averaged gradients
__group_128__new_moment_2 (optional) - T4: New averaged squared gradients
__group_128__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_129__new_weights (optional) - T2: New weights
__group_129__new_gradients (optional) - T3: New gradients
__group_129__new_moment_1 (optional) - T4: New averaged gradients
__group_129__new_moment_2 (optional) - T4: New averaged squared gradients
__group_129__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_130__new_weights (optional) - T2: New weights
__group_130__new_gradients (optional) - T3: New gradients
__group_130__new_moment_1 (optional) - T4: New averaged gradients
__group_130__new_moment_2 (optional) - T4: New averaged squared gradients
__group_130__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_131__new_weights (optional) - T2: New weights
__group_131__new_gradients (optional) - T3: New gradients
__group_131__new_moment_1 (optional) - T4: New averaged gradients
__group_131__new_moment_2 (optional) - T4: New averaged squared gradients
__group_131__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_132__new_weights (optional) - T2: New weights
__group_132__new_gradients (optional) - T3: New gradients
__group_132__new_moment_1 (optional) - T4: New averaged gradients
__group_132__new_moment_2 (optional) - T4: New averaged squared gradients
__group_132__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_133__new_weights (optional) - T2: New weights
__group_133__new_gradients (optional) - T3: New gradients
__group_133__new_moment_1 (optional) - T4: New averaged gradients
__group_133__new_moment_2 (optional) - T4: New averaged squared gradients
__group_133__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_134__new_weights (optional) - T2: New weights
__group_134__new_gradients (optional) - T3: New gradients
__group_134__new_moment_1 (optional) - T4: New averaged gradients
__group_134__new_moment_2 (optional) - T4: New averaged squared gradients
__group_134__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_135__new_weights (optional) - T2: New weights
__group_135__new_gradients (optional) - T3: New gradients
__group_135__new_moment_1 (optional) - T4: New averaged gradients
__group_135__new_moment_2 (optional) - T4: New averaged squared gradients
__group_135__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_136__new_weights (optional) - T2: New weights
__group_136__new_gradients (optional) - T3: New gradients
__group_136__new_moment_1 (optional) - T4: New averaged gradients
__group_136__new_moment_2 (optional) - T4: New averaged squared gradients
__group_136__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_137__new_weights (optional) - T2: New weights
__group_137__new_gradients (optional) - T3: New gradients
__group_137__new_moment_1 (optional) - T4: New averaged gradients
__group_137__new_moment_2 (optional) - T4: New averaged squared gradients
__group_137__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_138__new_weights (optional) - T2: New weights
__group_138__new_gradients (optional) - T3: New gradients
__group_138__new_moment_1 (optional) - T4: New averaged gradients
__group_138__new_moment_2 (optional) - T4: New averaged squared gradients
__group_138__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_139__new_weights (optional) - T2: New weights
__group_139__new_gradients (optional) - T3: New gradients
__group_139__new_moment_1 (optional) - T4: New averaged gradients
__group_139__new_moment_2 (optional) - T4: New averaged squared gradients
__group_139__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_140__new_weights (optional) - T2: New weights
__group_140__new_gradients (optional) - T3: New gradients
__group_140__new_moment_1 (optional) - T4: New averaged gradients
__group_140__new_moment_2 (optional) - T4: New averaged squared gradients
__group_140__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_141__new_weights (optional) - T2: New weights
__group_141__new_gradients (optional) - T3: New gradients
__group_141__new_moment_1 (optional) - T4: New averaged gradients
__group_141__new_moment_2 (optional) - T4: New averaged squared gradients
__group_141__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_142__new_weights (optional) - T2: New weights
__group_142__new_gradients (optional) - T3: New gradients
__group_142__new_moment_1 (optional) - T4: New averaged gradients
__group_142__new_moment_2 (optional) - T4: New averaged squared gradients
__group_142__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_143__new_weights (optional) - T2: New weights
__group_143__new_gradients (optional) - T3: New gradients
__group_143__new_moment_1 (optional) - T4: New averaged gradients
__group_143__new_moment_2 (optional) - T4: New averaged squared gradients
__group_143__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_144__new_weights (optional) - T2: New weights
__group_144__new_gradients (optional) - T3: New gradients
__group_144__new_moment_1 (optional) - T4: New averaged gradients
__group_144__new_moment_2 (optional) - T4: New averaged squared gradients
__group_144__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_145__new_weights (optional) - T2: New weights
__group_145__new_gradients (optional) - T3: New gradients
__group_145__new_moment_1 (optional) - T4: New averaged gradients
__group_145__new_moment_2 (optional) - T4: New averaged squared gradients
__group_145__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_146__new_weights (optional) - T2: New weights
__group_146__new_gradients (optional) - T3: New gradients
__group_146__new_moment_1 (optional) - T4: New averaged gradients
__group_146__new_moment_2 (optional) - T4: New averaged squared gradients
__group_146__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_147__new_weights (optional) - T2: New weights
__group_147__new_gradients (optional) - T3: New gradients
__group_147__new_moment_1 (optional) - T4: New averaged gradients
__group_147__new_moment_2 (optional) - T4: New averaged squared gradients
__group_147__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_148__new_weights (optional) - T2: New weights
__group_148__new_gradients (optional) - T3: New gradients
__group_148__new_moment_1 (optional) - T4: New averaged gradients
__group_148__new_moment_2 (optional) - T4: New averaged squared gradients
__group_148__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_149__new_weights (optional) - T2: New weights
__group_149__new_gradients (optional) - T3: New gradients
__group_149__new_moment_1 (optional) - T4: New averaged gradients
__group_149__new_moment_2 (optional) - T4: New averaged squared gradients
__group_149__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_150__new_weights (optional) - T2: New weights
__group_150__new_gradients (optional) - T3: New gradients
__group_150__new_moment_1 (optional) - T4: New averaged gradients
__group_150__new_moment_2 (optional) - T4: New averaged squared gradients
__group_150__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_151__new_weights (optional) - T2: New weights
__group_151__new_gradients (optional) - T3: New gradients
__group_151__new_moment_1 (optional) - T4: New averaged gradients
__group_151__new_moment_2 (optional) - T4: New averaged squared gradients
__group_151__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_152__new_weights (optional) - T2: New weights
__group_152__new_gradients (optional) - T3: New gradients
__group_152__new_moment_1 (optional) - T4: New averaged gradients
__group_152__new_moment_2 (optional) - T4: New averaged squared gradients
__group_152__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_153__new_weights (optional) - T2: New weights
__group_153__new_gradients (optional) - T3: New gradients
__group_153__new_moment_1 (optional) - T4: New averaged gradients
__group_153__new_moment_2 (optional) - T4: New averaged squared gradients
__group_153__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_154__new_weights (optional) - T2: New weights
__group_154__new_gradients (optional) - T3: New gradients
__group_154__new_moment_1 (optional) - T4: New averaged gradients
__group_154__new_moment_2 (optional) - T4: New averaged squared gradients
__group_154__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_155__new_weights (optional) - T2: New weights
__group_155__new_gradients (optional) - T3: New gradients
__group_155__new_moment_1 (optional) - T4: New averaged gradients
__group_155__new_moment_2 (optional) - T4: New averaged squared gradients
__group_155__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_156__new_weights (optional) - T2: New weights
__group_156__new_gradients (optional) - T3: New gradients
__group_156__new_moment_1 (optional) - T4: New averaged gradients
__group_156__new_moment_2 (optional) - T4: New averaged squared gradients
__group_156__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_157__new_weights (optional) - T2: New weights
__group_157__new_gradients (optional) - T3: New gradients
__group_157__new_moment_1 (optional) - T4: New averaged gradients
__group_157__new_moment_2 (optional) - T4: New averaged squared gradients
__group_157__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_158__new_weights (optional) - T2: New weights
__group_158__new_gradients (optional) - T3: New gradients
__group_158__new_moment_1 (optional) - T4: New averaged gradients
__group_158__new_moment_2 (optional) - T4: New averaged squared gradients
__group_158__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_159__new_weights (optional) - T2: New weights
__group_159__new_gradients (optional) - T3: New gradients
__group_159__new_moment_1 (optional) - T4: New averaged gradients
__group_159__new_moment_2 (optional) - T4: New averaged squared gradients
__group_159__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_160__new_weights (optional) - T2: New weights
__group_160__new_gradients (optional) - T3: New gradients
__group_160__new_moment_1 (optional) - T4: New averaged gradients
__group_160__new_moment_2 (optional) - T4: New averaged squared gradients
__group_160__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_161__new_weights (optional) - T2: New weights
__group_161__new_gradients (optional) - T3: New gradients
__group_161__new_moment_1 (optional) - T4: New averaged gradients
__group_161__new_moment_2 (optional) - T4: New averaged squared gradients
__group_161__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_162__new_weights (optional) - T2: New weights
__group_162__new_gradients (optional) - T3: New gradients
__group_162__new_moment_1 (optional) - T4: New averaged gradients
__group_162__new_moment_2 (optional) - T4: New averaged squared gradients
__group_162__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_163__new_weights (optional) - T2: New weights
__group_163__new_gradients (optional) - T3: New gradients
__group_163__new_moment_1 (optional) - T4: New averaged gradients
__group_163__new_moment_2 (optional) - T4: New averaged squared gradients
__group_163__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_164__new_weights (optional) - T2: New weights
__group_164__new_gradients (optional) - T3: New gradients
__group_164__new_moment_1 (optional) - T4: New averaged gradients
__group_164__new_moment_2 (optional) - T4: New averaged squared gradients
__group_164__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_165__new_weights (optional) - T2: New weights
__group_165__new_gradients (optional) - T3: New gradients
__group_165__new_moment_1 (optional) - T4: New averaged gradients
__group_165__new_moment_2 (optional) - T4: New averaged squared gradients
__group_165__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_166__new_weights (optional) - T2: New weights
__group_166__new_gradients (optional) - T3: New gradients
__group_166__new_moment_1 (optional) - T4: New averaged gradients
__group_166__new_moment_2 (optional) - T4: New averaged squared gradients
__group_166__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_167__new_weights (optional) - T2: New weights
__group_167__new_gradients (optional) - T3: New gradients
__group_167__new_moment_1 (optional) - T4: New averaged gradients
__group_167__new_moment_2 (optional) - T4: New averaged squared gradients
__group_167__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_168__new_weights (optional) - T2: New weights
__group_168__new_gradients (optional) - T3: New gradients
__group_168__new_moment_1 (optional) - T4: New averaged gradients
__group_168__new_moment_2 (optional) - T4: New averaged squared gradients
__group_168__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_169__new_weights (optional) - T2: New weights
__group_169__new_gradients (optional) - T3: New gradients
__group_169__new_moment_1 (optional) - T4: New averaged gradients
__group_169__new_moment_2 (optional) - T4: New averaged squared gradients
__group_169__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_170__new_weights (optional) - T2: New weights
__group_170__new_gradients (optional) - T3: New gradients
__group_170__new_moment_1 (optional) - T4: New averaged gradients
__group_170__new_moment_2 (optional) - T4: New averaged squared gradients
__group_170__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_171__new_weights (optional) - T2: New weights
__group_171__new_gradients (optional) - T3: New gradients
__group_171__new_moment_1 (optional) - T4: New averaged gradients
__group_171__new_moment_2 (optional) - T4: New averaged squared gradients
__group_171__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_172__new_weights (optional) - T2: New weights
__group_172__new_gradients (optional) - T3: New gradients
__group_172__new_moment_1 (optional) - T4: New averaged gradients
__group_172__new_moment_2 (optional) - T4: New averaged squared gradients
__group_172__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_173__new_weights (optional) - T2: New weights
__group_173__new_gradients (optional) - T3: New gradients
__group_173__new_moment_1 (optional) - T4: New averaged gradients
__group_173__new_moment_2 (optional) - T4: New averaged squared gradients
__group_173__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_174__new_weights (optional) - T2: New weights
__group_174__new_gradients (optional) - T3: New gradients
__group_174__new_moment_1 (optional) - T4: New averaged gradients
__group_174__new_moment_2 (optional) - T4: New averaged squared gradients
__group_174__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_175__new_weights (optional) - T2: New weights
__group_175__new_gradients (optional) - T3: New gradients
__group_175__new_moment_1 (optional) - T4: New averaged gradients
__group_175__new_moment_2 (optional) - T4: New averaged squared gradients
__group_175__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_176__new_weights (optional) - T2: New weights
__group_176__new_gradients (optional) - T3: New gradients
__group_176__new_moment_1 (optional) - T4: New averaged gradients
__group_176__new_moment_2 (optional) - T4: New averaged squared gradients
__group_176__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_177__new_weights (optional) - T2: New weights
__group_177__new_gradients (optional) - T3: New gradients
__group_177__new_moment_1 (optional) - T4: New averaged gradients
__group_177__new_moment_2 (optional) - T4: New averaged squared gradients
__group_177__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_178__new_weights (optional) - T2: New weights
__group_178__new_gradients (optional) - T3: New gradients
__group_178__new_moment_1 (optional) - T4: New averaged gradients
__group_178__new_moment_2 (optional) - T4: New averaged squared gradients
__group_178__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_179__new_weights (optional) - T2: New weights
__group_179__new_gradients (optional) - T3: New gradients
__group_179__new_moment_1 (optional) - T4: New averaged gradients
__group_179__new_moment_2 (optional) - T4: New averaged squared gradients
__group_179__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_180__new_weights (optional) - T2: New weights
__group_180__new_gradients (optional) - T3: New gradients
__group_180__new_moment_1 (optional) - T4: New averaged gradients
__group_180__new_moment_2 (optional) - T4: New averaged squared gradients
__group_180__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_181__new_weights (optional) - T2: New weights
__group_181__new_gradients (optional) - T3: New gradients
__group_181__new_moment_1 (optional) - T4: New averaged gradients
__group_181__new_moment_2 (optional) - T4: New averaged squared gradients
__group_181__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_182__new_weights (optional) - T2: New weights
__group_182__new_gradients (optional) - T3: New gradients
__group_182__new_moment_1 (optional) - T4: New averaged gradients
__group_182__new_moment_2 (optional) - T4: New averaged squared gradients
__group_182__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_183__new_weights (optional) - T2: New weights
__group_183__new_gradients (optional) - T3: New gradients
__group_183__new_moment_1 (optional) - T4: New averaged gradients
__group_183__new_moment_2 (optional) - T4: New averaged squared gradients
__group_183__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_184__new_weights (optional) - T2: New weights
__group_184__new_gradients (optional) - T3: New gradients
__group_184__new_moment_1 (optional) - T4: New averaged gradients
__group_184__new_moment_2 (optional) - T4: New averaged squared gradients
__group_184__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_185__new_weights (optional) - T2: New weights
__group_185__new_gradients (optional) - T3: New gradients
__group_185__new_moment_1 (optional) - T4: New averaged gradients
__group_185__new_moment_2 (optional) - T4: New averaged squared gradients
__group_185__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_186__new_weights (optional) - T2: New weights
__group_186__new_gradients (optional) - T3: New gradients
__group_186__new_moment_1 (optional) - T4: New averaged gradients
__group_186__new_moment_2 (optional) - T4: New averaged squared gradients
__group_186__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_187__new_weights (optional) - T2: New weights
__group_187__new_gradients (optional) - T3: New gradients
__group_187__new_moment_1 (optional) - T4: New averaged gradients
__group_187__new_moment_2 (optional) - T4: New averaged squared gradients
__group_187__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_188__new_weights (optional) - T2: New weights
__group_188__new_gradients (optional) - T3: New gradients
__group_188__new_moment_1 (optional) - T4: New averaged gradients
__group_188__new_moment_2 (optional) - T4: New averaged squared gradients
__group_188__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_189__new_weights (optional) - T2: New weights
__group_189__new_gradients (optional) - T3: New gradients
__group_189__new_moment_1 (optional) - T4: New averaged gradients
__group_189__new_moment_2 (optional) - T4: New averaged squared gradients
__group_189__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_190__new_weights (optional) - T2: New weights
__group_190__new_gradients (optional) - T3: New gradients
__group_190__new_moment_1 (optional) - T4: New averaged gradients
__group_190__new_moment_2 (optional) - T4: New averaged squared gradients
__group_190__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_191__new_weights (optional) - T2: New weights
__group_191__new_gradients (optional) - T3: New gradients
__group_191__new_moment_1 (optional) - T4: New averaged gradients
__group_191__new_moment_2 (optional) - T4: New averaged squared gradients
__group_191__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_192__new_weights (optional) - T2: New weights
__group_192__new_gradients (optional) - T3: New gradients
__group_192__new_moment_1 (optional) - T4: New averaged gradients
__group_192__new_moment_2 (optional) - T4: New averaged squared gradients
__group_192__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_193__new_weights (optional) - T2: New weights
__group_193__new_gradients (optional) - T3: New gradients
__group_193__new_moment_1 (optional) - T4: New averaged gradients
__group_193__new_moment_2 (optional) - T4: New averaged squared gradients
__group_193__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_194__new_weights (optional) - T2: New weights
__group_194__new_gradients (optional) - T3: New gradients
__group_194__new_moment_1 (optional) - T4: New averaged gradients
__group_194__new_moment_2 (optional) - T4: New averaged squared gradients
__group_194__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_195__new_weights (optional) - T2: New weights
__group_195__new_gradients (optional) - T3: New gradients
__group_195__new_moment_1 (optional) - T4: New averaged gradients
__group_195__new_moment_2 (optional) - T4: New averaged squared gradients
__group_195__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_196__new_weights (optional) - T2: New weights
__group_196__new_gradients (optional) - T3: New gradients
__group_196__new_moment_1 (optional) - T4: New averaged gradients
__group_196__new_moment_2 (optional) - T4: New averaged squared gradients
__group_196__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_197__new_weights (optional) - T2: New weights
__group_197__new_gradients (optional) - T3: New gradients
__group_197__new_moment_1 (optional) - T4: New averaged gradients
__group_197__new_moment_2 (optional) - T4: New averaged squared gradients
__group_197__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_198__new_weights (optional) - T2: New weights
__group_198__new_gradients (optional) - T3: New gradients
__group_198__new_moment_1 (optional) - T4: New averaged gradients
__group_198__new_moment_2 (optional) - T4: New averaged squared gradients
__group_198__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_199__new_weights (optional) - T2: New weights
__group_199__new_gradients (optional) - T3: New gradients
__group_199__new_moment_1 (optional) - T4: New averaged gradients
__group_199__new_moment_2 (optional) - T4: New averaged squared gradients
__group_199__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_200__new_weights (optional) - T2: New weights
__group_200__new_gradients (optional) - T3: New gradients
__group_200__new_moment_1 (optional) - T4: New averaged gradients
__group_200__new_moment_2 (optional) - T4: New averaged squared gradients
__group_200__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_201__new_weights (optional) - T2: New weights
__group_201__new_gradients (optional) - T3: New gradients
__group_201__new_moment_1 (optional) - T4: New averaged gradients
__group_201__new_moment_2 (optional) - T4: New averaged squared gradients
__group_201__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_202__new_weights (optional) - T2: New weights
__group_202__new_gradients (optional) - T3: New gradients
__group_202__new_moment_1 (optional) - T4: New averaged gradients
__group_202__new_moment_2 (optional) - T4: New averaged squared gradients
__group_202__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_203__new_weights (optional) - T2: New weights
__group_203__new_gradients (optional) - T3: New gradients
__group_203__new_moment_1 (optional) - T4: New averaged gradients
__group_203__new_moment_2 (optional) - T4: New averaged squared gradients
__group_203__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_204__new_weights (optional) - T2: New weights
__group_204__new_gradients (optional) - T3: New gradients
__group_204__new_moment_1 (optional) - T4: New averaged gradients
__group_204__new_moment_2 (optional) - T4: New averaged squared gradients
__group_204__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_205__new_weights (optional) - T2: New weights
__group_205__new_gradients (optional) - T3: New gradients
__group_205__new_moment_1 (optional) - T4: New averaged gradients
__group_205__new_moment_2 (optional) - T4: New averaged squared gradients
__group_205__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_206__new_weights (optional) - T2: New weights
__group_206__new_gradients (optional) - T3: New gradients
__group_206__new_moment_1 (optional) - T4: New averaged gradients
__group_206__new_moment_2 (optional) - T4: New averaged squared gradients
__group_206__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_207__new_weights (optional) - T2: New weights
__group_207__new_gradients (optional) - T3: New gradients
__group_207__new_moment_1 (optional) - T4: New averaged gradients
__group_207__new_moment_2 (optional) - T4: New averaged squared gradients
__group_207__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_208__new_weights (optional) - T2: New weights
__group_208__new_gradients (optional) - T3: New gradients
__group_208__new_moment_1 (optional) - T4: New averaged gradients
__group_208__new_moment_2 (optional) - T4: New averaged squared gradients
__group_208__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_209__new_weights (optional) - T2: New weights
__group_209__new_gradients (optional) - T3: New gradients
__group_209__new_moment_1 (optional) - T4: New averaged gradients
__group_209__new_moment_2 (optional) - T4: New averaged squared gradients
__group_209__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_210__new_weights (optional) - T2: New weights
__group_210__new_gradients (optional) - T3: New gradients
__group_210__new_moment_1 (optional) - T4: New averaged gradients
__group_210__new_moment_2 (optional) - T4: New averaged squared gradients
__group_210__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_211__new_weights (optional) - T2: New weights
__group_211__new_gradients (optional) - T3: New gradients
__group_211__new_moment_1 (optional) - T4: New averaged gradients
__group_211__new_moment_2 (optional) - T4: New averaged squared gradients
__group_211__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_212__new_weights (optional) - T2: New weights
__group_212__new_gradients (optional) - T3: New gradients
__group_212__new_moment_1 (optional) - T4: New averaged gradients
__group_212__new_moment_2 (optional) - T4: New averaged squared gradients
__group_212__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_213__new_weights (optional) - T2: New weights
__group_213__new_gradients (optional) - T3: New gradients
__group_213__new_moment_1 (optional) - T4: New averaged gradients
__group_213__new_moment_2 (optional) - T4: New averaged squared gradients
__group_213__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_214__new_weights (optional) - T2: New weights
__group_214__new_gradients (optional) - T3: New gradients
__group_214__new_moment_1 (optional) - T4: New averaged gradients
__group_214__new_moment_2 (optional) - T4: New averaged squared gradients
__group_214__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_215__new_weights (optional) - T2: New weights
__group_215__new_gradients (optional) - T3: New gradients
__group_215__new_moment_1 (optional) - T4: New averaged gradients
__group_215__new_moment_2 (optional) - T4: New averaged squared gradients
__group_215__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_216__new_weights (optional) - T2: New weights
__group_216__new_gradients (optional) - T3: New gradients
__group_216__new_moment_1 (optional) - T4: New averaged gradients
__group_216__new_moment_2 (optional) - T4: New averaged squared gradients
__group_216__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_217__new_weights (optional) - T2: New weights
__group_217__new_gradients (optional) - T3: New gradients
__group_217__new_moment_1 (optional) - T4: New averaged gradients
__group_217__new_moment_2 (optional) - T4: New averaged squared gradients
__group_217__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_218__new_weights (optional) - T2: New weights
__group_218__new_gradients (optional) - T3: New gradients
__group_218__new_moment_1 (optional) - T4: New averaged gradients
__group_218__new_moment_2 (optional) - T4: New averaged squared gradients
__group_218__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_219__new_weights (optional) - T2: New weights
__group_219__new_gradients (optional) - T3: New gradients
__group_219__new_moment_1 (optional) - T4: New averaged gradients
__group_219__new_moment_2 (optional) - T4: New averaged squared gradients
__group_219__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_220__new_weights (optional) - T2: New weights
__group_220__new_gradients (optional) - T3: New gradients
__group_220__new_moment_1 (optional) - T4: New averaged gradients
__group_220__new_moment_2 (optional) - T4: New averaged squared gradients
__group_220__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_221__new_weights (optional) - T2: New weights
__group_221__new_gradients (optional) - T3: New gradients
__group_221__new_moment_1 (optional) - T4: New averaged gradients
__group_221__new_moment_2 (optional) - T4: New averaged squared gradients
__group_221__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_222__new_weights (optional) - T2: New weights
__group_222__new_gradients (optional) - T3: New gradients
__group_222__new_moment_1 (optional) - T4: New averaged gradients
__group_222__new_moment_2 (optional) - T4: New averaged squared gradients
__group_222__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_223__new_weights (optional) - T2: New weights
__group_223__new_gradients (optional) - T3: New gradients
__group_223__new_moment_1 (optional) - T4: New averaged gradients
__group_223__new_moment_2 (optional) - T4: New averaged squared gradients
__group_223__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_224__new_weights (optional) - T2: New weights
__group_224__new_gradients (optional) - T3: New gradients
__group_224__new_moment_1 (optional) - T4: New averaged gradients
__group_224__new_moment_2 (optional) - T4: New averaged squared gradients
__group_224__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_225__new_weights (optional) - T2: New weights
__group_225__new_gradients (optional) - T3: New gradients
__group_225__new_moment_1 (optional) - T4: New averaged gradients
__group_225__new_moment_2 (optional) - T4: New averaged squared gradients
__group_225__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_226__new_weights (optional) - T2: New weights
__group_226__new_gradients (optional) - T3: New gradients
__group_226__new_moment_1 (optional) - T4: New averaged gradients
__group_226__new_moment_2 (optional) - T4: New averaged squared gradients
__group_226__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_227__new_weights (optional) - T2: New weights
__group_227__new_gradients (optional) - T3: New gradients
__group_227__new_moment_1 (optional) - T4: New averaged gradients
__group_227__new_moment_2 (optional) - T4: New averaged squared gradients
__group_227__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_228__new_weights (optional) - T2: New weights
__group_228__new_gradients (optional) - T3: New gradients
__group_228__new_moment_1 (optional) - T4: New averaged gradients
__group_228__new_moment_2 (optional) - T4: New averaged squared gradients
__group_228__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_229__new_weights (optional) - T2: New weights
__group_229__new_gradients (optional) - T3: New gradients
__group_229__new_moment_1 (optional) - T4: New averaged gradients
__group_229__new_moment_2 (optional) - T4: New averaged squared gradients
__group_229__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_230__new_weights (optional) - T2: New weights
__group_230__new_gradients (optional) - T3: New gradients
__group_230__new_moment_1 (optional) - T4: New averaged gradients
__group_230__new_moment_2 (optional) - T4: New averaged squared gradients
__group_230__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_231__new_weights (optional) - T2: New weights
__group_231__new_gradients (optional) - T3: New gradients
__group_231__new_moment_1 (optional) - T4: New averaged gradients
__group_231__new_moment_2 (optional) - T4: New averaged squared gradients
__group_231__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_232__new_weights (optional) - T2: New weights
__group_232__new_gradients (optional) - T3: New gradients
__group_232__new_moment_1 (optional) - T4: New averaged gradients
__group_232__new_moment_2 (optional) - T4: New averaged squared gradients
__group_232__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_233__new_weights (optional) - T2: New weights
__group_233__new_gradients (optional) - T3: New gradients
__group_233__new_moment_1 (optional) - T4: New averaged gradients
__group_233__new_moment_2 (optional) - T4: New averaged squared gradients
__group_233__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_234__new_weights (optional) - T2: New weights
__group_234__new_gradients (optional) - T3: New gradients
__group_234__new_moment_1 (optional) - T4: New averaged gradients
__group_234__new_moment_2 (optional) - T4: New averaged squared gradients
__group_234__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_235__new_weights (optional) - T2: New weights
__group_235__new_gradients (optional) - T3: New gradients
__group_235__new_moment_1 (optional) - T4: New averaged gradients
__group_235__new_moment_2 (optional) - T4: New averaged squared gradients
__group_235__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_236__new_weights (optional) - T2: New weights
__group_236__new_gradients (optional) - T3: New gradients
__group_236__new_moment_1 (optional) - T4: New averaged gradients
__group_236__new_moment_2 (optional) - T4: New averaged squared gradients
__group_236__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_237__new_weights (optional) - T2: New weights
__group_237__new_gradients (optional) - T3: New gradients
__group_237__new_moment_1 (optional) - T4: New averaged gradients
__group_237__new_moment_2 (optional) - T4: New averaged squared gradients
__group_237__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_238__new_weights (optional) - T2: New weights
__group_238__new_gradients (optional) - T3: New gradients
__group_238__new_moment_1 (optional) - T4: New averaged gradients
__group_238__new_moment_2 (optional) - T4: New averaged squared gradients
__group_238__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_239__new_weights (optional) - T2: New weights
__group_239__new_gradients (optional) - T3: New gradients
__group_239__new_moment_1 (optional) - T4: New averaged gradients
__group_239__new_moment_2 (optional) - T4: New averaged squared gradients
__group_239__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_240__new_weights (optional) - T2: New weights
__group_240__new_gradients (optional) - T3: New gradients
__group_240__new_moment_1 (optional) - T4: New averaged gradients
__group_240__new_moment_2 (optional) - T4: New averaged squared gradients
__group_240__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_241__new_weights (optional) - T2: New weights
__group_241__new_gradients (optional) - T3: New gradients
__group_241__new_moment_1 (optional) - T4: New averaged gradients
__group_241__new_moment_2 (optional) - T4: New averaged squared gradients
__group_241__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_242__new_weights (optional) - T2: New weights
__group_242__new_gradients (optional) - T3: New gradients
__group_242__new_moment_1 (optional) - T4: New averaged gradients
__group_242__new_moment_2 (optional) - T4: New averaged squared gradients
__group_242__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_243__new_weights (optional) - T2: New weights
__group_243__new_gradients (optional) - T3: New gradients
__group_243__new_moment_1 (optional) - T4: New averaged gradients
__group_243__new_moment_2 (optional) - T4: New averaged squared gradients
__group_243__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_244__new_weights (optional) - T2: New weights
__group_244__new_gradients (optional) - T3: New gradients
__group_244__new_moment_1 (optional) - T4: New averaged gradients
__group_244__new_moment_2 (optional) - T4: New averaged squared gradients
__group_244__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_245__new_weights (optional) - T2: New weights
__group_245__new_gradients (optional) - T3: New gradients
__group_245__new_moment_1 (optional) - T4: New averaged gradients
__group_245__new_moment_2 (optional) - T4: New averaged squared gradients
__group_245__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_246__new_weights (optional) - T2: New weights
__group_246__new_gradients (optional) - T3: New gradients
__group_246__new_moment_1 (optional) - T4: New averaged gradients
__group_246__new_moment_2 (optional) - T4: New averaged squared gradients
__group_246__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_247__new_weights (optional) - T2: New weights
__group_247__new_gradients (optional) - T3: New gradients
__group_247__new_moment_1 (optional) - T4: New averaged gradients
__group_247__new_moment_2 (optional) - T4: New averaged squared gradients
__group_247__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_248__new_weights (optional) - T2: New weights
__group_248__new_gradients (optional) - T3: New gradients
__group_248__new_moment_1 (optional) - T4: New averaged gradients
__group_248__new_moment_2 (optional) - T4: New averaged squared gradients
__group_248__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_249__new_weights (optional) - T2: New weights
__group_249__new_gradients (optional) - T3: New gradients
__group_249__new_moment_1 (optional) - T4: New averaged gradients
__group_249__new_moment_2 (optional) - T4: New averaged squared gradients
__group_249__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_250__new_weights (optional) - T2: New weights
__group_250__new_gradients (optional) - T3: New gradients
__group_250__new_moment_1 (optional) - T4: New averaged gradients
__group_250__new_moment_2 (optional) - T4: New averaged squared gradients
__group_250__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_251__new_weights (optional) - T2: New weights
__group_251__new_gradients (optional) - T3: New gradients
__group_251__new_moment_1 (optional) - T4: New averaged gradients
__group_251__new_moment_2 (optional) - T4: New averaged squared gradients
__group_251__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_252__new_weights (optional) - T2: New weights
__group_252__new_gradients (optional) - T3: New gradients
__group_252__new_moment_1 (optional) - T4: New averaged gradients
__group_252__new_moment_2 (optional) - T4: New averaged squared gradients
__group_252__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_253__new_weights (optional) - T2: New weights
__group_253__new_gradients (optional) - T3: New gradients
__group_253__new_moment_1 (optional) - T4: New averaged gradients
__group_253__new_moment_2 (optional) - T4: New averaged squared gradients
__group_253__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_254__new_weights (optional) - T2: New weights
__group_254__new_gradients (optional) - T3: New gradients
__group_254__new_moment_1 (optional) - T4: New averaged gradients
__group_254__new_moment_2 (optional) - T4: New averaged squared gradients
__group_254__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_255__new_weights (optional) - T2: New weights
__group_255__new_gradients (optional) - T3: New gradients
__group_255__new_moment_1 (optional) - T4: New averaged gradients
__group_255__new_moment_2 (optional) - T4: New averaged squared gradients
__group_255__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_256__new_weights (optional) - T2: New weights
__group_256__new_gradients (optional) - T3: New gradients
__group_256__new_moment_1 (optional) - T4: New averaged gradients
__group_256__new_moment_2 (optional) - T4: New averaged squared gradients
__group_256__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_257__new_weights (optional) - T2: New weights
__group_257__new_gradients (optional) - T3: New gradients
__group_257__new_moment_1 (optional) - T4: New averaged gradients
__group_257__new_moment_2 (optional) - T4: New averaged squared gradients
__group_257__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_258__new_weights (optional) - T2: New weights
__group_258__new_gradients (optional) - T3: New gradients
__group_258__new_moment_1 (optional) - T4: New averaged gradients
__group_258__new_moment_2 (optional) - T4: New averaged squared gradients
__group_258__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_259__new_weights (optional) - T2: New weights
__group_259__new_gradients (optional) - T3: New gradients
__group_259__new_moment_1 (optional) - T4: New averaged gradients
__group_259__new_moment_2 (optional) - T4: New averaged squared gradients
__group_259__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_260__new_weights (optional) - T2: New weights
__group_260__new_gradients (optional) - T3: New gradients
__group_260__new_moment_1 (optional) - T4: New averaged gradients
__group_260__new_moment_2 (optional) - T4: New averaged squared gradients
__group_260__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_261__new_weights (optional) - T2: New weights
__group_261__new_gradients (optional) - T3: New gradients
__group_261__new_moment_1 (optional) - T4: New averaged gradients
__group_261__new_moment_2 (optional) - T4: New averaged squared gradients
__group_261__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_262__new_weights (optional) - T2: New weights
__group_262__new_gradients (optional) - T3: New gradients
__group_262__new_moment_1 (optional) - T4: New averaged gradients
__group_262__new_moment_2 (optional) - T4: New averaged squared gradients
__group_262__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_263__new_weights (optional) - T2: New weights
__group_263__new_gradients (optional) - T3: New gradients
__group_263__new_moment_1 (optional) - T4: New averaged gradients
__group_263__new_moment_2 (optional) - T4: New averaged squared gradients
__group_263__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_264__new_weights (optional) - T2: New weights
__group_264__new_gradients (optional) - T3: New gradients
__group_264__new_moment_1 (optional) - T4: New averaged gradients
__group_264__new_moment_2 (optional) - T4: New averaged squared gradients
__group_264__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_265__new_weights (optional) - T2: New weights
__group_265__new_gradients (optional) - T3: New gradients
__group_265__new_moment_1 (optional) - T4: New averaged gradients
__group_265__new_moment_2 (optional) - T4: New averaged squared gradients
__group_265__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_266__new_weights (optional) - T2: New weights
__group_266__new_gradients (optional) - T3: New gradients
__group_266__new_moment_1 (optional) - T4: New averaged gradients
__group_266__new_moment_2 (optional) - T4: New averaged squared gradients
__group_266__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_267__new_weights (optional) - T2: New weights
__group_267__new_gradients (optional) - T3: New gradients
__group_267__new_moment_1 (optional) - T4: New averaged gradients
__group_267__new_moment_2 (optional) - T4: New averaged squared gradients
__group_267__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_268__new_weights (optional) - T2: New weights
__group_268__new_gradients (optional) - T3: New gradients
__group_268__new_moment_1 (optional) - T4: New averaged gradients
__group_268__new_moment_2 (optional) - T4: New averaged squared gradients
__group_268__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_269__new_weights (optional) - T2: New weights
__group_269__new_gradients (optional) - T3: New gradients
__group_269__new_moment_1 (optional) - T4: New averaged gradients
__group_269__new_moment_2 (optional) - T4: New averaged squared gradients
__group_269__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_270__new_weights (optional) - T2: New weights
__group_270__new_gradients (optional) - T3: New gradients
__group_270__new_moment_1 (optional) - T4: New averaged gradients
__group_270__new_moment_2 (optional) - T4: New averaged squared gradients
__group_270__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_271__new_weights (optional) - T2: New weights
__group_271__new_gradients (optional) - T3: New gradients
__group_271__new_moment_1 (optional) - T4: New averaged gradients
__group_271__new_moment_2 (optional) - T4: New averaged squared gradients
__group_271__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_272__new_weights (optional) - T2: New weights
__group_272__new_gradients (optional) - T3: New gradients
__group_272__new_moment_1 (optional) - T4: New averaged gradients
__group_272__new_moment_2 (optional) - T4: New averaged squared gradients
__group_272__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_273__new_weights (optional) - T2: New weights
__group_273__new_gradients (optional) - T3: New gradients
__group_273__new_moment_1 (optional) - T4: New averaged gradients
__group_273__new_moment_2 (optional) - T4: New averaged squared gradients
__group_273__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_274__new_weights (optional) - T2: New weights
__group_274__new_gradients (optional) - T3: New gradients
__group_274__new_moment_1 (optional) - T4: New averaged gradients
__group_274__new_moment_2 (optional) - T4: New averaged squared gradients
__group_274__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_275__new_weights (optional) - T2: New weights
__group_275__new_gradients (optional) - T3: New gradients
__group_275__new_moment_1 (optional) - T4: New averaged gradients
__group_275__new_moment_2 (optional) - T4: New averaged squared gradients
__group_275__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_276__new_weights (optional) - T2: New weights
__group_276__new_gradients (optional) - T3: New gradients
__group_276__new_moment_1 (optional) - T4: New averaged gradients
__group_276__new_moment_2 (optional) - T4: New averaged squared gradients
__group_276__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_277__new_weights (optional) - T2: New weights
__group_277__new_gradients (optional) - T3: New gradients
__group_277__new_moment_1 (optional) - T4: New averaged gradients
__group_277__new_moment_2 (optional) - T4: New averaged squared gradients
__group_277__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_278__new_weights (optional) - T2: New weights
__group_278__new_gradients (optional) - T3: New gradients
__group_278__new_moment_1 (optional) - T4: New averaged gradients
__group_278__new_moment_2 (optional) - T4: New averaged squared gradients
__group_278__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_279__new_weights (optional) - T2: New weights
__group_279__new_gradients (optional) - T3: New gradients
__group_279__new_moment_1 (optional) - T4: New averaged gradients
__group_279__new_moment_2 (optional) - T4: New averaged squared gradients
__group_279__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_280__new_weights (optional) - T2: New weights
__group_280__new_gradients (optional) - T3: New gradients
__group_280__new_moment_1 (optional) - T4: New averaged gradients
__group_280__new_moment_2 (optional) - T4: New averaged squared gradients
__group_280__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_281__new_weights (optional) - T2: New weights
__group_281__new_gradients (optional) - T3: New gradients
__group_281__new_moment_1 (optional) - T4: New averaged gradients
__group_281__new_moment_2 (optional) - T4: New averaged squared gradients
__group_281__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_282__new_weights (optional) - T2: New weights
__group_282__new_gradients (optional) - T3: New gradients
__group_282__new_moment_1 (optional) - T4: New averaged gradients
__group_282__new_moment_2 (optional) - T4: New averaged squared gradients
__group_282__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_283__new_weights (optional) - T2: New weights
__group_283__new_gradients (optional) - T3: New gradients
__group_283__new_moment_1 (optional) - T4: New averaged gradients
__group_283__new_moment_2 (optional) - T4: New averaged squared gradients
__group_283__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_284__new_weights (optional) - T2: New weights
__group_284__new_gradients (optional) - T3: New gradients
__group_284__new_moment_1 (optional) - T4: New averaged gradients
__group_284__new_moment_2 (optional) - T4: New averaged squared gradients
__group_284__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_285__new_weights (optional) - T2: New weights
__group_285__new_gradients (optional) - T3: New gradients
__group_285__new_moment_1 (optional) - T4: New averaged gradients
__group_285__new_moment_2 (optional) - T4: New averaged squared gradients
__group_285__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_286__new_weights (optional) - T2: New weights
__group_286__new_gradients (optional) - T3: New gradients
__group_286__new_moment_1 (optional) - T4: New averaged gradients
__group_286__new_moment_2 (optional) - T4: New averaged squared gradients
__group_286__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_287__new_weights (optional) - T2: New weights
__group_287__new_gradients (optional) - T3: New gradients
__group_287__new_moment_1 (optional) - T4: New averaged gradients
__group_287__new_moment_2 (optional) - T4: New averaged squared gradients
__group_287__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_288__new_weights (optional) - T2: New weights
__group_288__new_gradients (optional) - T3: New gradients
__group_288__new_moment_1 (optional) - T4: New averaged gradients
__group_288__new_moment_2 (optional) - T4: New averaged squared gradients
__group_288__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_289__new_weights (optional) - T2: New weights
__group_289__new_gradients (optional) - T3: New gradients
__group_289__new_moment_1 (optional) - T4: New averaged gradients
__group_289__new_moment_2 (optional) - T4: New averaged squared gradients
__group_289__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_290__new_weights (optional) - T2: New weights
__group_290__new_gradients (optional) - T3: New gradients
__group_290__new_moment_1 (optional) - T4: New averaged gradients
__group_290__new_moment_2 (optional) - T4: New averaged squared gradients
__group_290__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_291__new_weights (optional) - T2: New weights
__group_291__new_gradients (optional) - T3: New gradients
__group_291__new_moment_1 (optional) - T4: New averaged gradients
__group_291__new_moment_2 (optional) - T4: New averaged squared gradients
__group_291__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_292__new_weights (optional) - T2: New weights
__group_292__new_gradients (optional) - T3: New gradients
__group_292__new_moment_1 (optional) - T4: New averaged gradients
__group_292__new_moment_2 (optional) - T4: New averaged squared gradients
__group_292__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_293__new_weights (optional) - T2: New weights
__group_293__new_gradients (optional) - T3: New gradients
__group_293__new_moment_1 (optional) - T4: New averaged gradients
__group_293__new_moment_2 (optional) - T4: New averaged squared gradients
__group_293__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_294__new_weights (optional) - T2: New weights
__group_294__new_gradients (optional) - T3: New gradients
__group_294__new_moment_1 (optional) - T4: New averaged gradients
__group_294__new_moment_2 (optional) - T4: New averaged squared gradients
__group_294__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_295__new_weights (optional) - T2: New weights
__group_295__new_gradients (optional) - T3: New gradients
__group_295__new_moment_1 (optional) - T4: New averaged gradients
__group_295__new_moment_2 (optional) - T4: New averaged squared gradients
__group_295__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_296__new_weights (optional) - T2: New weights
__group_296__new_gradients (optional) - T3: New gradients
__group_296__new_moment_1 (optional) - T4: New averaged gradients
__group_296__new_moment_2 (optional) - T4: New averaged squared gradients
__group_296__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_297__new_weights (optional) - T2: New weights
__group_297__new_gradients (optional) - T3: New gradients
__group_297__new_moment_1 (optional) - T4: New averaged gradients
__group_297__new_moment_2 (optional) - T4: New averaged squared gradients
__group_297__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_298__new_weights (optional) - T2: New weights
__group_298__new_gradients (optional) - T3: New gradients
__group_298__new_moment_1 (optional) - T4: New averaged gradients
__group_298__new_moment_2 (optional) - T4: New averaged squared gradients
__group_298__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_299__new_weights (optional) - T2: New weights
__group_299__new_gradients (optional) - T3: New gradients
__group_299__new_moment_1 (optional) - T4: New averaged gradients
__group_299__new_moment_2 (optional) - T4: New averaged squared gradients
__group_299__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_300__new_weights (optional) - T2: New weights
__group_300__new_gradients (optional) - T3: New gradients
__group_300__new_moment_1 (optional) - T4: New averaged gradients
__group_300__new_moment_2 (optional) - T4: New averaged squared gradients
__group_300__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_301__new_weights (optional) - T2: New weights
__group_301__new_gradients (optional) - T3: New gradients
__group_301__new_moment_1 (optional) - T4: New averaged gradients
__group_301__new_moment_2 (optional) - T4: New averaged squared gradients
__group_301__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_302__new_weights (optional) - T2: New weights
__group_302__new_gradients (optional) - T3: New gradients
__group_302__new_moment_1 (optional) - T4: New averaged gradients
__group_302__new_moment_2 (optional) - T4: New averaged squared gradients
__group_302__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_303__new_weights (optional) - T2: New weights
__group_303__new_gradients (optional) - T3: New gradients
__group_303__new_moment_1 (optional) - T4: New averaged gradients
__group_303__new_moment_2 (optional) - T4: New averaged squared gradients
__group_303__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_304__new_weights (optional) - T2: New weights
__group_304__new_gradients (optional) - T3: New gradients
__group_304__new_moment_1 (optional) - T4: New averaged gradients
__group_304__new_moment_2 (optional) - T4: New averaged squared gradients
__group_304__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_305__new_weights (optional) - T2: New weights
__group_305__new_gradients (optional) - T3: New gradients
__group_305__new_moment_1 (optional) - T4: New averaged gradients
__group_305__new_moment_2 (optional) - T4: New averaged squared gradients
__group_305__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_306__new_weights (optional) - T2: New weights
__group_306__new_gradients (optional) - T3: New gradients
__group_306__new_moment_1 (optional) - T4: New averaged gradients
__group_306__new_moment_2 (optional) - T4: New averaged squared gradients
__group_306__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_307__new_weights (optional) - T2: New weights
__group_307__new_gradients (optional) - T3: New gradients
__group_307__new_moment_1 (optional) - T4: New averaged gradients
__group_307__new_moment_2 (optional) - T4: New averaged squared gradients
__group_307__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_308__new_weights (optional) - T2: New weights
__group_308__new_gradients (optional) - T3: New gradients
__group_308__new_moment_1 (optional) - T4: New averaged gradients
__group_308__new_moment_2 (optional) - T4: New averaged squared gradients
__group_308__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_309__new_weights (optional) - T2: New weights
__group_309__new_gradients (optional) - T3: New gradients
__group_309__new_moment_1 (optional) - T4: New averaged gradients
__group_309__new_moment_2 (optional) - T4: New averaged squared gradients
__group_309__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_310__new_weights (optional) - T2: New weights
__group_310__new_gradients (optional) - T3: New gradients
__group_310__new_moment_1 (optional) - T4: New averaged gradients
__group_310__new_moment_2 (optional) - T4: New averaged squared gradients
__group_310__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_311__new_weights (optional) - T2: New weights
__group_311__new_gradients (optional) - T3: New gradients
__group_311__new_moment_1 (optional) - T4: New averaged gradients
__group_311__new_moment_2 (optional) - T4: New averaged squared gradients
__group_311__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_312__new_weights (optional) - T2: New weights
__group_312__new_gradients (optional) - T3: New gradients
__group_312__new_moment_1 (optional) - T4: New averaged gradients
__group_312__new_moment_2 (optional) - T4: New averaged squared gradients
__group_312__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_313__new_weights (optional) - T2: New weights
__group_313__new_gradients (optional) - T3: New gradients
__group_313__new_moment_1 (optional) - T4: New averaged gradients
__group_313__new_moment_2 (optional) - T4: New averaged squared gradients
__group_313__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_314__new_weights (optional) - T2: New weights
__group_314__new_gradients (optional) - T3: New gradients
__group_314__new_moment_1 (optional) - T4: New averaged gradients
__group_314__new_moment_2 (optional) - T4: New averaged squared gradients
__group_314__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_315__new_weights (optional) - T2: New weights
__group_315__new_gradients (optional) - T3: New gradients
__group_315__new_moment_1 (optional) - T4: New averaged gradients
__group_315__new_moment_2 (optional) - T4: New averaged squared gradients
__group_315__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_316__new_weights (optional) - T2: New weights
__group_316__new_gradients (optional) - T3: New gradients
__group_316__new_moment_1 (optional) - T4: New averaged gradients
__group_316__new_moment_2 (optional) - T4: New averaged squared gradients
__group_316__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_317__new_weights (optional) - T2: New weights
__group_317__new_gradients (optional) - T3: New gradients
__group_317__new_moment_1 (optional) - T4: New averaged gradients
__group_317__new_moment_2 (optional) - T4: New averaged squared gradients
__group_317__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_318__new_weights (optional) - T2: New weights
__group_318__new_gradients (optional) - T3: New gradients
__group_318__new_moment_1 (optional) - T4: New averaged gradients
__group_318__new_moment_2 (optional) - T4: New averaged squared gradients
__group_318__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_319__new_weights (optional) - T2: New weights
__group_319__new_gradients (optional) - T3: New gradients
__group_319__new_moment_1 (optional) - T4: New averaged gradients
__group_319__new_moment_2 (optional) - T4: New averaged squared gradients
__group_319__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_320__new_weights (optional) - T2: New weights
__group_320__new_gradients (optional) - T3: New gradients
__group_320__new_moment_1 (optional) - T4: New averaged gradients
__group_320__new_moment_2 (optional) - T4: New averaged squared gradients
__group_320__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_321__new_weights (optional) - T2: New weights
__group_321__new_gradients (optional) - T3: New gradients
__group_321__new_moment_1 (optional) - T4: New averaged gradients
__group_321__new_moment_2 (optional) - T4: New averaged squared gradients
__group_321__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_322__new_weights (optional) - T2: New weights
__group_322__new_gradients (optional) - T3: New gradients
__group_322__new_moment_1 (optional) - T4: New averaged gradients
__group_322__new_moment_2 (optional) - T4: New averaged squared gradients
__group_322__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_323__new_weights (optional) - T2: New weights
__group_323__new_gradients (optional) - T3: New gradients
__group_323__new_moment_1 (optional) - T4: New averaged gradients
__group_323__new_moment_2 (optional) - T4: New averaged squared gradients
__group_323__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_324__new_weights (optional) - T2: New weights
__group_324__new_gradients (optional) - T3: New gradients
__group_324__new_moment_1 (optional) - T4: New averaged gradients
__group_324__new_moment_2 (optional) - T4: New averaged squared gradients
__group_324__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_325__new_weights (optional) - T2: New weights
__group_325__new_gradients (optional) - T3: New gradients
__group_325__new_moment_1 (optional) - T4: New averaged gradients
__group_325__new_moment_2 (optional) - T4: New averaged squared gradients
__group_325__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_326__new_weights (optional) - T2: New weights
__group_326__new_gradients (optional) - T3: New gradients
__group_326__new_moment_1 (optional) - T4: New averaged gradients
__group_326__new_moment_2 (optional) - T4: New averaged squared gradients
__group_326__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_327__new_weights (optional) - T2: New weights
__group_327__new_gradients (optional) - T3: New gradients
__group_327__new_moment_1 (optional) - T4: New averaged gradients
__group_327__new_moment_2 (optional) - T4: New averaged squared gradients
__group_327__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_328__new_weights (optional) - T2: New weights
__group_328__new_gradients (optional) - T3: New gradients
__group_328__new_moment_1 (optional) - T4: New averaged gradients
__group_328__new_moment_2 (optional) - T4: New averaged squared gradients
__group_328__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_329__new_weights (optional) - T2: New weights
__group_329__new_gradients (optional) - T3: New gradients
__group_329__new_moment_1 (optional) - T4: New averaged gradients
__group_329__new_moment_2 (optional) - T4: New averaged squared gradients
__group_329__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_330__new_weights (optional) - T2: New weights
__group_330__new_gradients (optional) - T3: New gradients
__group_330__new_moment_1 (optional) - T4: New averaged gradients
__group_330__new_moment_2 (optional) - T4: New averaged squared gradients
__group_330__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_331__new_weights (optional) - T2: New weights
__group_331__new_gradients (optional) - T3: New gradients
__group_331__new_moment_1 (optional) - T4: New averaged gradients
__group_331__new_moment_2 (optional) - T4: New averaged squared gradients
__group_331__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_332__new_weights (optional) - T2: New weights
__group_332__new_gradients (optional) - T3: New gradients
__group_332__new_moment_1 (optional) - T4: New averaged gradients
__group_332__new_moment_2 (optional) - T4: New averaged squared gradients
__group_332__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_333__new_weights (optional) - T2: New weights
__group_333__new_gradients (optional) - T3: New gradients
__group_333__new_moment_1 (optional) - T4: New averaged gradients
__group_333__new_moment_2 (optional) - T4: New averaged squared gradients
__group_333__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_334__new_weights (optional) - T2: New weights
__group_334__new_gradients (optional) - T3: New gradients
__group_334__new_moment_1 (optional) - T4: New averaged gradients
__group_334__new_moment_2 (optional) - T4: New averaged squared gradients
__group_334__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_335__new_weights (optional) - T2: New weights
__group_335__new_gradients (optional) - T3: New gradients
__group_335__new_moment_1 (optional) - T4: New averaged gradients
__group_335__new_moment_2 (optional) - T4: New averaged squared gradients
__group_335__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_336__new_weights (optional) - T2: New weights
__group_336__new_gradients (optional) - T3: New gradients
__group_336__new_moment_1 (optional) - T4: New averaged gradients
__group_336__new_moment_2 (optional) - T4: New averaged squared gradients
__group_336__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_337__new_weights (optional) - T2: New weights
__group_337__new_gradients (optional) - T3: New gradients
__group_337__new_moment_1 (optional) - T4: New averaged gradients
__group_337__new_moment_2 (optional) - T4: New averaged squared gradients
__group_337__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_338__new_weights (optional) - T2: New weights
__group_338__new_gradients (optional) - T3: New gradients
__group_338__new_moment_1 (optional) - T4: New averaged gradients
__group_338__new_moment_2 (optional) - T4: New averaged squared gradients
__group_338__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_339__new_weights (optional) - T2: New weights
__group_339__new_gradients (optional) - T3: New gradients
__group_339__new_moment_1 (optional) - T4: New averaged gradients
__group_339__new_moment_2 (optional) - T4: New averaged squared gradients
__group_339__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_340__new_weights (optional) - T2: New weights
__group_340__new_gradients (optional) - T3: New gradients
__group_340__new_moment_1 (optional) - T4: New averaged gradients
__group_340__new_moment_2 (optional) - T4: New averaged squared gradients
__group_340__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_341__new_weights (optional) - T2: New weights
__group_341__new_gradients (optional) - T3: New gradients
__group_341__new_moment_1 (optional) - T4: New averaged gradients
__group_341__new_moment_2 (optional) - T4: New averaged squared gradients
__group_341__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_342__new_weights (optional) - T2: New weights
__group_342__new_gradients (optional) - T3: New gradients
__group_342__new_moment_1 (optional) - T4: New averaged gradients
__group_342__new_moment_2 (optional) - T4: New averaged squared gradients
__group_342__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_343__new_weights (optional) - T2: New weights
__group_343__new_gradients (optional) - T3: New gradients
__group_343__new_moment_1 (optional) - T4: New averaged gradients
__group_343__new_moment_2 (optional) - T4: New averaged squared gradients
__group_343__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_344__new_weights (optional) - T2: New weights
__group_344__new_gradients (optional) - T3: New gradients
__group_344__new_moment_1 (optional) - T4: New averaged gradients
__group_344__new_moment_2 (optional) - T4: New averaged squared gradients
__group_344__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_345__new_weights (optional) - T2: New weights
__group_345__new_gradients (optional) - T3: New gradients
__group_345__new_moment_1 (optional) - T4: New averaged gradients
__group_345__new_moment_2 (optional) - T4: New averaged squared gradients
__group_345__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_346__new_weights (optional) - T2: New weights
__group_346__new_gradients (optional) - T3: New gradients
__group_346__new_moment_1 (optional) - T4: New averaged gradients
__group_346__new_moment_2 (optional) - T4: New averaged squared gradients
__group_346__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_347__new_weights (optional) - T2: New weights
__group_347__new_gradients (optional) - T3: New gradients
__group_347__new_moment_1 (optional) - T4: New averaged gradients
__group_347__new_moment_2 (optional) - T4: New averaged squared gradients
__group_347__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_348__new_weights (optional) - T2: New weights
__group_348__new_gradients (optional) - T3: New gradients
__group_348__new_moment_1 (optional) - T4: New averaged gradients
__group_348__new_moment_2 (optional) - T4: New averaged squared gradients
__group_348__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_349__new_weights (optional) - T2: New weights
__group_349__new_gradients (optional) - T3: New gradients
__group_349__new_moment_1 (optional) - T4: New averaged gradients
__group_349__new_moment_2 (optional) - T4: New averaged squared gradients
__group_349__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_350__new_weights (optional) - T2: New weights
__group_350__new_gradients (optional) - T3: New gradients
__group_350__new_moment_1 (optional) - T4: New averaged gradients
__group_350__new_moment_2 (optional) - T4: New averaged squared gradients
__group_350__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_351__new_weights (optional) - T2: New weights
__group_351__new_gradients (optional) - T3: New gradients
__group_351__new_moment_1 (optional) - T4: New averaged gradients
__group_351__new_moment_2 (optional) - T4: New averaged squared gradients
__group_351__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_352__new_weights (optional) - T2: New weights
__group_352__new_gradients (optional) - T3: New gradients
__group_352__new_moment_1 (optional) - T4: New averaged gradients
__group_352__new_moment_2 (optional) - T4: New averaged squared gradients
__group_352__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_353__new_weights (optional) - T2: New weights
__group_353__new_gradients (optional) - T3: New gradients
__group_353__new_moment_1 (optional) - T4: New averaged gradients
__group_353__new_moment_2 (optional) - T4: New averaged squared gradients
__group_353__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_354__new_weights (optional) - T2: New weights
__group_354__new_gradients (optional) - T3: New gradients
__group_354__new_moment_1 (optional) - T4: New averaged gradients
__group_354__new_moment_2 (optional) - T4: New averaged squared gradients
__group_354__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_355__new_weights (optional) - T2: New weights
__group_355__new_gradients (optional) - T3: New gradients
__group_355__new_moment_1 (optional) - T4: New averaged gradients
__group_355__new_moment_2 (optional) - T4: New averaged squared gradients
__group_355__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_356__new_weights (optional) - T2: New weights
__group_356__new_gradients (optional) - T3: New gradients
__group_356__new_moment_1 (optional) - T4: New averaged gradients
__group_356__new_moment_2 (optional) - T4: New averaged squared gradients
__group_356__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_357__new_weights (optional) - T2: New weights
__group_357__new_gradients (optional) - T3: New gradients
__group_357__new_moment_1 (optional) - T4: New averaged gradients
__group_357__new_moment_2 (optional) - T4: New averaged squared gradients
__group_357__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_358__new_weights (optional) - T2: New weights
__group_358__new_gradients (optional) - T3: New gradients
__group_358__new_moment_1 (optional) - T4: New averaged gradients
__group_358__new_moment_2 (optional) - T4: New averaged squared gradients
__group_358__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_359__new_weights (optional) - T2: New weights
__group_359__new_gradients (optional) - T3: New gradients
__group_359__new_moment_1 (optional) - T4: New averaged gradients
__group_359__new_moment_2 (optional) - T4: New averaged squared gradients
__group_359__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_360__new_weights (optional) - T2: New weights
__group_360__new_gradients (optional) - T3: New gradients
__group_360__new_moment_1 (optional) - T4: New averaged gradients
__group_360__new_moment_2 (optional) - T4: New averaged squared gradients
__group_360__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_361__new_weights (optional) - T2: New weights
__group_361__new_gradients (optional) - T3: New gradients
__group_361__new_moment_1 (optional) - T4: New averaged gradients
__group_361__new_moment_2 (optional) - T4: New averaged squared gradients
__group_361__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_362__new_weights (optional) - T2: New weights
__group_362__new_gradients (optional) - T3: New gradients
__group_362__new_moment_1 (optional) - T4: New averaged gradients
__group_362__new_moment_2 (optional) - T4: New averaged squared gradients
__group_362__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_363__new_weights (optional) - T2: New weights
__group_363__new_gradients (optional) - T3: New gradients
__group_363__new_moment_1 (optional) - T4: New averaged gradients
__group_363__new_moment_2 (optional) - T4: New averaged squared gradients
__group_363__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_364__new_weights (optional) - T2: New weights
__group_364__new_gradients (optional) - T3: New gradients
__group_364__new_moment_1 (optional) - T4: New averaged gradients
__group_364__new_moment_2 (optional) - T4: New averaged squared gradients
__group_364__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_365__new_weights (optional) - T2: New weights
__group_365__new_gradients (optional) - T3: New gradients
__group_365__new_moment_1 (optional) - T4: New averaged gradients
__group_365__new_moment_2 (optional) - T4: New averaged squared gradients
__group_365__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_366__new_weights (optional) - T2: New weights
__group_366__new_gradients (optional) - T3: New gradients
__group_366__new_moment_1 (optional) - T4: New averaged gradients
__group_366__new_moment_2 (optional) - T4: New averaged squared gradients
__group_366__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_367__new_weights (optional) - T2: New weights
__group_367__new_gradients (optional) - T3: New gradients
__group_367__new_moment_1 (optional) - T4: New averaged gradients
__group_367__new_moment_2 (optional) - T4: New averaged squared gradients
__group_367__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_368__new_weights (optional) - T2: New weights
__group_368__new_gradients (optional) - T3: New gradients
__group_368__new_moment_1 (optional) - T4: New averaged gradients
__group_368__new_moment_2 (optional) - T4: New averaged squared gradients
__group_368__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_369__new_weights (optional) - T2: New weights
__group_369__new_gradients (optional) - T3: New gradients
__group_369__new_moment_1 (optional) - T4: New averaged gradients
__group_369__new_moment_2 (optional) - T4: New averaged squared gradients
__group_369__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_370__new_weights (optional) - T2: New weights
__group_370__new_gradients (optional) - T3: New gradients
__group_370__new_moment_1 (optional) - T4: New averaged gradients
__group_370__new_moment_2 (optional) - T4: New averaged squared gradients
__group_370__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_371__new_weights (optional) - T2: New weights
__group_371__new_gradients (optional) - T3: New gradients
__group_371__new_moment_1 (optional) - T4: New averaged gradients
__group_371__new_moment_2 (optional) - T4: New averaged squared gradients
__group_371__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_372__new_weights (optional) - T2: New weights
__group_372__new_gradients (optional) - T3: New gradients
__group_372__new_moment_1 (optional) - T4: New averaged gradients
__group_372__new_moment_2 (optional) - T4: New averaged squared gradients
__group_372__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_373__new_weights (optional) - T2: New weights
__group_373__new_gradients (optional) - T3: New gradients
__group_373__new_moment_1 (optional) - T4: New averaged gradients
__group_373__new_moment_2 (optional) - T4: New averaged squared gradients
__group_373__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_374__new_weights (optional) - T2: New weights
__group_374__new_gradients (optional) - T3: New gradients
__group_374__new_moment_1 (optional) - T4: New averaged gradients
__group_374__new_moment_2 (optional) - T4: New averaged squared gradients
__group_374__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_375__new_weights (optional) - T2: New weights
__group_375__new_gradients (optional) - T3: New gradients
__group_375__new_moment_1 (optional) - T4: New averaged gradients
__group_375__new_moment_2 (optional) - T4: New averaged squared gradients
__group_375__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_376__new_weights (optional) - T2: New weights
__group_376__new_gradients (optional) - T3: New gradients
__group_376__new_moment_1 (optional) - T4: New averaged gradients
__group_376__new_moment_2 (optional) - T4: New averaged squared gradients
__group_376__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_377__new_weights (optional) - T2: New weights
__group_377__new_gradients (optional) - T3: New gradients
__group_377__new_moment_1 (optional) - T4: New averaged gradients
__group_377__new_moment_2 (optional) - T4: New averaged squared gradients
__group_377__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_378__new_weights (optional) - T2: New weights
__group_378__new_gradients (optional) - T3: New gradients
__group_378__new_moment_1 (optional) - T4: New averaged gradients
__group_378__new_moment_2 (optional) - T4: New averaged squared gradients
__group_378__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_379__new_weights (optional) - T2: New weights
__group_379__new_gradients (optional) - T3: New gradients
__group_379__new_moment_1 (optional) - T4: New averaged gradients
__group_379__new_moment_2 (optional) - T4: New averaged squared gradients
__group_379__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_380__new_weights (optional) - T2: New weights
__group_380__new_gradients (optional) - T3: New gradients
__group_380__new_moment_1 (optional) - T4: New averaged gradients
__group_380__new_moment_2 (optional) - T4: New averaged squared gradients
__group_380__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_381__new_weights (optional) - T2: New weights
__group_381__new_gradients (optional) - T3: New gradients
__group_381__new_moment_1 (optional) - T4: New averaged gradients
__group_381__new_moment_2 (optional) - T4: New averaged squared gradients
__group_381__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_382__new_weights (optional) - T2: New weights
__group_382__new_gradients (optional) - T3: New gradients
__group_382__new_moment_1 (optional) - T4: New averaged gradients
__group_382__new_moment_2 (optional) - T4: New averaged squared gradients
__group_382__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_383__new_weights (optional) - T2: New weights
__group_383__new_gradients (optional) - T3: New gradients
__group_383__new_moment_1 (optional) - T4: New averaged gradients
__group_383__new_moment_2 (optional) - T4: New averaged squared gradients
__group_383__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_384__new_weights (optional) - T2: New weights
__group_384__new_gradients (optional) - T3: New gradients
__group_384__new_moment_1 (optional) - T4: New averaged gradients
__group_384__new_moment_2 (optional) - T4: New averaged squared gradients
__group_384__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_385__new_weights (optional) - T2: New weights
__group_385__new_gradients (optional) - T3: New gradients
__group_385__new_moment_1 (optional) - T4: New averaged gradients
__group_385__new_moment_2 (optional) - T4: New averaged squared gradients
__group_385__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_386__new_weights (optional) - T2: New weights
__group_386__new_gradients (optional) - T3: New gradients
__group_386__new_moment_1 (optional) - T4: New averaged gradients
__group_386__new_moment_2 (optional) - T4: New averaged squared gradients
__group_386__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_387__new_weights (optional) - T2: New weights
__group_387__new_gradients (optional) - T3: New gradients
__group_387__new_moment_1 (optional) - T4: New averaged gradients
__group_387__new_moment_2 (optional) - T4: New averaged squared gradients
__group_387__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_388__new_weights (optional) - T2: New weights
__group_388__new_gradients (optional) - T3: New gradients
__group_388__new_moment_1 (optional) - T4: New averaged gradients
__group_388__new_moment_2 (optional) - T4: New averaged squared gradients
__group_388__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_389__new_weights (optional) - T2: New weights
__group_389__new_gradients (optional) - T3: New gradients
__group_389__new_moment_1 (optional) - T4: New averaged gradients
__group_389__new_moment_2 (optional) - T4: New averaged squared gradients
__group_389__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_390__new_weights (optional) - T2: New weights
__group_390__new_gradients (optional) - T3: New gradients
__group_390__new_moment_1 (optional) - T4: New averaged gradients
__group_390__new_moment_2 (optional) - T4: New averaged squared gradients
__group_390__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_391__new_weights (optional) - T2: New weights
__group_391__new_gradients (optional) - T3: New gradients
__group_391__new_moment_1 (optional) - T4: New averaged gradients
__group_391__new_moment_2 (optional) - T4: New averaged squared gradients
__group_391__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_392__new_weights (optional) - T2: New weights
__group_392__new_gradients (optional) - T3: New gradients
__group_392__new_moment_1 (optional) - T4: New averaged gradients
__group_392__new_moment_2 (optional) - T4: New averaged squared gradients
__group_392__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_393__new_weights (optional) - T2: New weights
__group_393__new_gradients (optional) - T3: New gradients
__group_393__new_moment_1 (optional) - T4: New averaged gradients
__group_393__new_moment_2 (optional) - T4: New averaged squared gradients
__group_393__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_394__new_weights (optional) - T2: New weights
__group_394__new_gradients (optional) - T3: New gradients
__group_394__new_moment_1 (optional) - T4: New averaged gradients
__group_394__new_moment_2 (optional) - T4: New averaged squared gradients
__group_394__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_395__new_weights (optional) - T2: New weights
__group_395__new_gradients (optional) - T3: New gradients
__group_395__new_moment_1 (optional) - T4: New averaged gradients
__group_395__new_moment_2 (optional) - T4: New averaged squared gradients
__group_395__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_396__new_weights (optional) - T2: New weights
__group_396__new_gradients (optional) - T3: New gradients
__group_396__new_moment_1 (optional) - T4: New averaged gradients
__group_396__new_moment_2 (optional) - T4: New averaged squared gradients
__group_396__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_397__new_weights (optional) - T2: New weights
__group_397__new_gradients (optional) - T3: New gradients
__group_397__new_moment_1 (optional) - T4: New averaged gradients
__group_397__new_moment_2 (optional) - T4: New averaged squared gradients
__group_397__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_398__new_weights (optional) - T2: New weights
__group_398__new_gradients (optional) - T3: New gradients
__group_398__new_moment_1 (optional) - T4: New averaged gradients
__group_398__new_moment_2 (optional) - T4: New averaged squared gradients
__group_398__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_399__new_weights (optional) - T2: New weights
__group_399__new_gradients (optional) - T3: New gradients
__group_399__new_moment_1 (optional) - T4: New averaged gradients
__group_399__new_moment_2 (optional) - T4: New averaged squared gradients
__group_399__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_400__new_weights (optional) - T2: New weights
__group_400__new_gradients (optional) - T3: New gradients
__group_400__new_moment_1 (optional) - T4: New averaged gradients
__group_400__new_moment_2 (optional) - T4: New averaged squared gradients
__group_400__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_401__new_weights (optional) - T2: New weights
__group_401__new_gradients (optional) - T3: New gradients
__group_401__new_moment_1 (optional) - T4: New averaged gradients
__group_401__new_moment_2 (optional) - T4: New averaged squared gradients
__group_401__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_402__new_weights (optional) - T2: New weights
__group_402__new_gradients (optional) - T3: New gradients
__group_402__new_moment_1 (optional) - T4: New averaged gradients
__group_402__new_moment_2 (optional) - T4: New averaged squared gradients
__group_402__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_403__new_weights (optional) - T2: New weights
__group_403__new_gradients (optional) - T3: New gradients
__group_403__new_moment_1 (optional) - T4: New averaged gradients
__group_403__new_moment_2 (optional) - T4: New averaged squared gradients
__group_403__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_404__new_weights (optional) - T2: New weights
__group_404__new_gradients (optional) - T3: New gradients
__group_404__new_moment_1 (optional) - T4: New averaged gradients
__group_404__new_moment_2 (optional) - T4: New averaged squared gradients
__group_404__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_405__new_weights (optional) - T2: New weights
__group_405__new_gradients (optional) - T3: New gradients
__group_405__new_moment_1 (optional) - T4: New averaged gradients
__group_405__new_moment_2 (optional) - T4: New averaged squared gradients
__group_405__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_406__new_weights (optional) - T2: New weights
__group_406__new_gradients (optional) - T3: New gradients
__group_406__new_moment_1 (optional) - T4: New averaged gradients
__group_406__new_moment_2 (optional) - T4: New averaged squared gradients
__group_406__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_407__new_weights (optional) - T2: New weights
__group_407__new_gradients (optional) - T3: New gradients
__group_407__new_moment_1 (optional) - T4: New averaged gradients
__group_407__new_moment_2 (optional) - T4: New averaged squared gradients
__group_407__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_408__new_weights (optional) - T2: New weights
__group_408__new_gradients (optional) - T3: New gradients
__group_408__new_moment_1 (optional) - T4: New averaged gradients
__group_408__new_moment_2 (optional) - T4: New averaged squared gradients
__group_408__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_409__new_weights (optional) - T2: New weights
__group_409__new_gradients (optional) - T3: New gradients
__group_409__new_moment_1 (optional) - T4: New averaged gradients
__group_409__new_moment_2 (optional) - T4: New averaged squared gradients
__group_409__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_410__new_weights (optional) - T2: New weights
__group_410__new_gradients (optional) - T3: New gradients
__group_410__new_moment_1 (optional) - T4: New averaged gradients
__group_410__new_moment_2 (optional) - T4: New averaged squared gradients
__group_410__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_411__new_weights (optional) - T2: New weights
__group_411__new_gradients (optional) - T3: New gradients
__group_411__new_moment_1 (optional) - T4: New averaged gradients
__group_411__new_moment_2 (optional) - T4: New averaged squared gradients
__group_411__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_412__new_weights (optional) - T2: New weights
__group_412__new_gradients (optional) - T3: New gradients
__group_412__new_moment_1 (optional) - T4: New averaged gradients
__group_412__new_moment_2 (optional) - T4: New averaged squared gradients
__group_412__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_413__new_weights (optional) - T2: New weights
__group_413__new_gradients (optional) - T3: New gradients
__group_413__new_moment_1 (optional) - T4: New averaged gradients
__group_413__new_moment_2 (optional) - T4: New averaged squared gradients
__group_413__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_414__new_weights (optional) - T2: New weights
__group_414__new_gradients (optional) - T3: New gradients
__group_414__new_moment_1 (optional) - T4: New averaged gradients
__group_414__new_moment_2 (optional) - T4: New averaged squared gradients
__group_414__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_415__new_weights (optional) - T2: New weights
__group_415__new_gradients (optional) - T3: New gradients
__group_415__new_moment_1 (optional) - T4: New averaged gradients
__group_415__new_moment_2 (optional) - T4: New averaged squared gradients
__group_415__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_416__new_weights (optional) - T2: New weights
__group_416__new_gradients (optional) - T3: New gradients
__group_416__new_moment_1 (optional) - T4: New averaged gradients
__group_416__new_moment_2 (optional) - T4: New averaged squared gradients
__group_416__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_417__new_weights (optional) - T2: New weights
__group_417__new_gradients (optional) - T3: New gradients
__group_417__new_moment_1 (optional) - T4: New averaged gradients
__group_417__new_moment_2 (optional) - T4: New averaged squared gradients
__group_417__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_418__new_weights (optional) - T2: New weights
__group_418__new_gradients (optional) - T3: New gradients
__group_418__new_moment_1 (optional) - T4: New averaged gradients
__group_418__new_moment_2 (optional) - T4: New averaged squared gradients
__group_418__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_419__new_weights (optional) - T2: New weights
__group_419__new_gradients (optional) - T3: New gradients
__group_419__new_moment_1 (optional) - T4: New averaged gradients
__group_419__new_moment_2 (optional) - T4: New averaged squared gradients
__group_419__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_420__new_weights (optional) - T2: New weights
__group_420__new_gradients (optional) - T3: New gradients
__group_420__new_moment_1 (optional) - T4: New averaged gradients
__group_420__new_moment_2 (optional) - T4: New averaged squared gradients
__group_420__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_421__new_weights (optional) - T2: New weights
__group_421__new_gradients (optional) - T3: New gradients
__group_421__new_moment_1 (optional) - T4: New averaged gradients
__group_421__new_moment_2 (optional) - T4: New averaged squared gradients
__group_421__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_422__new_weights (optional) - T2: New weights
__group_422__new_gradients (optional) - T3: New gradients
__group_422__new_moment_1 (optional) - T4: New averaged gradients
__group_422__new_moment_2 (optional) - T4: New averaged squared gradients
__group_422__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_423__new_weights (optional) - T2: New weights
__group_423__new_gradients (optional) - T3: New gradients
__group_423__new_moment_1 (optional) - T4: New averaged gradients
__group_423__new_moment_2 (optional) - T4: New averaged squared gradients
__group_423__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_424__new_weights (optional) - T2: New weights
__group_424__new_gradients (optional) - T3: New gradients
__group_424__new_moment_1 (optional) - T4: New averaged gradients
__group_424__new_moment_2 (optional) - T4: New averaged squared gradients
__group_424__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_425__new_weights (optional) - T2: New weights
__group_425__new_gradients (optional) - T3: New gradients
__group_425__new_moment_1 (optional) - T4: New averaged gradients
__group_425__new_moment_2 (optional) - T4: New averaged squared gradients
__group_425__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_426__new_weights (optional) - T2: New weights
__group_426__new_gradients (optional) - T3: New gradients
__group_426__new_moment_1 (optional) - T4: New averaged gradients
__group_426__new_moment_2 (optional) - T4: New averaged squared gradients
__group_426__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_427__new_weights (optional) - T2: New weights
__group_427__new_gradients (optional) - T3: New gradients
__group_427__new_moment_1 (optional) - T4: New averaged gradients
__group_427__new_moment_2 (optional) - T4: New averaged squared gradients
__group_427__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_428__new_weights (optional) - T2: New weights
__group_428__new_gradients (optional) - T3: New gradients
__group_428__new_moment_1 (optional) - T4: New averaged gradients
__group_428__new_moment_2 (optional) - T4: New averaged squared gradients
__group_428__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_429__new_weights (optional) - T2: New weights
__group_429__new_gradients (optional) - T3: New gradients
__group_429__new_moment_1 (optional) - T4: New averaged gradients
__group_429__new_moment_2 (optional) - T4: New averaged squared gradients
__group_429__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_430__new_weights (optional) - T2: New weights
__group_430__new_gradients (optional) - T3: New gradients
__group_430__new_moment_1 (optional) - T4: New averaged gradients
__group_430__new_moment_2 (optional) - T4: New averaged squared gradients
__group_430__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_431__new_weights (optional) - T2: New weights
__group_431__new_gradients (optional) - T3: New gradients
__group_431__new_moment_1 (optional) - T4: New averaged gradients
__group_431__new_moment_2 (optional) - T4: New averaged squared gradients
__group_431__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_432__new_weights (optional) - T2: New weights
__group_432__new_gradients (optional) - T3: New gradients
__group_432__new_moment_1 (optional) - T4: New averaged gradients
__group_432__new_moment_2 (optional) - T4: New averaged squared gradients
__group_432__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_433__new_weights (optional) - T2: New weights
__group_433__new_gradients (optional) - T3: New gradients
__group_433__new_moment_1 (optional) - T4: New averaged gradients
__group_433__new_moment_2 (optional) - T4: New averaged squared gradients
__group_433__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_434__new_weights (optional) - T2: New weights
__group_434__new_gradients (optional) - T3: New gradients
__group_434__new_moment_1 (optional) - T4: New averaged gradients
__group_434__new_moment_2 (optional) - T4: New averaged squared gradients
__group_434__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_435__new_weights (optional) - T2: New weights
__group_435__new_gradients (optional) - T3: New gradients
__group_435__new_moment_1 (optional) - T4: New averaged gradients
__group_435__new_moment_2 (optional) - T4: New averaged squared gradients
__group_435__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_436__new_weights (optional) - T2: New weights
__group_436__new_gradients (optional) - T3: New gradients
__group_436__new_moment_1 (optional) - T4: New averaged gradients
__group_436__new_moment_2 (optional) - T4: New averaged squared gradients
__group_436__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_437__new_weights (optional) - T2: New weights
__group_437__new_gradients (optional) - T3: New gradients
__group_437__new_moment_1 (optional) - T4: New averaged gradients
__group_437__new_moment_2 (optional) - T4: New averaged squared gradients
__group_437__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_438__new_weights (optional) - T2: New weights
__group_438__new_gradients (optional) - T3: New gradients
__group_438__new_moment_1 (optional) - T4: New averaged gradients
__group_438__new_moment_2 (optional) - T4: New averaged squared gradients
__group_438__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_439__new_weights (optional) - T2: New weights
__group_439__new_gradients (optional) - T3: New gradients
__group_439__new_moment_1 (optional) - T4: New averaged gradients
__group_439__new_moment_2 (optional) - T4: New averaged squared gradients
__group_439__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_440__new_weights (optional) - T2: New weights
__group_440__new_gradients (optional) - T3: New gradients
__group_440__new_moment_1 (optional) - T4: New averaged gradients
__group_440__new_moment_2 (optional) - T4: New averaged squared gradients
__group_440__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_441__new_weights (optional) - T2: New weights
__group_441__new_gradients (optional) - T3: New gradients
__group_441__new_moment_1 (optional) - T4: New averaged gradients
__group_441__new_moment_2 (optional) - T4: New averaged squared gradients
__group_441__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_442__new_weights (optional) - T2: New weights
__group_442__new_gradients (optional) - T3: New gradients
__group_442__new_moment_1 (optional) - T4: New averaged gradients
__group_442__new_moment_2 (optional) - T4: New averaged squared gradients
__group_442__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_443__new_weights (optional) - T2: New weights
__group_443__new_gradients (optional) - T3: New gradients
__group_443__new_moment_1 (optional) - T4: New averaged gradients
__group_443__new_moment_2 (optional) - T4: New averaged squared gradients
__group_443__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_444__new_weights (optional) - T2: New weights
__group_444__new_gradients (optional) - T3: New gradients
__group_444__new_moment_1 (optional) - T4: New averaged gradients
__group_444__new_moment_2 (optional) - T4: New averaged squared gradients
__group_444__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_445__new_weights (optional) - T2: New weights
__group_445__new_gradients (optional) - T3: New gradients
__group_445__new_moment_1 (optional) - T4: New averaged gradients
__group_445__new_moment_2 (optional) - T4: New averaged squared gradients
__group_445__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_446__new_weights (optional) - T2: New weights
__group_446__new_gradients (optional) - T3: New gradients
__group_446__new_moment_1 (optional) - T4: New averaged gradients
__group_446__new_moment_2 (optional) - T4: New averaged squared gradients
__group_446__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_447__new_weights (optional) - T2: New weights
__group_447__new_gradients (optional) - T3: New gradients
__group_447__new_moment_1 (optional) - T4: New averaged gradients
__group_447__new_moment_2 (optional) - T4: New averaged squared gradients
__group_447__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_448__new_weights (optional) - T2: New weights
__group_448__new_gradients (optional) - T3: New gradients
__group_448__new_moment_1 (optional) - T4: New averaged gradients
__group_448__new_moment_2 (optional) - T4: New averaged squared gradients
__group_448__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_449__new_weights (optional) - T2: New weights
__group_449__new_gradients (optional) - T3: New gradients
__group_449__new_moment_1 (optional) - T4: New averaged gradients
__group_449__new_moment_2 (optional) - T4: New averaged squared gradients
__group_449__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_450__new_weights (optional) - T2: New weights
__group_450__new_gradients (optional) - T3: New gradients
__group_450__new_moment_1 (optional) - T4: New averaged gradients
__group_450__new_moment_2 (optional) - T4: New averaged squared gradients
__group_450__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_451__new_weights (optional) - T2: New weights
__group_451__new_gradients (optional) - T3: New gradients
__group_451__new_moment_1 (optional) - T4: New averaged gradients
__group_451__new_moment_2 (optional) - T4: New averaged squared gradients
__group_451__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_452__new_weights (optional) - T2: New weights
__group_452__new_gradients (optional) - T3: New gradients
__group_452__new_moment_1 (optional) - T4: New averaged gradients
__group_452__new_moment_2 (optional) - T4: New averaged squared gradients
__group_452__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_453__new_weights (optional) - T2: New weights
__group_453__new_gradients (optional) - T3: New gradients
__group_453__new_moment_1 (optional) - T4: New averaged gradients
__group_453__new_moment_2 (optional) - T4: New averaged squared gradients
__group_453__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_454__new_weights (optional) - T2: New weights
__group_454__new_gradients (optional) - T3: New gradients
__group_454__new_moment_1 (optional) - T4: New averaged gradients
__group_454__new_moment_2 (optional) - T4: New averaged squared gradients
__group_454__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_455__new_weights (optional) - T2: New weights
__group_455__new_gradients (optional) - T3: New gradients
__group_455__new_moment_1 (optional) - T4: New averaged gradients
__group_455__new_moment_2 (optional) - T4: New averaged squared gradients
__group_455__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_456__new_weights (optional) - T2: New weights
__group_456__new_gradients (optional) - T3: New gradients
__group_456__new_moment_1 (optional) - T4: New averaged gradients
__group_456__new_moment_2 (optional) - T4: New averaged squared gradients
__group_456__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_457__new_weights (optional) - T2: New weights
__group_457__new_gradients (optional) - T3: New gradients
__group_457__new_moment_1 (optional) - T4: New averaged gradients
__group_457__new_moment_2 (optional) - T4: New averaged squared gradients
__group_457__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_458__new_weights (optional) - T2: New weights
__group_458__new_gradients (optional) - T3: New gradients
__group_458__new_moment_1 (optional) - T4: New averaged gradients
__group_458__new_moment_2 (optional) - T4: New averaged squared gradients
__group_458__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_459__new_weights (optional) - T2: New weights
__group_459__new_gradients (optional) - T3: New gradients
__group_459__new_moment_1 (optional) - T4: New averaged gradients
__group_459__new_moment_2 (optional) - T4: New averaged squared gradients
__group_459__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_460__new_weights (optional) - T2: New weights
__group_460__new_gradients (optional) - T3: New gradients
__group_460__new_moment_1 (optional) - T4: New averaged gradients
__group_460__new_moment_2 (optional) - T4: New averaged squared gradients
__group_460__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_461__new_weights (optional) - T2: New weights
__group_461__new_gradients (optional) - T3: New gradients
__group_461__new_moment_1 (optional) - T4: New averaged gradients
__group_461__new_moment_2 (optional) - T4: New averaged squared gradients
__group_461__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_462__new_weights (optional) - T2: New weights
__group_462__new_gradients (optional) - T3: New gradients
__group_462__new_moment_1 (optional) - T4: New averaged gradients
__group_462__new_moment_2 (optional) - T4: New averaged squared gradients
__group_462__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_463__new_weights (optional) - T2: New weights
__group_463__new_gradients (optional) - T3: New gradients
__group_463__new_moment_1 (optional) - T4: New averaged gradients
__group_463__new_moment_2 (optional) - T4: New averaged squared gradients
__group_463__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_464__new_weights (optional) - T2: New weights
__group_464__new_gradients (optional) - T3: New gradients
__group_464__new_moment_1 (optional) - T4: New averaged gradients
__group_464__new_moment_2 (optional) - T4: New averaged squared gradients
__group_464__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_465__new_weights (optional) - T2: New weights
__group_465__new_gradients (optional) - T3: New gradients
__group_465__new_moment_1 (optional) - T4: New averaged gradients
__group_465__new_moment_2 (optional) - T4: New averaged squared gradients
__group_465__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_466__new_weights (optional) - T2: New weights
__group_466__new_gradients (optional) - T3: New gradients
__group_466__new_moment_1 (optional) - T4: New averaged gradients
__group_466__new_moment_2 (optional) - T4: New averaged squared gradients
__group_466__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_467__new_weights (optional) - T2: New weights
__group_467__new_gradients (optional) - T3: New gradients
__group_467__new_moment_1 (optional) - T4: New averaged gradients
__group_467__new_moment_2 (optional) - T4: New averaged squared gradients
__group_467__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_468__new_weights (optional) - T2: New weights
__group_468__new_gradients (optional) - T3: New gradients
__group_468__new_moment_1 (optional) - T4: New averaged gradients
__group_468__new_moment_2 (optional) - T4: New averaged squared gradients
__group_468__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_469__new_weights (optional) - T2: New weights
__group_469__new_gradients (optional) - T3: New gradients
__group_469__new_moment_1 (optional) - T4: New averaged gradients
__group_469__new_moment_2 (optional) - T4: New averaged squared gradients
__group_469__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_470__new_weights (optional) - T2: New weights
__group_470__new_gradients (optional) - T3: New gradients
__group_470__new_moment_1 (optional) - T4: New averaged gradients
__group_470__new_moment_2 (optional) - T4: New averaged squared gradients
__group_470__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_471__new_weights (optional) - T2: New weights
__group_471__new_gradients (optional) - T3: New gradients
__group_471__new_moment_1 (optional) - T4: New averaged gradients
__group_471__new_moment_2 (optional) - T4: New averaged squared gradients
__group_471__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_472__new_weights (optional) - T2: New weights
__group_472__new_gradients (optional) - T3: New gradients
__group_472__new_moment_1 (optional) - T4: New averaged gradients
__group_472__new_moment_2 (optional) - T4: New averaged squared gradients
__group_472__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_473__new_weights (optional) - T2: New weights
__group_473__new_gradients (optional) - T3: New gradients
__group_473__new_moment_1 (optional) - T4: New averaged gradients
__group_473__new_moment_2 (optional) - T4: New averaged squared gradients
__group_473__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_474__new_weights (optional) - T2: New weights
__group_474__new_gradients (optional) - T3: New gradients
__group_474__new_moment_1 (optional) - T4: New averaged gradients
__group_474__new_moment_2 (optional) - T4: New averaged squared gradients
__group_474__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_475__new_weights (optional) - T2: New weights
__group_475__new_gradients (optional) - T3: New gradients
__group_475__new_moment_1 (optional) - T4: New averaged gradients
__group_475__new_moment_2 (optional) - T4: New averaged squared gradients
__group_475__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_476__new_weights (optional) - T2: New weights
__group_476__new_gradients (optional) - T3: New gradients
__group_476__new_moment_1 (optional) - T4: New averaged gradients
__group_476__new_moment_2 (optional) - T4: New averaged squared gradients
__group_476__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_477__new_weights (optional) - T2: New weights
__group_477__new_gradients (optional) - T3: New gradients
__group_477__new_moment_1 (optional) - T4: New averaged gradients
__group_477__new_moment_2 (optional) - T4: New averaged squared gradients
__group_477__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_478__new_weights (optional) - T2: New weights
__group_478__new_gradients (optional) - T3: New gradients
__group_478__new_moment_1 (optional) - T4: New averaged gradients
__group_478__new_moment_2 (optional) - T4: New averaged squared gradients
__group_478__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_479__new_weights (optional) - T2: New weights
__group_479__new_gradients (optional) - T3: New gradients
__group_479__new_moment_1 (optional) - T4: New averaged gradients
__group_479__new_moment_2 (optional) - T4: New averaged squared gradients
__group_479__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_480__new_weights (optional) - T2: New weights
__group_480__new_gradients (optional) - T3: New gradients
__group_480__new_moment_1 (optional) - T4: New averaged gradients
__group_480__new_moment_2 (optional) - T4: New averaged squared gradients
__group_480__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_481__new_weights (optional) - T2: New weights
__group_481__new_gradients (optional) - T3: New gradients
__group_481__new_moment_1 (optional) - T4: New averaged gradients
__group_481__new_moment_2 (optional) - T4: New averaged squared gradients
__group_481__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_482__new_weights (optional) - T2: New weights
__group_482__new_gradients (optional) - T3: New gradients
__group_482__new_moment_1 (optional) - T4: New averaged gradients
__group_482__new_moment_2 (optional) - T4: New averaged squared gradients
__group_482__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_483__new_weights (optional) - T2: New weights
__group_483__new_gradients (optional) - T3: New gradients
__group_483__new_moment_1 (optional) - T4: New averaged gradients
__group_483__new_moment_2 (optional) - T4: New averaged squared gradients
__group_483__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_484__new_weights (optional) - T2: New weights
__group_484__new_gradients (optional) - T3: New gradients
__group_484__new_moment_1 (optional) - T4: New averaged gradients
__group_484__new_moment_2 (optional) - T4: New averaged squared gradients
__group_484__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_485__new_weights (optional) - T2: New weights
__group_485__new_gradients (optional) - T3: New gradients
__group_485__new_moment_1 (optional) - T4: New averaged gradients
__group_485__new_moment_2 (optional) - T4: New averaged squared gradients
__group_485__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_486__new_weights (optional) - T2: New weights
__group_486__new_gradients (optional) - T3: New gradients
__group_486__new_moment_1 (optional) - T4: New averaged gradients
__group_486__new_moment_2 (optional) - T4: New averaged squared gradients
__group_486__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_487__new_weights (optional) - T2: New weights
__group_487__new_gradients (optional) - T3: New gradients
__group_487__new_moment_1 (optional) - T4: New averaged gradients
__group_487__new_moment_2 (optional) - T4: New averaged squared gradients
__group_487__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_488__new_weights (optional) - T2: New weights
__group_488__new_gradients (optional) - T3: New gradients
__group_488__new_moment_1 (optional) - T4: New averaged gradients
__group_488__new_moment_2 (optional) - T4: New averaged squared gradients
__group_488__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_489__new_weights (optional) - T2: New weights
__group_489__new_gradients (optional) - T3: New gradients
__group_489__new_moment_1 (optional) - T4: New averaged gradients
__group_489__new_moment_2 (optional) - T4: New averaged squared gradients
__group_489__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_490__new_weights (optional) - T2: New weights
__group_490__new_gradients (optional) - T3: New gradients
__group_490__new_moment_1 (optional) - T4: New averaged gradients
__group_490__new_moment_2 (optional) - T4: New averaged squared gradients
__group_490__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_491__new_weights (optional) - T2: New weights
__group_491__new_gradients (optional) - T3: New gradients
__group_491__new_moment_1 (optional) - T4: New averaged gradients
__group_491__new_moment_2 (optional) - T4: New averaged squared gradients
__group_491__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_492__new_weights (optional) - T2: New weights
__group_492__new_gradients (optional) - T3: New gradients
__group_492__new_moment_1 (optional) - T4: New averaged gradients
__group_492__new_moment_2 (optional) - T4: New averaged squared gradients
__group_492__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_493__new_weights (optional) - T2: New weights
__group_493__new_gradients (optional) - T3: New gradients
__group_493__new_moment_1 (optional) - T4: New averaged gradients
__group_493__new_moment_2 (optional) - T4: New averaged squared gradients
__group_493__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_494__new_weights (optional) - T2: New weights
__group_494__new_gradients (optional) - T3: New gradients
__group_494__new_moment_1 (optional) - T4: New averaged gradients
__group_494__new_moment_2 (optional) - T4: New averaged squared gradients
__group_494__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_495__new_weights (optional) - T2: New weights
__group_495__new_gradients (optional) - T3: New gradients
__group_495__new_moment_1 (optional) - T4: New averaged gradients
__group_495__new_moment_2 (optional) - T4: New averaged squared gradients
__group_495__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_496__new_weights (optional) - T2: New weights
__group_496__new_gradients (optional) - T3: New gradients
__group_496__new_moment_1 (optional) - T4: New averaged gradients
__group_496__new_moment_2 (optional) - T4: New averaged squared gradients
__group_496__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_497__new_weights (optional) - T2: New weights
__group_497__new_gradients (optional) - T3: New gradients
__group_497__new_moment_1 (optional) - T4: New averaged gradients
__group_497__new_moment_2 (optional) - T4: New averaged squared gradients
__group_497__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_498__new_weights (optional) - T2: New weights
__group_498__new_gradients (optional) - T3: New gradients
__group_498__new_moment_1 (optional) - T4: New averaged gradients
__group_498__new_moment_2 (optional) - T4: New averaged squared gradients
__group_498__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_499__new_weights (optional) - T2: New weights
__group_499__new_gradients (optional) - T3: New gradients
__group_499__new_moment_1 (optional) - T4: New averaged gradients
__group_499__new_moment_2 (optional) - T4: New averaged squared gradients
__group_499__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_500__new_weights (optional) - T2: New weights
__group_500__new_gradients (optional) - T3: New gradients
__group_500__new_moment_1 (optional) - T4: New averaged gradients
__group_500__new_moment_2 (optional) - T4: New averaged squared gradients
__group_500__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_501__new_weights (optional) - T2: New weights
__group_501__new_gradients (optional) - T3: New gradients
__group_501__new_moment_1 (optional) - T4: New averaged gradients
__group_501__new_moment_2 (optional) - T4: New averaged squared gradients
__group_501__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_502__new_weights (optional) - T2: New weights
__group_502__new_gradients (optional) - T3: New gradients
__group_502__new_moment_1 (optional) - T4: New averaged gradients
__group_502__new_moment_2 (optional) - T4: New averaged squared gradients
__group_502__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_503__new_weights (optional) - T2: New weights
__group_503__new_gradients (optional) - T3: New gradients
__group_503__new_moment_1 (optional) - T4: New averaged gradients
__group_503__new_moment_2 (optional) - T4: New averaged squared gradients
__group_503__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_504__new_weights (optional) - T2: New weights
__group_504__new_gradients (optional) - T3: New gradients
__group_504__new_moment_1 (optional) - T4: New averaged gradients
__group_504__new_moment_2 (optional) - T4: New averaged squared gradients
__group_504__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_505__new_weights (optional) - T2: New weights
__group_505__new_gradients (optional) - T3: New gradients
__group_505__new_moment_1 (optional) - T4: New averaged gradients
__group_505__new_moment_2 (optional) - T4: New averaged squared gradients
__group_505__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_506__new_weights (optional) - T2: New weights
__group_506__new_gradients (optional) - T3: New gradients
__group_506__new_moment_1 (optional) - T4: New averaged gradients
__group_506__new_moment_2 (optional) - T4: New averaged squared gradients
__group_506__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_507__new_weights (optional) - T2: New weights
__group_507__new_gradients (optional) - T3: New gradients
__group_507__new_moment_1 (optional) - T4: New averaged gradients
__group_507__new_moment_2 (optional) - T4: New averaged squared gradients
__group_507__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_508__new_weights (optional) - T2: New weights
__group_508__new_gradients (optional) - T3: New gradients
__group_508__new_moment_1 (optional) - T4: New averaged gradients
__group_508__new_moment_2 (optional) - T4: New averaged squared gradients
__group_508__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_509__new_weights (optional) - T2: New weights
__group_509__new_gradients (optional) - T3: New gradients
__group_509__new_moment_1 (optional) - T4: New averaged gradients
__group_509__new_moment_2 (optional) - T4: New averaged squared gradients
__group_509__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_510__new_weights (optional) - T2: New weights
__group_510__new_gradients (optional) - T3: New gradients
__group_510__new_moment_1 (optional) - T4: New averaged gradients
__group_510__new_moment_2 (optional) - T4: New averaged squared gradients
__group_510__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_511__new_weights (optional) - T2: New weights
__group_511__new_gradients (optional) - T3: New gradients
__group_511__new_moment_1 (optional) - T4: New averaged gradients
__group_511__new_moment_2 (optional) - T4: New averaged squared gradients
__group_511__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_512__new_weights (optional) - T2: New weights
__group_512__new_gradients (optional) - T3: New gradients
__group_512__new_moment_1 (optional) - T4: New averaged gradients
__group_512__new_moment_2 (optional) - T4: New averaged squared gradients
__group_512__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_513__new_weights (optional) - T2: New weights
__group_513__new_gradients (optional) - T3: New gradients
__group_513__new_moment_1 (optional) - T4: New averaged gradients
__group_513__new_moment_2 (optional) - T4: New averaged squared gradients
__group_513__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_514__new_weights (optional) - T2: New weights
__group_514__new_gradients (optional) - T3: New gradients
__group_514__new_moment_1 (optional) - T4: New averaged gradients
__group_514__new_moment_2 (optional) - T4: New averaged squared gradients
__group_514__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_515__new_weights (optional) - T2: New weights
__group_515__new_gradients (optional) - T3: New gradients
__group_515__new_moment_1 (optional) - T4: New averaged gradients
__group_515__new_moment_2 (optional) - T4: New averaged squared gradients
__group_515__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_516__new_weights (optional) - T2: New weights
__group_516__new_gradients (optional) - T3: New gradients
__group_516__new_moment_1 (optional) - T4: New averaged gradients
__group_516__new_moment_2 (optional) - T4: New averaged squared gradients
__group_516__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_517__new_weights (optional) - T2: New weights
__group_517__new_gradients (optional) - T3: New gradients
__group_517__new_moment_1 (optional) - T4: New averaged gradients
__group_517__new_moment_2 (optional) - T4: New averaged squared gradients
__group_517__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_518__new_weights (optional) - T2: New weights
__group_518__new_gradients (optional) - T3: New gradients
__group_518__new_moment_1 (optional) - T4: New averaged gradients
__group_518__new_moment_2 (optional) - T4: New averaged squared gradients
__group_518__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_519__new_weights (optional) - T2: New weights
__group_519__new_gradients (optional) - T3: New gradients
__group_519__new_moment_1 (optional) - T4: New averaged gradients
__group_519__new_moment_2 (optional) - T4: New averaged squared gradients
__group_519__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_520__new_weights (optional) - T2: New weights
__group_520__new_gradients (optional) - T3: New gradients
__group_520__new_moment_1 (optional) - T4: New averaged gradients
__group_520__new_moment_2 (optional) - T4: New averaged squared gradients
__group_520__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_521__new_weights (optional) - T2: New weights
__group_521__new_gradients (optional) - T3: New gradients
__group_521__new_moment_1 (optional) - T4: New averaged gradients
__group_521__new_moment_2 (optional) - T4: New averaged squared gradients
__group_521__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_522__new_weights (optional) - T2: New weights
__group_522__new_gradients (optional) - T3: New gradients
__group_522__new_moment_1 (optional) - T4: New averaged gradients
__group_522__new_moment_2 (optional) - T4: New averaged squared gradients
__group_522__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_523__new_weights (optional) - T2: New weights
__group_523__new_gradients (optional) - T3: New gradients
__group_523__new_moment_1 (optional) - T4: New averaged gradients
__group_523__new_moment_2 (optional) - T4: New averaged squared gradients
__group_523__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_524__new_weights (optional) - T2: New weights
__group_524__new_gradients (optional) - T3: New gradients
__group_524__new_moment_1 (optional) - T4: New averaged gradients
__group_524__new_moment_2 (optional) - T4: New averaged squared gradients
__group_524__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_525__new_weights (optional) - T2: New weights
__group_525__new_gradients (optional) - T3: New gradients
__group_525__new_moment_1 (optional) - T4: New averaged gradients
__group_525__new_moment_2 (optional) - T4: New averaged squared gradients
__group_525__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_526__new_weights (optional) - T2: New weights
__group_526__new_gradients (optional) - T3: New gradients
__group_526__new_moment_1 (optional) - T4: New averaged gradients
__group_526__new_moment_2 (optional) - T4: New averaged squared gradients
__group_526__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_527__new_weights (optional) - T2: New weights
__group_527__new_gradients (optional) - T3: New gradients
__group_527__new_moment_1 (optional) - T4: New averaged gradients
__group_527__new_moment_2 (optional) - T4: New averaged squared gradients
__group_527__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_528__new_weights (optional) - T2: New weights
__group_528__new_gradients (optional) - T3: New gradients
__group_528__new_moment_1 (optional) - T4: New averaged gradients
__group_528__new_moment_2 (optional) - T4: New averaged squared gradients
__group_528__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_529__new_weights (optional) - T2: New weights
__group_529__new_gradients (optional) - T3: New gradients
__group_529__new_moment_1 (optional) - T4: New averaged gradients
__group_529__new_moment_2 (optional) - T4: New averaged squared gradients
__group_529__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_530__new_weights (optional) - T2: New weights
__group_530__new_gradients (optional) - T3: New gradients
__group_530__new_moment_1 (optional) - T4: New averaged gradients
__group_530__new_moment_2 (optional) - T4: New averaged squared gradients
__group_530__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_531__new_weights (optional) - T2: New weights
__group_531__new_gradients (optional) - T3: New gradients
__group_531__new_moment_1 (optional) - T4: New averaged gradients
__group_531__new_moment_2 (optional) - T4: New averaged squared gradients
__group_531__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_532__new_weights (optional) - T2: New weights
__group_532__new_gradients (optional) - T3: New gradients
__group_532__new_moment_1 (optional) - T4: New averaged gradients
__group_532__new_moment_2 (optional) - T4: New averaged squared gradients
__group_532__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_533__new_weights (optional) - T2: New weights
__group_533__new_gradients (optional) - T3: New gradients
__group_533__new_moment_1 (optional) - T4: New averaged gradients
__group_533__new_moment_2 (optional) - T4: New averaged squared gradients
__group_533__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_534__new_weights (optional) - T2: New weights
__group_534__new_gradients (optional) - T3: New gradients
__group_534__new_moment_1 (optional) - T4: New averaged gradients
__group_534__new_moment_2 (optional) - T4: New averaged squared gradients
__group_534__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_535__new_weights (optional) - T2: New weights
__group_535__new_gradients (optional) - T3: New gradients
__group_535__new_moment_1 (optional) - T4: New averaged gradients
__group_535__new_moment_2 (optional) - T4: New averaged squared gradients
__group_535__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_536__new_weights (optional) - T2: New weights
__group_536__new_gradients (optional) - T3: New gradients
__group_536__new_moment_1 (optional) - T4: New averaged gradients
__group_536__new_moment_2 (optional) - T4: New averaged squared gradients
__group_536__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_537__new_weights (optional) - T2: New weights
__group_537__new_gradients (optional) - T3: New gradients
__group_537__new_moment_1 (optional) - T4: New averaged gradients
__group_537__new_moment_2 (optional) - T4: New averaged squared gradients
__group_537__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_538__new_weights (optional) - T2: New weights
__group_538__new_gradients (optional) - T3: New gradients
__group_538__new_moment_1 (optional) - T4: New averaged gradients
__group_538__new_moment_2 (optional) - T4: New averaged squared gradients
__group_538__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_539__new_weights (optional) - T2: New weights
__group_539__new_gradients (optional) - T3: New gradients
__group_539__new_moment_1 (optional) - T4: New averaged gradients
__group_539__new_moment_2 (optional) - T4: New averaged squared gradients
__group_539__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_540__new_weights (optional) - T2: New weights
__group_540__new_gradients (optional) - T3: New gradients
__group_540__new_moment_1 (optional) - T4: New averaged gradients
__group_540__new_moment_2 (optional) - T4: New averaged squared gradients
__group_540__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_541__new_weights (optional) - T2: New weights
__group_541__new_gradients (optional) - T3: New gradients
__group_541__new_moment_1 (optional) - T4: New averaged gradients
__group_541__new_moment_2 (optional) - T4: New averaged squared gradients
__group_541__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_542__new_weights (optional) - T2: New weights
__group_542__new_gradients (optional) - T3: New gradients
__group_542__new_moment_1 (optional) - T4: New averaged gradients
__group_542__new_moment_2 (optional) - T4: New averaged squared gradients
__group_542__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_543__new_weights (optional) - T2: New weights
__group_543__new_gradients (optional) - T3: New gradients
__group_543__new_moment_1 (optional) - T4: New averaged gradients
__group_543__new_moment_2 (optional) - T4: New averaged squared gradients
__group_543__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_544__new_weights (optional) - T2: New weights
__group_544__new_gradients (optional) - T3: New gradients
__group_544__new_moment_1 (optional) - T4: New averaged gradients
__group_544__new_moment_2 (optional) - T4: New averaged squared gradients
__group_544__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_545__new_weights (optional) - T2: New weights
__group_545__new_gradients (optional) - T3: New gradients
__group_545__new_moment_1 (optional) - T4: New averaged gradients
__group_545__new_moment_2 (optional) - T4: New averaged squared gradients
__group_545__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_546__new_weights (optional) - T2: New weights
__group_546__new_gradients (optional) - T3: New gradients
__group_546__new_moment_1 (optional) - T4: New averaged gradients
__group_546__new_moment_2 (optional) - T4: New averaged squared gradients
__group_546__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_547__new_weights (optional) - T2: New weights
__group_547__new_gradients (optional) - T3: New gradients
__group_547__new_moment_1 (optional) - T4: New averaged gradients
__group_547__new_moment_2 (optional) - T4: New averaged squared gradients
__group_547__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_548__new_weights (optional) - T2: New weights
__group_548__new_gradients (optional) - T3: New gradients
__group_548__new_moment_1 (optional) - T4: New averaged gradients
__group_548__new_moment_2 (optional) - T4: New averaged squared gradients
__group_548__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_549__new_weights (optional) - T2: New weights
__group_549__new_gradients (optional) - T3: New gradients
__group_549__new_moment_1 (optional) - T4: New averaged gradients
__group_549__new_moment_2 (optional) - T4: New averaged squared gradients
__group_549__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_550__new_weights (optional) - T2: New weights
__group_550__new_gradients (optional) - T3: New gradients
__group_550__new_moment_1 (optional) - T4: New averaged gradients
__group_550__new_moment_2 (optional) - T4: New averaged squared gradients
__group_550__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_551__new_weights (optional) - T2: New weights
__group_551__new_gradients (optional) - T3: New gradients
__group_551__new_moment_1 (optional) - T4: New averaged gradients
__group_551__new_moment_2 (optional) - T4: New averaged squared gradients
__group_551__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_552__new_weights (optional) - T2: New weights
__group_552__new_gradients (optional) - T3: New gradients
__group_552__new_moment_1 (optional) - T4: New averaged gradients
__group_552__new_moment_2 (optional) - T4: New averaged squared gradients
__group_552__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_553__new_weights (optional) - T2: New weights
__group_553__new_gradients (optional) - T3: New gradients
__group_553__new_moment_1 (optional) - T4: New averaged gradients
__group_553__new_moment_2 (optional) - T4: New averaged squared gradients
__group_553__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_554__new_weights (optional) - T2: New weights
__group_554__new_gradients (optional) - T3: New gradients
__group_554__new_moment_1 (optional) - T4: New averaged gradients
__group_554__new_moment_2 (optional) - T4: New averaged squared gradients
__group_554__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_555__new_weights (optional) - T2: New weights
__group_555__new_gradients (optional) - T3: New gradients
__group_555__new_moment_1 (optional) - T4: New averaged gradients
__group_555__new_moment_2 (optional) - T4: New averaged squared gradients
__group_555__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_556__new_weights (optional) - T2: New weights
__group_556__new_gradients (optional) - T3: New gradients
__group_556__new_moment_1 (optional) - T4: New averaged gradients
__group_556__new_moment_2 (optional) - T4: New averaged squared gradients
__group_556__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_557__new_weights (optional) - T2: New weights
__group_557__new_gradients (optional) - T3: New gradients
__group_557__new_moment_1 (optional) - T4: New averaged gradients
__group_557__new_moment_2 (optional) - T4: New averaged squared gradients
__group_557__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_558__new_weights (optional) - T2: New weights
__group_558__new_gradients (optional) - T3: New gradients
__group_558__new_moment_1 (optional) - T4: New averaged gradients
__group_558__new_moment_2 (optional) - T4: New averaged squared gradients
__group_558__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_559__new_weights (optional) - T2: New weights
__group_559__new_gradients (optional) - T3: New gradients
__group_559__new_moment_1 (optional) - T4: New averaged gradients
__group_559__new_moment_2 (optional) - T4: New averaged squared gradients
__group_559__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_560__new_weights (optional) - T2: New weights
__group_560__new_gradients (optional) - T3: New gradients
__group_560__new_moment_1 (optional) - T4: New averaged gradients
__group_560__new_moment_2 (optional) - T4: New averaged squared gradients
__group_560__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_561__new_weights (optional) - T2: New weights
__group_561__new_gradients (optional) - T3: New gradients
__group_561__new_moment_1 (optional) - T4: New averaged gradients
__group_561__new_moment_2 (optional) - T4: New averaged squared gradients
__group_561__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_562__new_weights (optional) - T2: New weights
__group_562__new_gradients (optional) - T3: New gradients
__group_562__new_moment_1 (optional) - T4: New averaged gradients
__group_562__new_moment_2 (optional) - T4: New averaged squared gradients
__group_562__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_563__new_weights (optional) - T2: New weights
__group_563__new_gradients (optional) - T3: New gradients
__group_563__new_moment_1 (optional) - T4: New averaged gradients
__group_563__new_moment_2 (optional) - T4: New averaged squared gradients
__group_563__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_564__new_weights (optional) - T2: New weights
__group_564__new_gradients (optional) - T3: New gradients
__group_564__new_moment_1 (optional) - T4: New averaged gradients
__group_564__new_moment_2 (optional) - T4: New averaged squared gradients
__group_564__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_565__new_weights (optional) - T2: New weights
__group_565__new_gradients (optional) - T3: New gradients
__group_565__new_moment_1 (optional) - T4: New averaged gradients
__group_565__new_moment_2 (optional) - T4: New averaged squared gradients
__group_565__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_566__new_weights (optional) - T2: New weights
__group_566__new_gradients (optional) - T3: New gradients
__group_566__new_moment_1 (optional) - T4: New averaged gradients
__group_566__new_moment_2 (optional) - T4: New averaged squared gradients
__group_566__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_567__new_weights (optional) - T2: New weights
__group_567__new_gradients (optional) - T3: New gradients
__group_567__new_moment_1 (optional) - T4: New averaged gradients
__group_567__new_moment_2 (optional) - T4: New averaged squared gradients
__group_567__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_568__new_weights (optional) - T2: New weights
__group_568__new_gradients (optional) - T3: New gradients
__group_568__new_moment_1 (optional) - T4: New averaged gradients
__group_568__new_moment_2 (optional) - T4: New averaged squared gradients
__group_568__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_569__new_weights (optional) - T2: New weights
__group_569__new_gradients (optional) - T3: New gradients
__group_569__new_moment_1 (optional) - T4: New averaged gradients
__group_569__new_moment_2 (optional) - T4: New averaged squared gradients
__group_569__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_570__new_weights (optional) - T2: New weights
__group_570__new_gradients (optional) - T3: New gradients
__group_570__new_moment_1 (optional) - T4: New averaged gradients
__group_570__new_moment_2 (optional) - T4: New averaged squared gradients
__group_570__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_571__new_weights (optional) - T2: New weights
__group_571__new_gradients (optional) - T3: New gradients
__group_571__new_moment_1 (optional) - T4: New averaged gradients
__group_571__new_moment_2 (optional) - T4: New averaged squared gradients
__group_571__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_572__new_weights (optional) - T2: New weights
__group_572__new_gradients (optional) - T3: New gradients
__group_572__new_moment_1 (optional) - T4: New averaged gradients
__group_572__new_moment_2 (optional) - T4: New averaged squared gradients
__group_572__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_573__new_weights (optional) - T2: New weights
__group_573__new_gradients (optional) - T3: New gradients
__group_573__new_moment_1 (optional) - T4: New averaged gradients
__group_573__new_moment_2 (optional) - T4: New averaged squared gradients
__group_573__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_574__new_weights (optional) - T2: New weights
__group_574__new_gradients (optional) - T3: New gradients
__group_574__new_moment_1 (optional) - T4: New averaged gradients
__group_574__new_moment_2 (optional) - T4: New averaged squared gradients
__group_574__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_575__new_weights (optional) - T2: New weights
__group_575__new_gradients (optional) - T3: New gradients
__group_575__new_moment_1 (optional) - T4: New averaged gradients
__group_575__new_moment_2 (optional) - T4: New averaged squared gradients
__group_575__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_576__new_weights (optional) - T2: New weights
__group_576__new_gradients (optional) - T3: New gradients
__group_576__new_moment_1 (optional) - T4: New averaged gradients
__group_576__new_moment_2 (optional) - T4: New averaged squared gradients
__group_576__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_577__new_weights (optional) - T2: New weights
__group_577__new_gradients (optional) - T3: New gradients
__group_577__new_moment_1 (optional) - T4: New averaged gradients
__group_577__new_moment_2 (optional) - T4: New averaged squared gradients
__group_577__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_578__new_weights (optional) - T2: New weights
__group_578__new_gradients (optional) - T3: New gradients
__group_578__new_moment_1 (optional) - T4: New averaged gradients
__group_578__new_moment_2 (optional) - T4: New averaged squared gradients
__group_578__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_579__new_weights (optional) - T2: New weights
__group_579__new_gradients (optional) - T3: New gradients
__group_579__new_moment_1 (optional) - T4: New averaged gradients
__group_579__new_moment_2 (optional) - T4: New averaged squared gradients
__group_579__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_580__new_weights (optional) - T2: New weights
__group_580__new_gradients (optional) - T3: New gradients
__group_580__new_moment_1 (optional) - T4: New averaged gradients
__group_580__new_moment_2 (optional) - T4: New averaged squared gradients
__group_580__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_581__new_weights (optional) - T2: New weights
__group_581__new_gradients (optional) - T3: New gradients
__group_581__new_moment_1 (optional) - T4: New averaged gradients
__group_581__new_moment_2 (optional) - T4: New averaged squared gradients
__group_581__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_582__new_weights (optional) - T2: New weights
__group_582__new_gradients (optional) - T3: New gradients
__group_582__new_moment_1 (optional) - T4: New averaged gradients
__group_582__new_moment_2 (optional) - T4: New averaged squared gradients
__group_582__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_583__new_weights (optional) - T2: New weights
__group_583__new_gradients (optional) - T3: New gradients
__group_583__new_moment_1 (optional) - T4: New averaged gradients
__group_583__new_moment_2 (optional) - T4: New averaged squared gradients
__group_583__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_584__new_weights (optional) - T2: New weights
__group_584__new_gradients (optional) - T3: New gradients
__group_584__new_moment_1 (optional) - T4: New averaged gradients
__group_584__new_moment_2 (optional) - T4: New averaged squared gradients
__group_584__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_585__new_weights (optional) - T2: New weights
__group_585__new_gradients (optional) - T3: New gradients
__group_585__new_moment_1 (optional) - T4: New averaged gradients
__group_585__new_moment_2 (optional) - T4: New averaged squared gradients
__group_585__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_586__new_weights (optional) - T2: New weights
__group_586__new_gradients (optional) - T3: New gradients
__group_586__new_moment_1 (optional) - T4: New averaged gradients
__group_586__new_moment_2 (optional) - T4: New averaged squared gradients
__group_586__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_587__new_weights (optional) - T2: New weights
__group_587__new_gradients (optional) - T3: New gradients
__group_587__new_moment_1 (optional) - T4: New averaged gradients
__group_587__new_moment_2 (optional) - T4: New averaged squared gradients
__group_587__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_588__new_weights (optional) - T2: New weights
__group_588__new_gradients (optional) - T3: New gradients
__group_588__new_moment_1 (optional) - T4: New averaged gradients
__group_588__new_moment_2 (optional) - T4: New averaged squared gradients
__group_588__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_589__new_weights (optional) - T2: New weights
__group_589__new_gradients (optional) - T3: New gradients
__group_589__new_moment_1 (optional) - T4: New averaged gradients
__group_589__new_moment_2 (optional) - T4: New averaged squared gradients
__group_589__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_590__new_weights (optional) - T2: New weights
__group_590__new_gradients (optional) - T3: New gradients
__group_590__new_moment_1 (optional) - T4: New averaged gradients
__group_590__new_moment_2 (optional) - T4: New averaged squared gradients
__group_590__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_591__new_weights (optional) - T2: New weights
__group_591__new_gradients (optional) - T3: New gradients
__group_591__new_moment_1 (optional) - T4: New averaged gradients
__group_591__new_moment_2 (optional) - T4: New averaged squared gradients
__group_591__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_592__new_weights (optional) - T2: New weights
__group_592__new_gradients (optional) - T3: New gradients
__group_592__new_moment_1 (optional) - T4: New averaged gradients
__group_592__new_moment_2 (optional) - T4: New averaged squared gradients
__group_592__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_593__new_weights (optional) - T2: New weights
__group_593__new_gradients (optional) - T3: New gradients
__group_593__new_moment_1 (optional) - T4: New averaged gradients
__group_593__new_moment_2 (optional) - T4: New averaged squared gradients
__group_593__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_594__new_weights (optional) - T2: New weights
__group_594__new_gradients (optional) - T3: New gradients
__group_594__new_moment_1 (optional) - T4: New averaged gradients
__group_594__new_moment_2 (optional) - T4: New averaged squared gradients
__group_594__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_595__new_weights (optional) - T2: New weights
__group_595__new_gradients (optional) - T3: New gradients
__group_595__new_moment_1 (optional) - T4: New averaged gradients
__group_595__new_moment_2 (optional) - T4: New averaged squared gradients
__group_595__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_596__new_weights (optional) - T2: New weights
__group_596__new_gradients (optional) - T3: New gradients
__group_596__new_moment_1 (optional) - T4: New averaged gradients
__group_596__new_moment_2 (optional) - T4: New averaged squared gradients
__group_596__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_597__new_weights (optional) - T2: New weights
__group_597__new_gradients (optional) - T3: New gradients
__group_597__new_moment_1 (optional) - T4: New averaged gradients
__group_597__new_moment_2 (optional) - T4: New averaged squared gradients
__group_597__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_598__new_weights (optional) - T2: New weights
__group_598__new_gradients (optional) - T3: New gradients
__group_598__new_moment_1 (optional) - T4: New averaged gradients
__group_598__new_moment_2 (optional) - T4: New averaged squared gradients
__group_598__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_599__new_weights (optional) - T2: New weights
__group_599__new_gradients (optional) - T3: New gradients
__group_599__new_moment_1 (optional) - T4: New averaged gradients
__group_599__new_moment_2 (optional) - T4: New averaged squared gradients
__group_599__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_600__new_weights (optional) - T2: New weights
__group_600__new_gradients (optional) - T3: New gradients
__group_600__new_moment_1 (optional) - T4: New averaged gradients
__group_600__new_moment_2 (optional) - T4: New averaged squared gradients
__group_600__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_601__new_weights (optional) - T2: New weights
__group_601__new_gradients (optional) - T3: New gradients
__group_601__new_moment_1 (optional) - T4: New averaged gradients
__group_601__new_moment_2 (optional) - T4: New averaged squared gradients
__group_601__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_602__new_weights (optional) - T2: New weights
__group_602__new_gradients (optional) - T3: New gradients
__group_602__new_moment_1 (optional) - T4: New averaged gradients
__group_602__new_moment_2 (optional) - T4: New averaged squared gradients
__group_602__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_603__new_weights (optional) - T2: New weights
__group_603__new_gradients (optional) - T3: New gradients
__group_603__new_moment_1 (optional) - T4: New averaged gradients
__group_603__new_moment_2 (optional) - T4: New averaged squared gradients
__group_603__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_604__new_weights (optional) - T2: New weights
__group_604__new_gradients (optional) - T3: New gradients
__group_604__new_moment_1 (optional) - T4: New averaged gradients
__group_604__new_moment_2 (optional) - T4: New averaged squared gradients
__group_604__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_605__new_weights (optional) - T2: New weights
__group_605__new_gradients (optional) - T3: New gradients
__group_605__new_moment_1 (optional) - T4: New averaged gradients
__group_605__new_moment_2 (optional) - T4: New averaged squared gradients
__group_605__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_606__new_weights (optional) - T2: New weights
__group_606__new_gradients (optional) - T3: New gradients
__group_606__new_moment_1 (optional) - T4: New averaged gradients
__group_606__new_moment_2 (optional) - T4: New averaged squared gradients
__group_606__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_607__new_weights (optional) - T2: New weights
__group_607__new_gradients (optional) - T3: New gradients
__group_607__new_moment_1 (optional) - T4: New averaged gradients
__group_607__new_moment_2 (optional) - T4: New averaged squared gradients
__group_607__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_608__new_weights (optional) - T2: New weights
__group_608__new_gradients (optional) - T3: New gradients
__group_608__new_moment_1 (optional) - T4: New averaged gradients
__group_608__new_moment_2 (optional) - T4: New averaged squared gradients
__group_608__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_609__new_weights (optional) - T2: New weights
__group_609__new_gradients (optional) - T3: New gradients
__group_609__new_moment_1 (optional) - T4: New averaged gradients
__group_609__new_moment_2 (optional) - T4: New averaged squared gradients
__group_609__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_610__new_weights (optional) - T2: New weights
__group_610__new_gradients (optional) - T3: New gradients
__group_610__new_moment_1 (optional) - T4: New averaged gradients
__group_610__new_moment_2 (optional) - T4: New averaged squared gradients
__group_610__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_611__new_weights (optional) - T2: New weights
__group_611__new_gradients (optional) - T3: New gradients
__group_611__new_moment_1 (optional) - T4: New averaged gradients
__group_611__new_moment_2 (optional) - T4: New averaged squared gradients
__group_611__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_612__new_weights (optional) - T2: New weights
__group_612__new_gradients (optional) - T3: New gradients
__group_612__new_moment_1 (optional) - T4: New averaged gradients
__group_612__new_moment_2 (optional) - T4: New averaged squared gradients
__group_612__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_613__new_weights (optional) - T2: New weights
__group_613__new_gradients (optional) - T3: New gradients
__group_613__new_moment_1 (optional) - T4: New averaged gradients
__group_613__new_moment_2 (optional) - T4: New averaged squared gradients
__group_613__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_614__new_weights (optional) - T2: New weights
__group_614__new_gradients (optional) - T3: New gradients
__group_614__new_moment_1 (optional) - T4: New averaged gradients
__group_614__new_moment_2 (optional) - T4: New averaged squared gradients
__group_614__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_615__new_weights (optional) - T2: New weights
__group_615__new_gradients (optional) - T3: New gradients
__group_615__new_moment_1 (optional) - T4: New averaged gradients
__group_615__new_moment_2 (optional) - T4: New averaged squared gradients
__group_615__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_616__new_weights (optional) - T2: New weights
__group_616__new_gradients (optional) - T3: New gradients
__group_616__new_moment_1 (optional) - T4: New averaged gradients
__group_616__new_moment_2 (optional) - T4: New averaged squared gradients
__group_616__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_617__new_weights (optional) - T2: New weights
__group_617__new_gradients (optional) - T3: New gradients
__group_617__new_moment_1 (optional) - T4: New averaged gradients
__group_617__new_moment_2 (optional) - T4: New averaged squared gradients
__group_617__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_618__new_weights (optional) - T2: New weights
__group_618__new_gradients (optional) - T3: New gradients
__group_618__new_moment_1 (optional) - T4: New averaged gradients
__group_618__new_moment_2 (optional) - T4: New averaged squared gradients
__group_618__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_619__new_weights (optional) - T2: New weights
__group_619__new_gradients (optional) - T3: New gradients
__group_619__new_moment_1 (optional) - T4: New averaged gradients
__group_619__new_moment_2 (optional) - T4: New averaged squared gradients
__group_619__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_620__new_weights (optional) - T2: New weights
__group_620__new_gradients (optional) - T3: New gradients
__group_620__new_moment_1 (optional) - T4: New averaged gradients
__group_620__new_moment_2 (optional) - T4: New averaged squared gradients
__group_620__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_621__new_weights (optional) - T2: New weights
__group_621__new_gradients (optional) - T3: New gradients
__group_621__new_moment_1 (optional) - T4: New averaged gradients
__group_621__new_moment_2 (optional) - T4: New averaged squared gradients
__group_621__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_622__new_weights (optional) - T2: New weights
__group_622__new_gradients (optional) - T3: New gradients
__group_622__new_moment_1 (optional) - T4: New averaged gradients
__group_622__new_moment_2 (optional) - T4: New averaged squared gradients
__group_622__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_623__new_weights (optional) - T2: New weights
__group_623__new_gradients (optional) - T3: New gradients
__group_623__new_moment_1 (optional) - T4: New averaged gradients
__group_623__new_moment_2 (optional) - T4: New averaged squared gradients
__group_623__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_624__new_weights (optional) - T2: New weights
__group_624__new_gradients (optional) - T3: New gradients
__group_624__new_moment_1 (optional) - T4: New averaged gradients
__group_624__new_moment_2 (optional) - T4: New averaged squared gradients
__group_624__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_625__new_weights (optional) - T2: New weights
__group_625__new_gradients (optional) - T3: New gradients
__group_625__new_moment_1 (optional) - T4: New averaged gradients
__group_625__new_moment_2 (optional) - T4: New averaged squared gradients
__group_625__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_626__new_weights (optional) - T2: New weights
__group_626__new_gradients (optional) - T3: New gradients
__group_626__new_moment_1 (optional) - T4: New averaged gradients
__group_626__new_moment_2 (optional) - T4: New averaged squared gradients
__group_626__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_627__new_weights (optional) - T2: New weights
__group_627__new_gradients (optional) - T3: New gradients
__group_627__new_moment_1 (optional) - T4: New averaged gradients
__group_627__new_moment_2 (optional) - T4: New averaged squared gradients
__group_627__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_628__new_weights (optional) - T2: New weights
__group_628__new_gradients (optional) - T3: New gradients
__group_628__new_moment_1 (optional) - T4: New averaged gradients
__group_628__new_moment_2 (optional) - T4: New averaged squared gradients
__group_628__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_629__new_weights (optional) - T2: New weights
__group_629__new_gradients (optional) - T3: New gradients
__group_629__new_moment_1 (optional) - T4: New averaged gradients
__group_629__new_moment_2 (optional) - T4: New averaged squared gradients
__group_629__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_630__new_weights (optional) - T2: New weights
__group_630__new_gradients (optional) - T3: New gradients
__group_630__new_moment_1 (optional) - T4: New averaged gradients
__group_630__new_moment_2 (optional) - T4: New averaged squared gradients
__group_630__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_631__new_weights (optional) - T2: New weights
__group_631__new_gradients (optional) - T3: New gradients
__group_631__new_moment_1 (optional) - T4: New averaged gradients
__group_631__new_moment_2 (optional) - T4: New averaged squared gradients
__group_631__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_632__new_weights (optional) - T2: New weights
__group_632__new_gradients (optional) - T3: New gradients
__group_632__new_moment_1 (optional) - T4: New averaged gradients
__group_632__new_moment_2 (optional) - T4: New averaged squared gradients
__group_632__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_633__new_weights (optional) - T2: New weights
__group_633__new_gradients (optional) - T3: New gradients
__group_633__new_moment_1 (optional) - T4: New averaged gradients
__group_633__new_moment_2 (optional) - T4: New averaged squared gradients
__group_633__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_634__new_weights (optional) - T2: New weights
__group_634__new_gradients (optional) - T3: New gradients
__group_634__new_moment_1 (optional) - T4: New averaged gradients
__group_634__new_moment_2 (optional) - T4: New averaged squared gradients
__group_634__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_635__new_weights (optional) - T2: New weights
__group_635__new_gradients (optional) - T3: New gradients
__group_635__new_moment_1 (optional) - T4: New averaged gradients
__group_635__new_moment_2 (optional) - T4: New averaged squared gradients
__group_635__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_636__new_weights (optional) - T2: New weights
__group_636__new_gradients (optional) - T3: New gradients
__group_636__new_moment_1 (optional) - T4: New averaged gradients
__group_636__new_moment_2 (optional) - T4: New averaged squared gradients
__group_636__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_637__new_weights (optional) - T2: New weights
__group_637__new_gradients (optional) - T3: New gradients
__group_637__new_moment_1 (optional) - T4: New averaged gradients
__group_637__new_moment_2 (optional) - T4: New averaged squared gradients
__group_637__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_638__new_weights (optional) - T2: New weights
__group_638__new_gradients (optional) - T3: New gradients
__group_638__new_moment_1 (optional) - T4: New averaged gradients
__group_638__new_moment_2 (optional) - T4: New averaged squared gradients
__group_638__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_639__new_weights (optional) - T2: New weights
__group_639__new_gradients (optional) - T3: New gradients
__group_639__new_moment_1 (optional) - T4: New averaged gradients
__group_639__new_moment_2 (optional) - T4: New averaged squared gradients
__group_639__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_640__new_weights (optional) - T2: New weights
__group_640__new_gradients (optional) - T3: New gradients
__group_640__new_moment_1 (optional) - T4: New averaged gradients
__group_640__new_moment_2 (optional) - T4: New averaged squared gradients
__group_640__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_641__new_weights (optional) - T2: New weights
__group_641__new_gradients (optional) - T3: New gradients
__group_641__new_moment_1 (optional) - T4: New averaged gradients
__group_641__new_moment_2 (optional) - T4: New averaged squared gradients
__group_641__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_642__new_weights (optional) - T2: New weights
__group_642__new_gradients (optional) - T3: New gradients
__group_642__new_moment_1 (optional) - T4: New averaged gradients
__group_642__new_moment_2 (optional) - T4: New averaged squared gradients
__group_642__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_643__new_weights (optional) - T2: New weights
__group_643__new_gradients (optional) - T3: New gradients
__group_643__new_moment_1 (optional) - T4: New averaged gradients
__group_643__new_moment_2 (optional) - T4: New averaged squared gradients
__group_643__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_644__new_weights (optional) - T2: New weights
__group_644__new_gradients (optional) - T3: New gradients
__group_644__new_moment_1 (optional) - T4: New averaged gradients
__group_644__new_moment_2 (optional) - T4: New averaged squared gradients
__group_644__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_645__new_weights (optional) - T2: New weights
__group_645__new_gradients (optional) - T3: New gradients
__group_645__new_moment_1 (optional) - T4: New averaged gradients
__group_645__new_moment_2 (optional) - T4: New averaged squared gradients
__group_645__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_646__new_weights (optional) - T2: New weights
__group_646__new_gradients (optional) - T3: New gradients
__group_646__new_moment_1 (optional) - T4: New averaged gradients
__group_646__new_moment_2 (optional) - T4: New averaged squared gradients
__group_646__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_647__new_weights (optional) - T2: New weights
__group_647__new_gradients (optional) - T3: New gradients
__group_647__new_moment_1 (optional) - T4: New averaged gradients
__group_647__new_moment_2 (optional) - T4: New averaged squared gradients
__group_647__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_648__new_weights (optional) - T2: New weights
__group_648__new_gradients (optional) - T3: New gradients
__group_648__new_moment_1 (optional) - T4: New averaged gradients
__group_648__new_moment_2 (optional) - T4: New averaged squared gradients
__group_648__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_649__new_weights (optional) - T2: New weights
__group_649__new_gradients (optional) - T3: New gradients
__group_649__new_moment_1 (optional) - T4: New averaged gradients
__group_649__new_moment_2 (optional) - T4: New averaged squared gradients
__group_649__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_650__new_weights (optional) - T2: New weights
__group_650__new_gradients (optional) - T3: New gradients
__group_650__new_moment_1 (optional) - T4: New averaged gradients
__group_650__new_moment_2 (optional) - T4: New averaged squared gradients
__group_650__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_651__new_weights (optional) - T2: New weights
__group_651__new_gradients (optional) - T3: New gradients
__group_651__new_moment_1 (optional) - T4: New averaged gradients
__group_651__new_moment_2 (optional) - T4: New averaged squared gradients
__group_651__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_652__new_weights (optional) - T2: New weights
__group_652__new_gradients (optional) - T3: New gradients
__group_652__new_moment_1 (optional) - T4: New averaged gradients
__group_652__new_moment_2 (optional) - T4: New averaged squared gradients
__group_652__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_653__new_weights (optional) - T2: New weights
__group_653__new_gradients (optional) - T3: New gradients
__group_653__new_moment_1 (optional) - T4: New averaged gradients
__group_653__new_moment_2 (optional) - T4: New averaged squared gradients
__group_653__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_654__new_weights (optional) - T2: New weights
__group_654__new_gradients (optional) - T3: New gradients
__group_654__new_moment_1 (optional) - T4: New averaged gradients
__group_654__new_moment_2 (optional) - T4: New averaged squared gradients
__group_654__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_655__new_weights (optional) - T2: New weights
__group_655__new_gradients (optional) - T3: New gradients
__group_655__new_moment_1 (optional) - T4: New averaged gradients
__group_655__new_moment_2 (optional) - T4: New averaged squared gradients
__group_655__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_656__new_weights (optional) - T2: New weights
__group_656__new_gradients (optional) - T3: New gradients
__group_656__new_moment_1 (optional) - T4: New averaged gradients
__group_656__new_moment_2 (optional) - T4: New averaged squared gradients
__group_656__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_657__new_weights (optional) - T2: New weights
__group_657__new_gradients (optional) - T3: New gradients
__group_657__new_moment_1 (optional) - T4: New averaged gradients
__group_657__new_moment_2 (optional) - T4: New averaged squared gradients
__group_657__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_658__new_weights (optional) - T2: New weights
__group_658__new_gradients (optional) - T3: New gradients
__group_658__new_moment_1 (optional) - T4: New averaged gradients
__group_658__new_moment_2 (optional) - T4: New averaged squared gradients
__group_658__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_659__new_weights (optional) - T2: New weights
__group_659__new_gradients (optional) - T3: New gradients
__group_659__new_moment_1 (optional) - T4: New averaged gradients
__group_659__new_moment_2 (optional) - T4: New averaged squared gradients
__group_659__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_660__new_weights (optional) - T2: New weights
__group_660__new_gradients (optional) - T3: New gradients
__group_660__new_moment_1 (optional) - T4: New averaged gradients
__group_660__new_moment_2 (optional) - T4: New averaged squared gradients
__group_660__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_661__new_weights (optional) - T2: New weights
__group_661__new_gradients (optional) - T3: New gradients
__group_661__new_moment_1 (optional) - T4: New averaged gradients
__group_661__new_moment_2 (optional) - T4: New averaged squared gradients
__group_661__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_662__new_weights (optional) - T2: New weights
__group_662__new_gradients (optional) - T3: New gradients
__group_662__new_moment_1 (optional) - T4: New averaged gradients
__group_662__new_moment_2 (optional) - T4: New averaged squared gradients
__group_662__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_663__new_weights (optional) - T2: New weights
__group_663__new_gradients (optional) - T3: New gradients
__group_663__new_moment_1 (optional) - T4: New averaged gradients
__group_663__new_moment_2 (optional) - T4: New averaged squared gradients
__group_663__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_664__new_weights (optional) - T2: New weights
__group_664__new_gradients (optional) - T3: New gradients
__group_664__new_moment_1 (optional) - T4: New averaged gradients
__group_664__new_moment_2 (optional) - T4: New averaged squared gradients
__group_664__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_665__new_weights (optional) - T2: New weights
__group_665__new_gradients (optional) - T3: New gradients
__group_665__new_moment_1 (optional) - T4: New averaged gradients
__group_665__new_moment_2 (optional) - T4: New averaged squared gradients
__group_665__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_666__new_weights (optional) - T2: New weights
__group_666__new_gradients (optional) - T3: New gradients
__group_666__new_moment_1 (optional) - T4: New averaged gradients
__group_666__new_moment_2 (optional) - T4: New averaged squared gradients
__group_666__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_667__new_weights (optional) - T2: New weights
__group_667__new_gradients (optional) - T3: New gradients
__group_667__new_moment_1 (optional) - T4: New averaged gradients
__group_667__new_moment_2 (optional) - T4: New averaged squared gradients
__group_667__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_668__new_weights (optional) - T2: New weights
__group_668__new_gradients (optional) - T3: New gradients
__group_668__new_moment_1 (optional) - T4: New averaged gradients
__group_668__new_moment_2 (optional) - T4: New averaged squared gradients
__group_668__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_669__new_weights (optional) - T2: New weights
__group_669__new_gradients (optional) - T3: New gradients
__group_669__new_moment_1 (optional) - T4: New averaged gradients
__group_669__new_moment_2 (optional) - T4: New averaged squared gradients
__group_669__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_670__new_weights (optional) - T2: New weights
__group_670__new_gradients (optional) - T3: New gradients
__group_670__new_moment_1 (optional) - T4: New averaged gradients
__group_670__new_moment_2 (optional) - T4: New averaged squared gradients
__group_670__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_671__new_weights (optional) - T2: New weights
__group_671__new_gradients (optional) - T3: New gradients
__group_671__new_moment_1 (optional) - T4: New averaged gradients
__group_671__new_moment_2 (optional) - T4: New averaged squared gradients
__group_671__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_672__new_weights (optional) - T2: New weights
__group_672__new_gradients (optional) - T3: New gradients
__group_672__new_moment_1 (optional) - T4: New averaged gradients
__group_672__new_moment_2 (optional) - T4: New averaged squared gradients
__group_672__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_673__new_weights (optional) - T2: New weights
__group_673__new_gradients (optional) - T3: New gradients
__group_673__new_moment_1 (optional) - T4: New averaged gradients
__group_673__new_moment_2 (optional) - T4: New averaged squared gradients
__group_673__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_674__new_weights (optional) - T2: New weights
__group_674__new_gradients (optional) - T3: New gradients
__group_674__new_moment_1 (optional) - T4: New averaged gradients
__group_674__new_moment_2 (optional) - T4: New averaged squared gradients
__group_674__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_675__new_weights (optional) - T2: New weights
__group_675__new_gradients (optional) - T3: New gradients
__group_675__new_moment_1 (optional) - T4: New averaged gradients
__group_675__new_moment_2 (optional) - T4: New averaged squared gradients
__group_675__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_676__new_weights (optional) - T2: New weights
__group_676__new_gradients (optional) - T3: New gradients
__group_676__new_moment_1 (optional) - T4: New averaged gradients
__group_676__new_moment_2 (optional) - T4: New averaged squared gradients
__group_676__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_677__new_weights (optional) - T2: New weights
__group_677__new_gradients (optional) - T3: New gradients
__group_677__new_moment_1 (optional) - T4: New averaged gradients
__group_677__new_moment_2 (optional) - T4: New averaged squared gradients
__group_677__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_678__new_weights (optional) - T2: New weights
__group_678__new_gradients (optional) - T3: New gradients
__group_678__new_moment_1 (optional) - T4: New averaged gradients
__group_678__new_moment_2 (optional) - T4: New averaged squared gradients
__group_678__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_679__new_weights (optional) - T2: New weights
__group_679__new_gradients (optional) - T3: New gradients
__group_679__new_moment_1 (optional) - T4: New averaged gradients
__group_679__new_moment_2 (optional) - T4: New averaged squared gradients
__group_679__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_680__new_weights (optional) - T2: New weights
__group_680__new_gradients (optional) - T3: New gradients
__group_680__new_moment_1 (optional) - T4: New averaged gradients
__group_680__new_moment_2 (optional) - T4: New averaged squared gradients
__group_680__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_681__new_weights (optional) - T2: New weights
__group_681__new_gradients (optional) - T3: New gradients
__group_681__new_moment_1 (optional) - T4: New averaged gradients
__group_681__new_moment_2 (optional) - T4: New averaged squared gradients
__group_681__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_682__new_weights (optional) - T2: New weights
__group_682__new_gradients (optional) - T3: New gradients
__group_682__new_moment_1 (optional) - T4: New averaged gradients
__group_682__new_moment_2 (optional) - T4: New averaged squared gradients
__group_682__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_683__new_weights (optional) - T2: New weights
__group_683__new_gradients (optional) - T3: New gradients
__group_683__new_moment_1 (optional) - T4: New averaged gradients
__group_683__new_moment_2 (optional) - T4: New averaged squared gradients
__group_683__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_684__new_weights (optional) - T2: New weights
__group_684__new_gradients (optional) - T3: New gradients
__group_684__new_moment_1 (optional) - T4: New averaged gradients
__group_684__new_moment_2 (optional) - T4: New averaged squared gradients
__group_684__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_685__new_weights (optional) - T2: New weights
__group_685__new_gradients (optional) - T3: New gradients
__group_685__new_moment_1 (optional) - T4: New averaged gradients
__group_685__new_moment_2 (optional) - T4: New averaged squared gradients
__group_685__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_686__new_weights (optional) - T2: New weights
__group_686__new_gradients (optional) - T3: New gradients
__group_686__new_moment_1 (optional) - T4: New averaged gradients
__group_686__new_moment_2 (optional) - T4: New averaged squared gradients
__group_686__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_687__new_weights (optional) - T2: New weights
__group_687__new_gradients (optional) - T3: New gradients
__group_687__new_moment_1 (optional) - T4: New averaged gradients
__group_687__new_moment_2 (optional) - T4: New averaged squared gradients
__group_687__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_688__new_weights (optional) - T2: New weights
__group_688__new_gradients (optional) - T3: New gradients
__group_688__new_moment_1 (optional) - T4: New averaged gradients
__group_688__new_moment_2 (optional) - T4: New averaged squared gradients
__group_688__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_689__new_weights (optional) - T2: New weights
__group_689__new_gradients (optional) - T3: New gradients
__group_689__new_moment_1 (optional) - T4: New averaged gradients
__group_689__new_moment_2 (optional) - T4: New averaged squared gradients
__group_689__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_690__new_weights (optional) - T2: New weights
__group_690__new_gradients (optional) - T3: New gradients
__group_690__new_moment_1 (optional) - T4: New averaged gradients
__group_690__new_moment_2 (optional) - T4: New averaged squared gradients
__group_690__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_691__new_weights (optional) - T2: New weights
__group_691__new_gradients (optional) - T3: New gradients
__group_691__new_moment_1 (optional) - T4: New averaged gradients
__group_691__new_moment_2 (optional) - T4: New averaged squared gradients
__group_691__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_692__new_weights (optional) - T2: New weights
__group_692__new_gradients (optional) - T3: New gradients
__group_692__new_moment_1 (optional) - T4: New averaged gradients
__group_692__new_moment_2 (optional) - T4: New averaged squared gradients
__group_692__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_693__new_weights (optional) - T2: New weights
__group_693__new_gradients (optional) - T3: New gradients
__group_693__new_moment_1 (optional) - T4: New averaged gradients
__group_693__new_moment_2 (optional) - T4: New averaged squared gradients
__group_693__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_694__new_weights (optional) - T2: New weights
__group_694__new_gradients (optional) - T3: New gradients
__group_694__new_moment_1 (optional) - T4: New averaged gradients
__group_694__new_moment_2 (optional) - T4: New averaged squared gradients
__group_694__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_695__new_weights (optional) - T2: New weights
__group_695__new_gradients (optional) - T3: New gradients
__group_695__new_moment_1 (optional) - T4: New averaged gradients
__group_695__new_moment_2 (optional) - T4: New averaged squared gradients
__group_695__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_696__new_weights (optional) - T2: New weights
__group_696__new_gradients (optional) - T3: New gradients
__group_696__new_moment_1 (optional) - T4: New averaged gradients
__group_696__new_moment_2 (optional) - T4: New averaged squared gradients
__group_696__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_697__new_weights (optional) - T2: New weights
__group_697__new_gradients (optional) - T3: New gradients
__group_697__new_moment_1 (optional) - T4: New averaged gradients
__group_697__new_moment_2 (optional) - T4: New averaged squared gradients
__group_697__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_698__new_weights (optional) - T2: New weights
__group_698__new_gradients (optional) - T3: New gradients
__group_698__new_moment_1 (optional) - T4: New averaged gradients
__group_698__new_moment_2 (optional) - T4: New averaged squared gradients
__group_698__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_699__new_weights (optional) - T2: New weights
__group_699__new_gradients (optional) - T3: New gradients
__group_699__new_moment_1 (optional) - T4: New averaged gradients
__group_699__new_moment_2 (optional) - T4: New averaged squared gradients
__group_699__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_700__new_weights (optional) - T2: New weights
__group_700__new_gradients (optional) - T3: New gradients
__group_700__new_moment_1 (optional) - T4: New averaged gradients
__group_700__new_moment_2 (optional) - T4: New averaged squared gradients
__group_700__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_701__new_weights (optional) - T2: New weights
__group_701__new_gradients (optional) - T3: New gradients
__group_701__new_moment_1 (optional) - T4: New averaged gradients
__group_701__new_moment_2 (optional) - T4: New averaged squared gradients
__group_701__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_702__new_weights (optional) - T2: New weights
__group_702__new_gradients (optional) - T3: New gradients
__group_702__new_moment_1 (optional) - T4: New averaged gradients
__group_702__new_moment_2 (optional) - T4: New averaged squared gradients
__group_702__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_703__new_weights (optional) - T2: New weights
__group_703__new_gradients (optional) - T3: New gradients
__group_703__new_moment_1 (optional) - T4: New averaged gradients
__group_703__new_moment_2 (optional) - T4: New averaged squared gradients
__group_703__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_704__new_weights (optional) - T2: New weights
__group_704__new_gradients (optional) - T3: New gradients
__group_704__new_moment_1 (optional) - T4: New averaged gradients
__group_704__new_moment_2 (optional) - T4: New averaged squared gradients
__group_704__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_705__new_weights (optional) - T2: New weights
__group_705__new_gradients (optional) - T3: New gradients
__group_705__new_moment_1 (optional) - T4: New averaged gradients
__group_705__new_moment_2 (optional) - T4: New averaged squared gradients
__group_705__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_706__new_weights (optional) - T2: New weights
__group_706__new_gradients (optional) - T3: New gradients
__group_706__new_moment_1 (optional) - T4: New averaged gradients
__group_706__new_moment_2 (optional) - T4: New averaged squared gradients
__group_706__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_707__new_weights (optional) - T2: New weights
__group_707__new_gradients (optional) - T3: New gradients
__group_707__new_moment_1 (optional) - T4: New averaged gradients
__group_707__new_moment_2 (optional) - T4: New averaged squared gradients
__group_707__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_708__new_weights (optional) - T2: New weights
__group_708__new_gradients (optional) - T3: New gradients
__group_708__new_moment_1 (optional) - T4: New averaged gradients
__group_708__new_moment_2 (optional) - T4: New averaged squared gradients
__group_708__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_709__new_weights (optional) - T2: New weights
__group_709__new_gradients (optional) - T3: New gradients
__group_709__new_moment_1 (optional) - T4: New averaged gradients
__group_709__new_moment_2 (optional) - T4: New averaged squared gradients
__group_709__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_710__new_weights (optional) - T2: New weights
__group_710__new_gradients (optional) - T3: New gradients
__group_710__new_moment_1 (optional) - T4: New averaged gradients
__group_710__new_moment_2 (optional) - T4: New averaged squared gradients
__group_710__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_711__new_weights (optional) - T2: New weights
__group_711__new_gradients (optional) - T3: New gradients
__group_711__new_moment_1 (optional) - T4: New averaged gradients
__group_711__new_moment_2 (optional) - T4: New averaged squared gradients
__group_711__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_712__new_weights (optional) - T2: New weights
__group_712__new_gradients (optional) - T3: New gradients
__group_712__new_moment_1 (optional) - T4: New averaged gradients
__group_712__new_moment_2 (optional) - T4: New averaged squared gradients
__group_712__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_713__new_weights (optional) - T2: New weights
__group_713__new_gradients (optional) - T3: New gradients
__group_713__new_moment_1 (optional) - T4: New averaged gradients
__group_713__new_moment_2 (optional) - T4: New averaged squared gradients
__group_713__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_714__new_weights (optional) - T2: New weights
__group_714__new_gradients (optional) - T3: New gradients
__group_714__new_moment_1 (optional) - T4: New averaged gradients
__group_714__new_moment_2 (optional) - T4: New averaged squared gradients
__group_714__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_715__new_weights (optional) - T2: New weights
__group_715__new_gradients (optional) - T3: New gradients
__group_715__new_moment_1 (optional) - T4: New averaged gradients
__group_715__new_moment_2 (optional) - T4: New averaged squared gradients
__group_715__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_716__new_weights (optional) - T2: New weights
__group_716__new_gradients (optional) - T3: New gradients
__group_716__new_moment_1 (optional) - T4: New averaged gradients
__group_716__new_moment_2 (optional) - T4: New averaged squared gradients
__group_716__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_717__new_weights (optional) - T2: New weights
__group_717__new_gradients (optional) - T3: New gradients
__group_717__new_moment_1 (optional) - T4: New averaged gradients
__group_717__new_moment_2 (optional) - T4: New averaged squared gradients
__group_717__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_718__new_weights (optional) - T2: New weights
__group_718__new_gradients (optional) - T3: New gradients
__group_718__new_moment_1 (optional) - T4: New averaged gradients
__group_718__new_moment_2 (optional) - T4: New averaged squared gradients
__group_718__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_719__new_weights (optional) - T2: New weights
__group_719__new_gradients (optional) - T3: New gradients
__group_719__new_moment_1 (optional) - T4: New averaged gradients
__group_719__new_moment_2 (optional) - T4: New averaged squared gradients
__group_719__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_720__new_weights (optional) - T2: New weights
__group_720__new_gradients (optional) - T3: New gradients
__group_720__new_moment_1 (optional) - T4: New averaged gradients
__group_720__new_moment_2 (optional) - T4: New averaged squared gradients
__group_720__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_721__new_weights (optional) - T2: New weights
__group_721__new_gradients (optional) - T3: New gradients
__group_721__new_moment_1 (optional) - T4: New averaged gradients
__group_721__new_moment_2 (optional) - T4: New averaged squared gradients
__group_721__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_722__new_weights (optional) - T2: New weights
__group_722__new_gradients (optional) - T3: New gradients
__group_722__new_moment_1 (optional) - T4: New averaged gradients
__group_722__new_moment_2 (optional) - T4: New averaged squared gradients
__group_722__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_723__new_weights (optional) - T2: New weights
__group_723__new_gradients (optional) - T3: New gradients
__group_723__new_moment_1 (optional) - T4: New averaged gradients
__group_723__new_moment_2 (optional) - T4: New averaged squared gradients
__group_723__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_724__new_weights (optional) - T2: New weights
__group_724__new_gradients (optional) - T3: New gradients
__group_724__new_moment_1 (optional) - T4: New averaged gradients
__group_724__new_moment_2 (optional) - T4: New averaged squared gradients
__group_724__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_725__new_weights (optional) - T2: New weights
__group_725__new_gradients (optional) - T3: New gradients
__group_725__new_moment_1 (optional) - T4: New averaged gradients
__group_725__new_moment_2 (optional) - T4: New averaged squared gradients
__group_725__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_726__new_weights (optional) - T2: New weights
__group_726__new_gradients (optional) - T3: New gradients
__group_726__new_moment_1 (optional) - T4: New averaged gradients
__group_726__new_moment_2 (optional) - T4: New averaged squared gradients
__group_726__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_727__new_weights (optional) - T2: New weights
__group_727__new_gradients (optional) - T3: New gradients
__group_727__new_moment_1 (optional) - T4: New averaged gradients
__group_727__new_moment_2 (optional) - T4: New averaged squared gradients
__group_727__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_728__new_weights (optional) - T2: New weights
__group_728__new_gradients (optional) - T3: New gradients
__group_728__new_moment_1 (optional) - T4: New averaged gradients
__group_728__new_moment_2 (optional) - T4: New averaged squared gradients
__group_728__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_729__new_weights (optional) - T2: New weights
__group_729__new_gradients (optional) - T3: New gradients
__group_729__new_moment_1 (optional) - T4: New averaged gradients
__group_729__new_moment_2 (optional) - T4: New averaged squared gradients
__group_729__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_730__new_weights (optional) - T2: New weights
__group_730__new_gradients (optional) - T3: New gradients
__group_730__new_moment_1 (optional) - T4: New averaged gradients
__group_730__new_moment_2 (optional) - T4: New averaged squared gradients
__group_730__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_731__new_weights (optional) - T2: New weights
__group_731__new_gradients (optional) - T3: New gradients
__group_731__new_moment_1 (optional) - T4: New averaged gradients
__group_731__new_moment_2 (optional) - T4: New averaged squared gradients
__group_731__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_732__new_weights (optional) - T2: New weights
__group_732__new_gradients (optional) - T3: New gradients
__group_732__new_moment_1 (optional) - T4: New averaged gradients
__group_732__new_moment_2 (optional) - T4: New averaged squared gradients
__group_732__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_733__new_weights (optional) - T2: New weights
__group_733__new_gradients (optional) - T3: New gradients
__group_733__new_moment_1 (optional) - T4: New averaged gradients
__group_733__new_moment_2 (optional) - T4: New averaged squared gradients
__group_733__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_734__new_weights (optional) - T2: New weights
__group_734__new_gradients (optional) - T3: New gradients
__group_734__new_moment_1 (optional) - T4: New averaged gradients
__group_734__new_moment_2 (optional) - T4: New averaged squared gradients
__group_734__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_735__new_weights (optional) - T2: New weights
__group_735__new_gradients (optional) - T3: New gradients
__group_735__new_moment_1 (optional) - T4: New averaged gradients
__group_735__new_moment_2 (optional) - T4: New averaged squared gradients
__group_735__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_736__new_weights (optional) - T2: New weights
__group_736__new_gradients (optional) - T3: New gradients
__group_736__new_moment_1 (optional) - T4: New averaged gradients
__group_736__new_moment_2 (optional) - T4: New averaged squared gradients
__group_736__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_737__new_weights (optional) - T2: New weights
__group_737__new_gradients (optional) - T3: New gradients
__group_737__new_moment_1 (optional) - T4: New averaged gradients
__group_737__new_moment_2 (optional) - T4: New averaged squared gradients
__group_737__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_738__new_weights (optional) - T2: New weights
__group_738__new_gradients (optional) - T3: New gradients
__group_738__new_moment_1 (optional) - T4: New averaged gradients
__group_738__new_moment_2 (optional) - T4: New averaged squared gradients
__group_738__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_739__new_weights (optional) - T2: New weights
__group_739__new_gradients (optional) - T3: New gradients
__group_739__new_moment_1 (optional) - T4: New averaged gradients
__group_739__new_moment_2 (optional) - T4: New averaged squared gradients
__group_739__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_740__new_weights (optional) - T2: New weights
__group_740__new_gradients (optional) - T3: New gradients
__group_740__new_moment_1 (optional) - T4: New averaged gradients
__group_740__new_moment_2 (optional) - T4: New averaged squared gradients
__group_740__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_741__new_weights (optional) - T2: New weights
__group_741__new_gradients (optional) - T3: New gradients
__group_741__new_moment_1 (optional) - T4: New averaged gradients
__group_741__new_moment_2 (optional) - T4: New averaged squared gradients
__group_741__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_742__new_weights (optional) - T2: New weights
__group_742__new_gradients (optional) - T3: New gradients
__group_742__new_moment_1 (optional) - T4: New averaged gradients
__group_742__new_moment_2 (optional) - T4: New averaged squared gradients
__group_742__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_743__new_weights (optional) - T2: New weights
__group_743__new_gradients (optional) - T3: New gradients
__group_743__new_moment_1 (optional) - T4: New averaged gradients
__group_743__new_moment_2 (optional) - T4: New averaged squared gradients
__group_743__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_744__new_weights (optional) - T2: New weights
__group_744__new_gradients (optional) - T3: New gradients
__group_744__new_moment_1 (optional) - T4: New averaged gradients
__group_744__new_moment_2 (optional) - T4: New averaged squared gradients
__group_744__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_745__new_weights (optional) - T2: New weights
__group_745__new_gradients (optional) - T3: New gradients
__group_745__new_moment_1 (optional) - T4: New averaged gradients
__group_745__new_moment_2 (optional) - T4: New averaged squared gradients
__group_745__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_746__new_weights (optional) - T2: New weights
__group_746__new_gradients (optional) - T3: New gradients
__group_746__new_moment_1 (optional) - T4: New averaged gradients
__group_746__new_moment_2 (optional) - T4: New averaged squared gradients
__group_746__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_747__new_weights (optional) - T2: New weights
__group_747__new_gradients (optional) - T3: New gradients
__group_747__new_moment_1 (optional) - T4: New averaged gradients
__group_747__new_moment_2 (optional) - T4: New averaged squared gradients
__group_747__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_748__new_weights (optional) - T2: New weights
__group_748__new_gradients (optional) - T3: New gradients
__group_748__new_moment_1 (optional) - T4: New averaged gradients
__group_748__new_moment_2 (optional) - T4: New averaged squared gradients
__group_748__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_749__new_weights (optional) - T2: New weights
__group_749__new_gradients (optional) - T3: New gradients
__group_749__new_moment_1 (optional) - T4: New averaged gradients
__group_749__new_moment_2 (optional) - T4: New averaged squared gradients
__group_749__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_750__new_weights (optional) - T2: New weights
__group_750__new_gradients (optional) - T3: New gradients
__group_750__new_moment_1 (optional) - T4: New averaged gradients
__group_750__new_moment_2 (optional) - T4: New averaged squared gradients
__group_750__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_751__new_weights (optional) - T2: New weights
__group_751__new_gradients (optional) - T3: New gradients
__group_751__new_moment_1 (optional) - T4: New averaged gradients
__group_751__new_moment_2 (optional) - T4: New averaged squared gradients
__group_751__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_752__new_weights (optional) - T2: New weights
__group_752__new_gradients (optional) - T3: New gradients
__group_752__new_moment_1 (optional) - T4: New averaged gradients
__group_752__new_moment_2 (optional) - T4: New averaged squared gradients
__group_752__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_753__new_weights (optional) - T2: New weights
__group_753__new_gradients (optional) - T3: New gradients
__group_753__new_moment_1 (optional) - T4: New averaged gradients
__group_753__new_moment_2 (optional) - T4: New averaged squared gradients
__group_753__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_754__new_weights (optional) - T2: New weights
__group_754__new_gradients (optional) - T3: New gradients
__group_754__new_moment_1 (optional) - T4: New averaged gradients
__group_754__new_moment_2 (optional) - T4: New averaged squared gradients
__group_754__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_755__new_weights (optional) - T2: New weights
__group_755__new_gradients (optional) - T3: New gradients
__group_755__new_moment_1 (optional) - T4: New averaged gradients
__group_755__new_moment_2 (optional) - T4: New averaged squared gradients
__group_755__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_756__new_weights (optional) - T2: New weights
__group_756__new_gradients (optional) - T3: New gradients
__group_756__new_moment_1 (optional) - T4: New averaged gradients
__group_756__new_moment_2 (optional) - T4: New averaged squared gradients
__group_756__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_757__new_weights (optional) - T2: New weights
__group_757__new_gradients (optional) - T3: New gradients
__group_757__new_moment_1 (optional) - T4: New averaged gradients
__group_757__new_moment_2 (optional) - T4: New averaged squared gradients
__group_757__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_758__new_weights (optional) - T2: New weights
__group_758__new_gradients (optional) - T3: New gradients
__group_758__new_moment_1 (optional) - T4: New averaged gradients
__group_758__new_moment_2 (optional) - T4: New averaged squared gradients
__group_758__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_759__new_weights (optional) - T2: New weights
__group_759__new_gradients (optional) - T3: New gradients
__group_759__new_moment_1 (optional) - T4: New averaged gradients
__group_759__new_moment_2 (optional) - T4: New averaged squared gradients
__group_759__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_760__new_weights (optional) - T2: New weights
__group_760__new_gradients (optional) - T3: New gradients
__group_760__new_moment_1 (optional) - T4: New averaged gradients
__group_760__new_moment_2 (optional) - T4: New averaged squared gradients
__group_760__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_761__new_weights (optional) - T2: New weights
__group_761__new_gradients (optional) - T3: New gradients
__group_761__new_moment_1 (optional) - T4: New averaged gradients
__group_761__new_moment_2 (optional) - T4: New averaged squared gradients
__group_761__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_762__new_weights (optional) - T2: New weights
__group_762__new_gradients (optional) - T3: New gradients
__group_762__new_moment_1 (optional) - T4: New averaged gradients
__group_762__new_moment_2 (optional) - T4: New averaged squared gradients
__group_762__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_763__new_weights (optional) - T2: New weights
__group_763__new_gradients (optional) - T3: New gradients
__group_763__new_moment_1 (optional) - T4: New averaged gradients
__group_763__new_moment_2 (optional) - T4: New averaged squared gradients
__group_763__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_764__new_weights (optional) - T2: New weights
__group_764__new_gradients (optional) - T3: New gradients
__group_764__new_moment_1 (optional) - T4: New averaged gradients
__group_764__new_moment_2 (optional) - T4: New averaged squared gradients
__group_764__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_765__new_weights (optional) - T2: New weights
__group_765__new_gradients (optional) - T3: New gradients
__group_765__new_moment_1 (optional) - T4: New averaged gradients
__group_765__new_moment_2 (optional) - T4: New averaged squared gradients
__group_765__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_766__new_weights (optional) - T2: New weights
__group_766__new_gradients (optional) - T3: New gradients
__group_766__new_moment_1 (optional) - T4: New averaged gradients
__group_766__new_moment_2 (optional) - T4: New averaged squared gradients
__group_766__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_767__new_weights (optional) - T2: New weights
__group_767__new_gradients (optional) - T3: New gradients
__group_767__new_moment_1 (optional) - T4: New averaged gradients
__group_767__new_moment_2 (optional) - T4: New averaged squared gradients
__group_767__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_768__new_weights (optional) - T2: New weights
__group_768__new_gradients (optional) - T3: New gradients
__group_768__new_moment_1 (optional) - T4: New averaged gradients
__group_768__new_moment_2 (optional) - T4: New averaged squared gradients
__group_768__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_769__new_weights (optional) - T2: New weights
__group_769__new_gradients (optional) - T3: New gradients
__group_769__new_moment_1 (optional) - T4: New averaged gradients
__group_769__new_moment_2 (optional) - T4: New averaged squared gradients
__group_769__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_770__new_weights (optional) - T2: New weights
__group_770__new_gradients (optional) - T3: New gradients
__group_770__new_moment_1 (optional) - T4: New averaged gradients
__group_770__new_moment_2 (optional) - T4: New averaged squared gradients
__group_770__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_771__new_weights (optional) - T2: New weights
__group_771__new_gradients (optional) - T3: New gradients
__group_771__new_moment_1 (optional) - T4: New averaged gradients
__group_771__new_moment_2 (optional) - T4: New averaged squared gradients
__group_771__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_772__new_weights (optional) - T2: New weights
__group_772__new_gradients (optional) - T3: New gradients
__group_772__new_moment_1 (optional) - T4: New averaged gradients
__group_772__new_moment_2 (optional) - T4: New averaged squared gradients
__group_772__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_773__new_weights (optional) - T2: New weights
__group_773__new_gradients (optional) - T3: New gradients
__group_773__new_moment_1 (optional) - T4: New averaged gradients
__group_773__new_moment_2 (optional) - T4: New averaged squared gradients
__group_773__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_774__new_weights (optional) - T2: New weights
__group_774__new_gradients (optional) - T3: New gradients
__group_774__new_moment_1 (optional) - T4: New averaged gradients
__group_774__new_moment_2 (optional) - T4: New averaged squared gradients
__group_774__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_775__new_weights (optional) - T2: New weights
__group_775__new_gradients (optional) - T3: New gradients
__group_775__new_moment_1 (optional) - T4: New averaged gradients
__group_775__new_moment_2 (optional) - T4: New averaged squared gradients
__group_775__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_776__new_weights (optional) - T2: New weights
__group_776__new_gradients (optional) - T3: New gradients
__group_776__new_moment_1 (optional) - T4: New averaged gradients
__group_776__new_moment_2 (optional) - T4: New averaged squared gradients
__group_776__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_777__new_weights (optional) - T2: New weights
__group_777__new_gradients (optional) - T3: New gradients
__group_777__new_moment_1 (optional) - T4: New averaged gradients
__group_777__new_moment_2 (optional) - T4: New averaged squared gradients
__group_777__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_778__new_weights (optional) - T2: New weights
__group_778__new_gradients (optional) - T3: New gradients
__group_778__new_moment_1 (optional) - T4: New averaged gradients
__group_778__new_moment_2 (optional) - T4: New averaged squared gradients
__group_778__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_779__new_weights (optional) - T2: New weights
__group_779__new_gradients (optional) - T3: New gradients
__group_779__new_moment_1 (optional) - T4: New averaged gradients
__group_779__new_moment_2 (optional) - T4: New averaged squared gradients
__group_779__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_780__new_weights (optional) - T2: New weights
__group_780__new_gradients (optional) - T3: New gradients
__group_780__new_moment_1 (optional) - T4: New averaged gradients
__group_780__new_moment_2 (optional) - T4: New averaged squared gradients
__group_780__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_781__new_weights (optional) - T2: New weights
__group_781__new_gradients (optional) - T3: New gradients
__group_781__new_moment_1 (optional) - T4: New averaged gradients
__group_781__new_moment_2 (optional) - T4: New averaged squared gradients
__group_781__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_782__new_weights (optional) - T2: New weights
__group_782__new_gradients (optional) - T3: New gradients
__group_782__new_moment_1 (optional) - T4: New averaged gradients
__group_782__new_moment_2 (optional) - T4: New averaged squared gradients
__group_782__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_783__new_weights (optional) - T2: New weights
__group_783__new_gradients (optional) - T3: New gradients
__group_783__new_moment_1 (optional) - T4: New averaged gradients
__group_783__new_moment_2 (optional) - T4: New averaged squared gradients
__group_783__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_784__new_weights (optional) - T2: New weights
__group_784__new_gradients (optional) - T3: New gradients
__group_784__new_moment_1 (optional) - T4: New averaged gradients
__group_784__new_moment_2 (optional) - T4: New averaged squared gradients
__group_784__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_785__new_weights (optional) - T2: New weights
__group_785__new_gradients (optional) - T3: New gradients
__group_785__new_moment_1 (optional) - T4: New averaged gradients
__group_785__new_moment_2 (optional) - T4: New averaged squared gradients
__group_785__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_786__new_weights (optional) - T2: New weights
__group_786__new_gradients (optional) - T3: New gradients
__group_786__new_moment_1 (optional) - T4: New averaged gradients
__group_786__new_moment_2 (optional) - T4: New averaged squared gradients
__group_786__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_787__new_weights (optional) - T2: New weights
__group_787__new_gradients (optional) - T3: New gradients
__group_787__new_moment_1 (optional) - T4: New averaged gradients
__group_787__new_moment_2 (optional) - T4: New averaged squared gradients
__group_787__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_788__new_weights (optional) - T2: New weights
__group_788__new_gradients (optional) - T3: New gradients
__group_788__new_moment_1 (optional) - T4: New averaged gradients
__group_788__new_moment_2 (optional) - T4: New averaged squared gradients
__group_788__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_789__new_weights (optional) - T2: New weights
__group_789__new_gradients (optional) - T3: New gradients
__group_789__new_moment_1 (optional) - T4: New averaged gradients
__group_789__new_moment_2 (optional) - T4: New averaged squared gradients
__group_789__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_790__new_weights (optional) - T2: New weights
__group_790__new_gradients (optional) - T3: New gradients
__group_790__new_moment_1 (optional) - T4: New averaged gradients
__group_790__new_moment_2 (optional) - T4: New averaged squared gradients
__group_790__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_791__new_weights (optional) - T2: New weights
__group_791__new_gradients (optional) - T3: New gradients
__group_791__new_moment_1 (optional) - T4: New averaged gradients
__group_791__new_moment_2 (optional) - T4: New averaged squared gradients
__group_791__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_792__new_weights (optional) - T2: New weights
__group_792__new_gradients (optional) - T3: New gradients
__group_792__new_moment_1 (optional) - T4: New averaged gradients
__group_792__new_moment_2 (optional) - T4: New averaged squared gradients
__group_792__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_793__new_weights (optional) - T2: New weights
__group_793__new_gradients (optional) - T3: New gradients
__group_793__new_moment_1 (optional) - T4: New averaged gradients
__group_793__new_moment_2 (optional) - T4: New averaged squared gradients
__group_793__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_794__new_weights (optional) - T2: New weights
__group_794__new_gradients (optional) - T3: New gradients
__group_794__new_moment_1 (optional) - T4: New averaged gradients
__group_794__new_moment_2 (optional) - T4: New averaged squared gradients
__group_794__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_795__new_weights (optional) - T2: New weights
__group_795__new_gradients (optional) - T3: New gradients
__group_795__new_moment_1 (optional) - T4: New averaged gradients
__group_795__new_moment_2 (optional) - T4: New averaged squared gradients
__group_795__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_796__new_weights (optional) - T2: New weights
__group_796__new_gradients (optional) - T3: New gradients
__group_796__new_moment_1 (optional) - T4: New averaged gradients
__group_796__new_moment_2 (optional) - T4: New averaged squared gradients
__group_796__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_797__new_weights (optional) - T2: New weights
__group_797__new_gradients (optional) - T3: New gradients
__group_797__new_moment_1 (optional) - T4: New averaged gradients
__group_797__new_moment_2 (optional) - T4: New averaged squared gradients
__group_797__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_798__new_weights (optional) - T2: New weights
__group_798__new_gradients (optional) - T3: New gradients
__group_798__new_moment_1 (optional) - T4: New averaged gradients
__group_798__new_moment_2 (optional) - T4: New averaged squared gradients
__group_798__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_799__new_weights (optional) - T2: New weights
__group_799__new_gradients (optional) - T3: New gradients
__group_799__new_moment_1 (optional) - T4: New averaged gradients
__group_799__new_moment_2 (optional) - T4: New averaged squared gradients
__group_799__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_800__new_weights (optional) - T2: New weights
__group_800__new_gradients (optional) - T3: New gradients
__group_800__new_moment_1 (optional) - T4: New averaged gradients
__group_800__new_moment_2 (optional) - T4: New averaged squared gradients
__group_800__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_801__new_weights (optional) - T2: New weights
__group_801__new_gradients (optional) - T3: New gradients
__group_801__new_moment_1 (optional) - T4: New averaged gradients
__group_801__new_moment_2 (optional) - T4: New averaged squared gradients
__group_801__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_802__new_weights (optional) - T2: New weights
__group_802__new_gradients (optional) - T3: New gradients
__group_802__new_moment_1 (optional) - T4: New averaged gradients
__group_802__new_moment_2 (optional) - T4: New averaged squared gradients
__group_802__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_803__new_weights (optional) - T2: New weights
__group_803__new_gradients (optional) - T3: New gradients
__group_803__new_moment_1 (optional) - T4: New averaged gradients
__group_803__new_moment_2 (optional) - T4: New averaged squared gradients
__group_803__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_804__new_weights (optional) - T2: New weights
__group_804__new_gradients (optional) - T3: New gradients
__group_804__new_moment_1 (optional) - T4: New averaged gradients
__group_804__new_moment_2 (optional) - T4: New averaged squared gradients
__group_804__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_805__new_weights (optional) - T2: New weights
__group_805__new_gradients (optional) - T3: New gradients
__group_805__new_moment_1 (optional) - T4: New averaged gradients
__group_805__new_moment_2 (optional) - T4: New averaged squared gradients
__group_805__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_806__new_weights (optional) - T2: New weights
__group_806__new_gradients (optional) - T3: New gradients
__group_806__new_moment_1 (optional) - T4: New averaged gradients
__group_806__new_moment_2 (optional) - T4: New averaged squared gradients
__group_806__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_807__new_weights (optional) - T2: New weights
__group_807__new_gradients (optional) - T3: New gradients
__group_807__new_moment_1 (optional) - T4: New averaged gradients
__group_807__new_moment_2 (optional) - T4: New averaged squared gradients
__group_807__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_808__new_weights (optional) - T2: New weights
__group_808__new_gradients (optional) - T3: New gradients
__group_808__new_moment_1 (optional) - T4: New averaged gradients
__group_808__new_moment_2 (optional) - T4: New averaged squared gradients
__group_808__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_809__new_weights (optional) - T2: New weights
__group_809__new_gradients (optional) - T3: New gradients
__group_809__new_moment_1 (optional) - T4: New averaged gradients
__group_809__new_moment_2 (optional) - T4: New averaged squared gradients
__group_809__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_810__new_weights (optional) - T2: New weights
__group_810__new_gradients (optional) - T3: New gradients
__group_810__new_moment_1 (optional) - T4: New averaged gradients
__group_810__new_moment_2 (optional) - T4: New averaged squared gradients
__group_810__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_811__new_weights (optional) - T2: New weights
__group_811__new_gradients (optional) - T3: New gradients
__group_811__new_moment_1 (optional) - T4: New averaged gradients
__group_811__new_moment_2 (optional) - T4: New averaged squared gradients
__group_811__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_812__new_weights (optional) - T2: New weights
__group_812__new_gradients (optional) - T3: New gradients
__group_812__new_moment_1 (optional) - T4: New averaged gradients
__group_812__new_moment_2 (optional) - T4: New averaged squared gradients
__group_812__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_813__new_weights (optional) - T2: New weights
__group_813__new_gradients (optional) - T3: New gradients
__group_813__new_moment_1 (optional) - T4: New averaged gradients
__group_813__new_moment_2 (optional) - T4: New averaged squared gradients
__group_813__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_814__new_weights (optional) - T2: New weights
__group_814__new_gradients (optional) - T3: New gradients
__group_814__new_moment_1 (optional) - T4: New averaged gradients
__group_814__new_moment_2 (optional) - T4: New averaged squared gradients
__group_814__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_815__new_weights (optional) - T2: New weights
__group_815__new_gradients (optional) - T3: New gradients
__group_815__new_moment_1 (optional) - T4: New averaged gradients
__group_815__new_moment_2 (optional) - T4: New averaged squared gradients
__group_815__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_816__new_weights (optional) - T2: New weights
__group_816__new_gradients (optional) - T3: New gradients
__group_816__new_moment_1 (optional) - T4: New averaged gradients
__group_816__new_moment_2 (optional) - T4: New averaged squared gradients
__group_816__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_817__new_weights (optional) - T2: New weights
__group_817__new_gradients (optional) - T3: New gradients
__group_817__new_moment_1 (optional) - T4: New averaged gradients
__group_817__new_moment_2 (optional) - T4: New averaged squared gradients
__group_817__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_818__new_weights (optional) - T2: New weights
__group_818__new_gradients (optional) - T3: New gradients
__group_818__new_moment_1 (optional) - T4: New averaged gradients
__group_818__new_moment_2 (optional) - T4: New averaged squared gradients
__group_818__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_819__new_weights (optional) - T2: New weights
__group_819__new_gradients (optional) - T3: New gradients
__group_819__new_moment_1 (optional) - T4: New averaged gradients
__group_819__new_moment_2 (optional) - T4: New averaged squared gradients
__group_819__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_820__new_weights (optional) - T2: New weights
__group_820__new_gradients (optional) - T3: New gradients
__group_820__new_moment_1 (optional) - T4: New averaged gradients
__group_820__new_moment_2 (optional) - T4: New averaged squared gradients
__group_820__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_821__new_weights (optional) - T2: New weights
__group_821__new_gradients (optional) - T3: New gradients
__group_821__new_moment_1 (optional) - T4: New averaged gradients
__group_821__new_moment_2 (optional) - T4: New averaged squared gradients
__group_821__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_822__new_weights (optional) - T2: New weights
__group_822__new_gradients (optional) - T3: New gradients
__group_822__new_moment_1 (optional) - T4: New averaged gradients
__group_822__new_moment_2 (optional) - T4: New averaged squared gradients
__group_822__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_823__new_weights (optional) - T2: New weights
__group_823__new_gradients (optional) - T3: New gradients
__group_823__new_moment_1 (optional) - T4: New averaged gradients
__group_823__new_moment_2 (optional) - T4: New averaged squared gradients
__group_823__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_824__new_weights (optional) - T2: New weights
__group_824__new_gradients (optional) - T3: New gradients
__group_824__new_moment_1 (optional) - T4: New averaged gradients
__group_824__new_moment_2 (optional) - T4: New averaged squared gradients
__group_824__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_825__new_weights (optional) - T2: New weights
__group_825__new_gradients (optional) - T3: New gradients
__group_825__new_moment_1 (optional) - T4: New averaged gradients
__group_825__new_moment_2 (optional) - T4: New averaged squared gradients
__group_825__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_826__new_weights (optional) - T2: New weights
__group_826__new_gradients (optional) - T3: New gradients
__group_826__new_moment_1 (optional) - T4: New averaged gradients
__group_826__new_moment_2 (optional) - T4: New averaged squared gradients
__group_826__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_827__new_weights (optional) - T2: New weights
__group_827__new_gradients (optional) - T3: New gradients
__group_827__new_moment_1 (optional) - T4: New averaged gradients
__group_827__new_moment_2 (optional) - T4: New averaged squared gradients
__group_827__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_828__new_weights (optional) - T2: New weights
__group_828__new_gradients (optional) - T3: New gradients
__group_828__new_moment_1 (optional) - T4: New averaged gradients
__group_828__new_moment_2 (optional) - T4: New averaged squared gradients
__group_828__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_829__new_weights (optional) - T2: New weights
__group_829__new_gradients (optional) - T3: New gradients
__group_829__new_moment_1 (optional) - T4: New averaged gradients
__group_829__new_moment_2 (optional) - T4: New averaged squared gradients
__group_829__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_830__new_weights (optional) - T2: New weights
__group_830__new_gradients (optional) - T3: New gradients
__group_830__new_moment_1 (optional) - T4: New averaged gradients
__group_830__new_moment_2 (optional) - T4: New averaged squared gradients
__group_830__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_831__new_weights (optional) - T2: New weights
__group_831__new_gradients (optional) - T3: New gradients
__group_831__new_moment_1 (optional) - T4: New averaged gradients
__group_831__new_moment_2 (optional) - T4: New averaged squared gradients
__group_831__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_832__new_weights (optional) - T2: New weights
__group_832__new_gradients (optional) - T3: New gradients
__group_832__new_moment_1 (optional) - T4: New averaged gradients
__group_832__new_moment_2 (optional) - T4: New averaged squared gradients
__group_832__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_833__new_weights (optional) - T2: New weights
__group_833__new_gradients (optional) - T3: New gradients
__group_833__new_moment_1 (optional) - T4: New averaged gradients
__group_833__new_moment_2 (optional) - T4: New averaged squared gradients
__group_833__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_834__new_weights (optional) - T2: New weights
__group_834__new_gradients (optional) - T3: New gradients
__group_834__new_moment_1 (optional) - T4: New averaged gradients
__group_834__new_moment_2 (optional) - T4: New averaged squared gradients
__group_834__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_835__new_weights (optional) - T2: New weights
__group_835__new_gradients (optional) - T3: New gradients
__group_835__new_moment_1 (optional) - T4: New averaged gradients
__group_835__new_moment_2 (optional) - T4: New averaged squared gradients
__group_835__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_836__new_weights (optional) - T2: New weights
__group_836__new_gradients (optional) - T3: New gradients
__group_836__new_moment_1 (optional) - T4: New averaged gradients
__group_836__new_moment_2 (optional) - T4: New averaged squared gradients
__group_836__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_837__new_weights (optional) - T2: New weights
__group_837__new_gradients (optional) - T3: New gradients
__group_837__new_moment_1 (optional) - T4: New averaged gradients
__group_837__new_moment_2 (optional) - T4: New averaged squared gradients
__group_837__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_838__new_weights (optional) - T2: New weights
__group_838__new_gradients (optional) - T3: New gradients
__group_838__new_moment_1 (optional) - T4: New averaged gradients
__group_838__new_moment_2 (optional) - T4: New averaged squared gradients
__group_838__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_839__new_weights (optional) - T2: New weights
__group_839__new_gradients (optional) - T3: New gradients
__group_839__new_moment_1 (optional) - T4: New averaged gradients
__group_839__new_moment_2 (optional) - T4: New averaged squared gradients
__group_839__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_840__new_weights (optional) - T2: New weights
__group_840__new_gradients (optional) - T3: New gradients
__group_840__new_moment_1 (optional) - T4: New averaged gradients
__group_840__new_moment_2 (optional) - T4: New averaged squared gradients
__group_840__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_841__new_weights (optional) - T2: New weights
__group_841__new_gradients (optional) - T3: New gradients
__group_841__new_moment_1 (optional) - T4: New averaged gradients
__group_841__new_moment_2 (optional) - T4: New averaged squared gradients
__group_841__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_842__new_weights (optional) - T2: New weights
__group_842__new_gradients (optional) - T3: New gradients
__group_842__new_moment_1 (optional) - T4: New averaged gradients
__group_842__new_moment_2 (optional) - T4: New averaged squared gradients
__group_842__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_843__new_weights (optional) - T2: New weights
__group_843__new_gradients (optional) - T3: New gradients
__group_843__new_moment_1 (optional) - T4: New averaged gradients
__group_843__new_moment_2 (optional) - T4: New averaged squared gradients
__group_843__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_844__new_weights (optional) - T2: New weights
__group_844__new_gradients (optional) - T3: New gradients
__group_844__new_moment_1 (optional) - T4: New averaged gradients
__group_844__new_moment_2 (optional) - T4: New averaged squared gradients
__group_844__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_845__new_weights (optional) - T2: New weights
__group_845__new_gradients (optional) - T3: New gradients
__group_845__new_moment_1 (optional) - T4: New averaged gradients
__group_845__new_moment_2 (optional) - T4: New averaged squared gradients
__group_845__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_846__new_weights (optional) - T2: New weights
__group_846__new_gradients (optional) - T3: New gradients
__group_846__new_moment_1 (optional) - T4: New averaged gradients
__group_846__new_moment_2 (optional) - T4: New averaged squared gradients
__group_846__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_847__new_weights (optional) - T2: New weights
__group_847__new_gradients (optional) - T3: New gradients
__group_847__new_moment_1 (optional) - T4: New averaged gradients
__group_847__new_moment_2 (optional) - T4: New averaged squared gradients
__group_847__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_848__new_weights (optional) - T2: New weights
__group_848__new_gradients (optional) - T3: New gradients
__group_848__new_moment_1 (optional) - T4: New averaged gradients
__group_848__new_moment_2 (optional) - T4: New averaged squared gradients
__group_848__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_849__new_weights (optional) - T2: New weights
__group_849__new_gradients (optional) - T3: New gradients
__group_849__new_moment_1 (optional) - T4: New averaged gradients
__group_849__new_moment_2 (optional) - T4: New averaged squared gradients
__group_849__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_850__new_weights (optional) - T2: New weights
__group_850__new_gradients (optional) - T3: New gradients
__group_850__new_moment_1 (optional) - T4: New averaged gradients
__group_850__new_moment_2 (optional) - T4: New averaged squared gradients
__group_850__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_851__new_weights (optional) - T2: New weights
__group_851__new_gradients (optional) - T3: New gradients
__group_851__new_moment_1 (optional) - T4: New averaged gradients
__group_851__new_moment_2 (optional) - T4: New averaged squared gradients
__group_851__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_852__new_weights (optional) - T2: New weights
__group_852__new_gradients (optional) - T3: New gradients
__group_852__new_moment_1 (optional) - T4: New averaged gradients
__group_852__new_moment_2 (optional) - T4: New averaged squared gradients
__group_852__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_853__new_weights (optional) - T2: New weights
__group_853__new_gradients (optional) - T3: New gradients
__group_853__new_moment_1 (optional) - T4: New averaged gradients
__group_853__new_moment_2 (optional) - T4: New averaged squared gradients
__group_853__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_854__new_weights (optional) - T2: New weights
__group_854__new_gradients (optional) - T3: New gradients
__group_854__new_moment_1 (optional) - T4: New averaged gradients
__group_854__new_moment_2 (optional) - T4: New averaged squared gradients
__group_854__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_855__new_weights (optional) - T2: New weights
__group_855__new_gradients (optional) - T3: New gradients
__group_855__new_moment_1 (optional) - T4: New averaged gradients
__group_855__new_moment_2 (optional) - T4: New averaged squared gradients
__group_855__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_856__new_weights (optional) - T2: New weights
__group_856__new_gradients (optional) - T3: New gradients
__group_856__new_moment_1 (optional) - T4: New averaged gradients
__group_856__new_moment_2 (optional) - T4: New averaged squared gradients
__group_856__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_857__new_weights (optional) - T2: New weights
__group_857__new_gradients (optional) - T3: New gradients
__group_857__new_moment_1 (optional) - T4: New averaged gradients
__group_857__new_moment_2 (optional) - T4: New averaged squared gradients
__group_857__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_858__new_weights (optional) - T2: New weights
__group_858__new_gradients (optional) - T3: New gradients
__group_858__new_moment_1 (optional) - T4: New averaged gradients
__group_858__new_moment_2 (optional) - T4: New averaged squared gradients
__group_858__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_859__new_weights (optional) - T2: New weights
__group_859__new_gradients (optional) - T3: New gradients
__group_859__new_moment_1 (optional) - T4: New averaged gradients
__group_859__new_moment_2 (optional) - T4: New averaged squared gradients
__group_859__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_860__new_weights (optional) - T2: New weights
__group_860__new_gradients (optional) - T3: New gradients
__group_860__new_moment_1 (optional) - T4: New averaged gradients
__group_860__new_moment_2 (optional) - T4: New averaged squared gradients
__group_860__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_861__new_weights (optional) - T2: New weights
__group_861__new_gradients (optional) - T3: New gradients
__group_861__new_moment_1 (optional) - T4: New averaged gradients
__group_861__new_moment_2 (optional) - T4: New averaged squared gradients
__group_861__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_862__new_weights (optional) - T2: New weights
__group_862__new_gradients (optional) - T3: New gradients
__group_862__new_moment_1 (optional) - T4: New averaged gradients
__group_862__new_moment_2 (optional) - T4: New averaged squared gradients
__group_862__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_863__new_weights (optional) - T2: New weights
__group_863__new_gradients (optional) - T3: New gradients
__group_863__new_moment_1 (optional) - T4: New averaged gradients
__group_863__new_moment_2 (optional) - T4: New averaged squared gradients
__group_863__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_864__new_weights (optional) - T2: New weights
__group_864__new_gradients (optional) - T3: New gradients
__group_864__new_moment_1 (optional) - T4: New averaged gradients
__group_864__new_moment_2 (optional) - T4: New averaged squared gradients
__group_864__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_865__new_weights (optional) - T2: New weights
__group_865__new_gradients (optional) - T3: New gradients
__group_865__new_moment_1 (optional) - T4: New averaged gradients
__group_865__new_moment_2 (optional) - T4: New averaged squared gradients
__group_865__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_866__new_weights (optional) - T2: New weights
__group_866__new_gradients (optional) - T3: New gradients
__group_866__new_moment_1 (optional) - T4: New averaged gradients
__group_866__new_moment_2 (optional) - T4: New averaged squared gradients
__group_866__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_867__new_weights (optional) - T2: New weights
__group_867__new_gradients (optional) - T3: New gradients
__group_867__new_moment_1 (optional) - T4: New averaged gradients
__group_867__new_moment_2 (optional) - T4: New averaged squared gradients
__group_867__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_868__new_weights (optional) - T2: New weights
__group_868__new_gradients (optional) - T3: New gradients
__group_868__new_moment_1 (optional) - T4: New averaged gradients
__group_868__new_moment_2 (optional) - T4: New averaged squared gradients
__group_868__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_869__new_weights (optional) - T2: New weights
__group_869__new_gradients (optional) - T3: New gradients
__group_869__new_moment_1 (optional) - T4: New averaged gradients
__group_869__new_moment_2 (optional) - T4: New averaged squared gradients
__group_869__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_870__new_weights (optional) - T2: New weights
__group_870__new_gradients (optional) - T3: New gradients
__group_870__new_moment_1 (optional) - T4: New averaged gradients
__group_870__new_moment_2 (optional) - T4: New averaged squared gradients
__group_870__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_871__new_weights (optional) - T2: New weights
__group_871__new_gradients (optional) - T3: New gradients
__group_871__new_moment_1 (optional) - T4: New averaged gradients
__group_871__new_moment_2 (optional) - T4: New averaged squared gradients
__group_871__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_872__new_weights (optional) - T2: New weights
__group_872__new_gradients (optional) - T3: New gradients
__group_872__new_moment_1 (optional) - T4: New averaged gradients
__group_872__new_moment_2 (optional) - T4: New averaged squared gradients
__group_872__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_873__new_weights (optional) - T2: New weights
__group_873__new_gradients (optional) - T3: New gradients
__group_873__new_moment_1 (optional) - T4: New averaged gradients
__group_873__new_moment_2 (optional) - T4: New averaged squared gradients
__group_873__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_874__new_weights (optional) - T2: New weights
__group_874__new_gradients (optional) - T3: New gradients
__group_874__new_moment_1 (optional) - T4: New averaged gradients
__group_874__new_moment_2 (optional) - T4: New averaged squared gradients
__group_874__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_875__new_weights (optional) - T2: New weights
__group_875__new_gradients (optional) - T3: New gradients
__group_875__new_moment_1 (optional) - T4: New averaged gradients
__group_875__new_moment_2 (optional) - T4: New averaged squared gradients
__group_875__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_876__new_weights (optional) - T2: New weights
__group_876__new_gradients (optional) - T3: New gradients
__group_876__new_moment_1 (optional) - T4: New averaged gradients
__group_876__new_moment_2 (optional) - T4: New averaged squared gradients
__group_876__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_877__new_weights (optional) - T2: New weights
__group_877__new_gradients (optional) - T3: New gradients
__group_877__new_moment_1 (optional) - T4: New averaged gradients
__group_877__new_moment_2 (optional) - T4: New averaged squared gradients
__group_877__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_878__new_weights (optional) - T2: New weights
__group_878__new_gradients (optional) - T3: New gradients
__group_878__new_moment_1 (optional) - T4: New averaged gradients
__group_878__new_moment_2 (optional) - T4: New averaged squared gradients
__group_878__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_879__new_weights (optional) - T2: New weights
__group_879__new_gradients (optional) - T3: New gradients
__group_879__new_moment_1 (optional) - T4: New averaged gradients
__group_879__new_moment_2 (optional) - T4: New averaged squared gradients
__group_879__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_880__new_weights (optional) - T2: New weights
__group_880__new_gradients (optional) - T3: New gradients
__group_880__new_moment_1 (optional) - T4: New averaged gradients
__group_880__new_moment_2 (optional) - T4: New averaged squared gradients
__group_880__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_881__new_weights (optional) - T2: New weights
__group_881__new_gradients (optional) - T3: New gradients
__group_881__new_moment_1 (optional) - T4: New averaged gradients
__group_881__new_moment_2 (optional) - T4: New averaged squared gradients
__group_881__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_882__new_weights (optional) - T2: New weights
__group_882__new_gradients (optional) - T3: New gradients
__group_882__new_moment_1 (optional) - T4: New averaged gradients
__group_882__new_moment_2 (optional) - T4: New averaged squared gradients
__group_882__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_883__new_weights (optional) - T2: New weights
__group_883__new_gradients (optional) - T3: New gradients
__group_883__new_moment_1 (optional) - T4: New averaged gradients
__group_883__new_moment_2 (optional) - T4: New averaged squared gradients
__group_883__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_884__new_weights (optional) - T2: New weights
__group_884__new_gradients (optional) - T3: New gradients
__group_884__new_moment_1 (optional) - T4: New averaged gradients
__group_884__new_moment_2 (optional) - T4: New averaged squared gradients
__group_884__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_885__new_weights (optional) - T2: New weights
__group_885__new_gradients (optional) - T3: New gradients
__group_885__new_moment_1 (optional) - T4: New averaged gradients
__group_885__new_moment_2 (optional) - T4: New averaged squared gradients
__group_885__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_886__new_weights (optional) - T2: New weights
__group_886__new_gradients (optional) - T3: New gradients
__group_886__new_moment_1 (optional) - T4: New averaged gradients
__group_886__new_moment_2 (optional) - T4: New averaged squared gradients
__group_886__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_887__new_weights (optional) - T2: New weights
__group_887__new_gradients (optional) - T3: New gradients
__group_887__new_moment_1 (optional) - T4: New averaged gradients
__group_887__new_moment_2 (optional) - T4: New averaged squared gradients
__group_887__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_888__new_weights (optional) - T2: New weights
__group_888__new_gradients (optional) - T3: New gradients
__group_888__new_moment_1 (optional) - T4: New averaged gradients
__group_888__new_moment_2 (optional) - T4: New averaged squared gradients
__group_888__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_889__new_weights (optional) - T2: New weights
__group_889__new_gradients (optional) - T3: New gradients
__group_889__new_moment_1 (optional) - T4: New averaged gradients
__group_889__new_moment_2 (optional) - T4: New averaged squared gradients
__group_889__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_890__new_weights (optional) - T2: New weights
__group_890__new_gradients (optional) - T3: New gradients
__group_890__new_moment_1 (optional) - T4: New averaged gradients
__group_890__new_moment_2 (optional) - T4: New averaged squared gradients
__group_890__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_891__new_weights (optional) - T2: New weights
__group_891__new_gradients (optional) - T3: New gradients
__group_891__new_moment_1 (optional) - T4: New averaged gradients
__group_891__new_moment_2 (optional) - T4: New averaged squared gradients
__group_891__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_892__new_weights (optional) - T2: New weights
__group_892__new_gradients (optional) - T3: New gradients
__group_892__new_moment_1 (optional) - T4: New averaged gradients
__group_892__new_moment_2 (optional) - T4: New averaged squared gradients
__group_892__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_893__new_weights (optional) - T2: New weights
__group_893__new_gradients (optional) - T3: New gradients
__group_893__new_moment_1 (optional) - T4: New averaged gradients
__group_893__new_moment_2 (optional) - T4: New averaged squared gradients
__group_893__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_894__new_weights (optional) - T2: New weights
__group_894__new_gradients (optional) - T3: New gradients
__group_894__new_moment_1 (optional) - T4: New averaged gradients
__group_894__new_moment_2 (optional) - T4: New averaged squared gradients
__group_894__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_895__new_weights (optional) - T2: New weights
__group_895__new_gradients (optional) - T3: New gradients
__group_895__new_moment_1 (optional) - T4: New averaged gradients
__group_895__new_moment_2 (optional) - T4: New averaged squared gradients
__group_895__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_896__new_weights (optional) - T2: New weights
__group_896__new_gradients (optional) - T3: New gradients
__group_896__new_moment_1 (optional) - T4: New averaged gradients
__group_896__new_moment_2 (optional) - T4: New averaged squared gradients
__group_896__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_897__new_weights (optional) - T2: New weights
__group_897__new_gradients (optional) - T3: New gradients
__group_897__new_moment_1 (optional) - T4: New averaged gradients
__group_897__new_moment_2 (optional) - T4: New averaged squared gradients
__group_897__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_898__new_weights (optional) - T2: New weights
__group_898__new_gradients (optional) - T3: New gradients
__group_898__new_moment_1 (optional) - T4: New averaged gradients
__group_898__new_moment_2 (optional) - T4: New averaged squared gradients
__group_898__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_899__new_weights (optional) - T2: New weights
__group_899__new_gradients (optional) - T3: New gradients
__group_899__new_moment_1 (optional) - T4: New averaged gradients
__group_899__new_moment_2 (optional) - T4: New averaged squared gradients
__group_899__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_900__new_weights (optional) - T2: New weights
__group_900__new_gradients (optional) - T3: New gradients
__group_900__new_moment_1 (optional) - T4: New averaged gradients
__group_900__new_moment_2 (optional) - T4: New averaged squared gradients
__group_900__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_901__new_weights (optional) - T2: New weights
__group_901__new_gradients (optional) - T3: New gradients
__group_901__new_moment_1 (optional) - T4: New averaged gradients
__group_901__new_moment_2 (optional) - T4: New averaged squared gradients
__group_901__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_902__new_weights (optional) - T2: New weights
__group_902__new_gradients (optional) - T3: New gradients
__group_902__new_moment_1 (optional) - T4: New averaged gradients
__group_902__new_moment_2 (optional) - T4: New averaged squared gradients
__group_902__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_903__new_weights (optional) - T2: New weights
__group_903__new_gradients (optional) - T3: New gradients
__group_903__new_moment_1 (optional) - T4: New averaged gradients
__group_903__new_moment_2 (optional) - T4: New averaged squared gradients
__group_903__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_904__new_weights (optional) - T2: New weights
__group_904__new_gradients (optional) - T3: New gradients
__group_904__new_moment_1 (optional) - T4: New averaged gradients
__group_904__new_moment_2 (optional) - T4: New averaged squared gradients
__group_904__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_905__new_weights (optional) - T2: New weights
__group_905__new_gradients (optional) - T3: New gradients
__group_905__new_moment_1 (optional) - T4: New averaged gradients
__group_905__new_moment_2 (optional) - T4: New averaged squared gradients
__group_905__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_906__new_weights (optional) - T2: New weights
__group_906__new_gradients (optional) - T3: New gradients
__group_906__new_moment_1 (optional) - T4: New averaged gradients
__group_906__new_moment_2 (optional) - T4: New averaged squared gradients
__group_906__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_907__new_weights (optional) - T2: New weights
__group_907__new_gradients (optional) - T3: New gradients
__group_907__new_moment_1 (optional) - T4: New averaged gradients
__group_907__new_moment_2 (optional) - T4: New averaged squared gradients
__group_907__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_908__new_weights (optional) - T2: New weights
__group_908__new_gradients (optional) - T3: New gradients
__group_908__new_moment_1 (optional) - T4: New averaged gradients
__group_908__new_moment_2 (optional) - T4: New averaged squared gradients
__group_908__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_909__new_weights (optional) - T2: New weights
__group_909__new_gradients (optional) - T3: New gradients
__group_909__new_moment_1 (optional) - T4: New averaged gradients
__group_909__new_moment_2 (optional) - T4: New averaged squared gradients
__group_909__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_910__new_weights (optional) - T2: New weights
__group_910__new_gradients (optional) - T3: New gradients
__group_910__new_moment_1 (optional) - T4: New averaged gradients
__group_910__new_moment_2 (optional) - T4: New averaged squared gradients
__group_910__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_911__new_weights (optional) - T2: New weights
__group_911__new_gradients (optional) - T3: New gradients
__group_911__new_moment_1 (optional) - T4: New averaged gradients
__group_911__new_moment_2 (optional) - T4: New averaged squared gradients
__group_911__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_912__new_weights (optional) - T2: New weights
__group_912__new_gradients (optional) - T3: New gradients
__group_912__new_moment_1 (optional) - T4: New averaged gradients
__group_912__new_moment_2 (optional) - T4: New averaged squared gradients
__group_912__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_913__new_weights (optional) - T2: New weights
__group_913__new_gradients (optional) - T3: New gradients
__group_913__new_moment_1 (optional) - T4: New averaged gradients
__group_913__new_moment_2 (optional) - T4: New averaged squared gradients
__group_913__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_914__new_weights (optional) - T2: New weights
__group_914__new_gradients (optional) - T3: New gradients
__group_914__new_moment_1 (optional) - T4: New averaged gradients
__group_914__new_moment_2 (optional) - T4: New averaged squared gradients
__group_914__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_915__new_weights (optional) - T2: New weights
__group_915__new_gradients (optional) - T3: New gradients
__group_915__new_moment_1 (optional) - T4: New averaged gradients
__group_915__new_moment_2 (optional) - T4: New averaged squared gradients
__group_915__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_916__new_weights (optional) - T2: New weights
__group_916__new_gradients (optional) - T3: New gradients
__group_916__new_moment_1 (optional) - T4: New averaged gradients
__group_916__new_moment_2 (optional) - T4: New averaged squared gradients
__group_916__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_917__new_weights (optional) - T2: New weights
__group_917__new_gradients (optional) - T3: New gradients
__group_917__new_moment_1 (optional) - T4: New averaged gradients
__group_917__new_moment_2 (optional) - T4: New averaged squared gradients
__group_917__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_918__new_weights (optional) - T2: New weights
__group_918__new_gradients (optional) - T3: New gradients
__group_918__new_moment_1 (optional) - T4: New averaged gradients
__group_918__new_moment_2 (optional) - T4: New averaged squared gradients
__group_918__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_919__new_weights (optional) - T2: New weights
__group_919__new_gradients (optional) - T3: New gradients
__group_919__new_moment_1 (optional) - T4: New averaged gradients
__group_919__new_moment_2 (optional) - T4: New averaged squared gradients
__group_919__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_920__new_weights (optional) - T2: New weights
__group_920__new_gradients (optional) - T3: New gradients
__group_920__new_moment_1 (optional) - T4: New averaged gradients
__group_920__new_moment_2 (optional) - T4: New averaged squared gradients
__group_920__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_921__new_weights (optional) - T2: New weights
__group_921__new_gradients (optional) - T3: New gradients
__group_921__new_moment_1 (optional) - T4: New averaged gradients
__group_921__new_moment_2 (optional) - T4: New averaged squared gradients
__group_921__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_922__new_weights (optional) - T2: New weights
__group_922__new_gradients (optional) - T3: New gradients
__group_922__new_moment_1 (optional) - T4: New averaged gradients
__group_922__new_moment_2 (optional) - T4: New averaged squared gradients
__group_922__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_923__new_weights (optional) - T2: New weights
__group_923__new_gradients (optional) - T3: New gradients
__group_923__new_moment_1 (optional) - T4: New averaged gradients
__group_923__new_moment_2 (optional) - T4: New averaged squared gradients
__group_923__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_924__new_weights (optional) - T2: New weights
__group_924__new_gradients (optional) - T3: New gradients
__group_924__new_moment_1 (optional) - T4: New averaged gradients
__group_924__new_moment_2 (optional) - T4: New averaged squared gradients
__group_924__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_925__new_weights (optional) - T2: New weights
__group_925__new_gradients (optional) - T3: New gradients
__group_925__new_moment_1 (optional) - T4: New averaged gradients
__group_925__new_moment_2 (optional) - T4: New averaged squared gradients
__group_925__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_926__new_weights (optional) - T2: New weights
__group_926__new_gradients (optional) - T3: New gradients
__group_926__new_moment_1 (optional) - T4: New averaged gradients
__group_926__new_moment_2 (optional) - T4: New averaged squared gradients
__group_926__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_927__new_weights (optional) - T2: New weights
__group_927__new_gradients (optional) - T3: New gradients
__group_927__new_moment_1 (optional) - T4: New averaged gradients
__group_927__new_moment_2 (optional) - T4: New averaged squared gradients
__group_927__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_928__new_weights (optional) - T2: New weights
__group_928__new_gradients (optional) - T3: New gradients
__group_928__new_moment_1 (optional) - T4: New averaged gradients
__group_928__new_moment_2 (optional) - T4: New averaged squared gradients
__group_928__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_929__new_weights (optional) - T2: New weights
__group_929__new_gradients (optional) - T3: New gradients
__group_929__new_moment_1 (optional) - T4: New averaged gradients
__group_929__new_moment_2 (optional) - T4: New averaged squared gradients
__group_929__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_930__new_weights (optional) - T2: New weights
__group_930__new_gradients (optional) - T3: New gradients
__group_930__new_moment_1 (optional) - T4: New averaged gradients
__group_930__new_moment_2 (optional) - T4: New averaged squared gradients
__group_930__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_931__new_weights (optional) - T2: New weights
__group_931__new_gradients (optional) - T3: New gradients
__group_931__new_moment_1 (optional) - T4: New averaged gradients
__group_931__new_moment_2 (optional) - T4: New averaged squared gradients
__group_931__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_932__new_weights (optional) - T2: New weights
__group_932__new_gradients (optional) - T3: New gradients
__group_932__new_moment_1 (optional) - T4: New averaged gradients
__group_932__new_moment_2 (optional) - T4: New averaged squared gradients
__group_932__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_933__new_weights (optional) - T2: New weights
__group_933__new_gradients (optional) - T3: New gradients
__group_933__new_moment_1 (optional) - T4: New averaged gradients
__group_933__new_moment_2 (optional) - T4: New averaged squared gradients
__group_933__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_934__new_weights (optional) - T2: New weights
__group_934__new_gradients (optional) - T3: New gradients
__group_934__new_moment_1 (optional) - T4: New averaged gradients
__group_934__new_moment_2 (optional) - T4: New averaged squared gradients
__group_934__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_935__new_weights (optional) - T2: New weights
__group_935__new_gradients (optional) - T3: New gradients
__group_935__new_moment_1 (optional) - T4: New averaged gradients
__group_935__new_moment_2 (optional) - T4: New averaged squared gradients
__group_935__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_936__new_weights (optional) - T2: New weights
__group_936__new_gradients (optional) - T3: New gradients
__group_936__new_moment_1 (optional) - T4: New averaged gradients
__group_936__new_moment_2 (optional) - T4: New averaged squared gradients
__group_936__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_937__new_weights (optional) - T2: New weights
__group_937__new_gradients (optional) - T3: New gradients
__group_937__new_moment_1 (optional) - T4: New averaged gradients
__group_937__new_moment_2 (optional) - T4: New averaged squared gradients
__group_937__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_938__new_weights (optional) - T2: New weights
__group_938__new_gradients (optional) - T3: New gradients
__group_938__new_moment_1 (optional) - T4: New averaged gradients
__group_938__new_moment_2 (optional) - T4: New averaged squared gradients
__group_938__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_939__new_weights (optional) - T2: New weights
__group_939__new_gradients (optional) - T3: New gradients
__group_939__new_moment_1 (optional) - T4: New averaged gradients
__group_939__new_moment_2 (optional) - T4: New averaged squared gradients
__group_939__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_940__new_weights (optional) - T2: New weights
__group_940__new_gradients (optional) - T3: New gradients
__group_940__new_moment_1 (optional) - T4: New averaged gradients
__group_940__new_moment_2 (optional) - T4: New averaged squared gradients
__group_940__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_941__new_weights (optional) - T2: New weights
__group_941__new_gradients (optional) - T3: New gradients
__group_941__new_moment_1 (optional) - T4: New averaged gradients
__group_941__new_moment_2 (optional) - T4: New averaged squared gradients
__group_941__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_942__new_weights (optional) - T2: New weights
__group_942__new_gradients (optional) - T3: New gradients
__group_942__new_moment_1 (optional) - T4: New averaged gradients
__group_942__new_moment_2 (optional) - T4: New averaged squared gradients
__group_942__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_943__new_weights (optional) - T2: New weights
__group_943__new_gradients (optional) - T3: New gradients
__group_943__new_moment_1 (optional) - T4: New averaged gradients
__group_943__new_moment_2 (optional) - T4: New averaged squared gradients
__group_943__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_944__new_weights (optional) - T2: New weights
__group_944__new_gradients (optional) - T3: New gradients
__group_944__new_moment_1 (optional) - T4: New averaged gradients
__group_944__new_moment_2 (optional) - T4: New averaged squared gradients
__group_944__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_945__new_weights (optional) - T2: New weights
__group_945__new_gradients (optional) - T3: New gradients
__group_945__new_moment_1 (optional) - T4: New averaged gradients
__group_945__new_moment_2 (optional) - T4: New averaged squared gradients
__group_945__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_946__new_weights (optional) - T2: New weights
__group_946__new_gradients (optional) - T3: New gradients
__group_946__new_moment_1 (optional) - T4: New averaged gradients
__group_946__new_moment_2 (optional) - T4: New averaged squared gradients
__group_946__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_947__new_weights (optional) - T2: New weights
__group_947__new_gradients (optional) - T3: New gradients
__group_947__new_moment_1 (optional) - T4: New averaged gradients
__group_947__new_moment_2 (optional) - T4: New averaged squared gradients
__group_947__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_948__new_weights (optional) - T2: New weights
__group_948__new_gradients (optional) - T3: New gradients
__group_948__new_moment_1 (optional) - T4: New averaged gradients
__group_948__new_moment_2 (optional) - T4: New averaged squared gradients
__group_948__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_949__new_weights (optional) - T2: New weights
__group_949__new_gradients (optional) - T3: New gradients
__group_949__new_moment_1 (optional) - T4: New averaged gradients
__group_949__new_moment_2 (optional) - T4: New averaged squared gradients
__group_949__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_950__new_weights (optional) - T2: New weights
__group_950__new_gradients (optional) - T3: New gradients
__group_950__new_moment_1 (optional) - T4: New averaged gradients
__group_950__new_moment_2 (optional) - T4: New averaged squared gradients
__group_950__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_951__new_weights (optional) - T2: New weights
__group_951__new_gradients (optional) - T3: New gradients
__group_951__new_moment_1 (optional) - T4: New averaged gradients
__group_951__new_moment_2 (optional) - T4: New averaged squared gradients
__group_951__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_952__new_weights (optional) - T2: New weights
__group_952__new_gradients (optional) - T3: New gradients
__group_952__new_moment_1 (optional) - T4: New averaged gradients
__group_952__new_moment_2 (optional) - T4: New averaged squared gradients
__group_952__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_953__new_weights (optional) - T2: New weights
__group_953__new_gradients (optional) - T3: New gradients
__group_953__new_moment_1 (optional) - T4: New averaged gradients
__group_953__new_moment_2 (optional) - T4: New averaged squared gradients
__group_953__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_954__new_weights (optional) - T2: New weights
__group_954__new_gradients (optional) - T3: New gradients
__group_954__new_moment_1 (optional) - T4: New averaged gradients
__group_954__new_moment_2 (optional) - T4: New averaged squared gradients
__group_954__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_955__new_weights (optional) - T2: New weights
__group_955__new_gradients (optional) - T3: New gradients
__group_955__new_moment_1 (optional) - T4: New averaged gradients
__group_955__new_moment_2 (optional) - T4: New averaged squared gradients
__group_955__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_956__new_weights (optional) - T2: New weights
__group_956__new_gradients (optional) - T3: New gradients
__group_956__new_moment_1 (optional) - T4: New averaged gradients
__group_956__new_moment_2 (optional) - T4: New averaged squared gradients
__group_956__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_957__new_weights (optional) - T2: New weights
__group_957__new_gradients (optional) - T3: New gradients
__group_957__new_moment_1 (optional) - T4: New averaged gradients
__group_957__new_moment_2 (optional) - T4: New averaged squared gradients
__group_957__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_958__new_weights (optional) - T2: New weights
__group_958__new_gradients (optional) - T3: New gradients
__group_958__new_moment_1 (optional) - T4: New averaged gradients
__group_958__new_moment_2 (optional) - T4: New averaged squared gradients
__group_958__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_959__new_weights (optional) - T2: New weights
__group_959__new_gradients (optional) - T3: New gradients
__group_959__new_moment_1 (optional) - T4: New averaged gradients
__group_959__new_moment_2 (optional) - T4: New averaged squared gradients
__group_959__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_960__new_weights (optional) - T2: New weights
__group_960__new_gradients (optional) - T3: New gradients
__group_960__new_moment_1 (optional) - T4: New averaged gradients
__group_960__new_moment_2 (optional) - T4: New averaged squared gradients
__group_960__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_961__new_weights (optional) - T2: New weights
__group_961__new_gradients (optional) - T3: New gradients
__group_961__new_moment_1 (optional) - T4: New averaged gradients
__group_961__new_moment_2 (optional) - T4: New averaged squared gradients
__group_961__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_962__new_weights (optional) - T2: New weights
__group_962__new_gradients (optional) - T3: New gradients
__group_962__new_moment_1 (optional) - T4: New averaged gradients
__group_962__new_moment_2 (optional) - T4: New averaged squared gradients
__group_962__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_963__new_weights (optional) - T2: New weights
__group_963__new_gradients (optional) - T3: New gradients
__group_963__new_moment_1 (optional) - T4: New averaged gradients
__group_963__new_moment_2 (optional) - T4: New averaged squared gradients
__group_963__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_964__new_weights (optional) - T2: New weights
__group_964__new_gradients (optional) - T3: New gradients
__group_964__new_moment_1 (optional) - T4: New averaged gradients
__group_964__new_moment_2 (optional) - T4: New averaged squared gradients
__group_964__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_965__new_weights (optional) - T2: New weights
__group_965__new_gradients (optional) - T3: New gradients
__group_965__new_moment_1 (optional) - T4: New averaged gradients
__group_965__new_moment_2 (optional) - T4: New averaged squared gradients
__group_965__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_966__new_weights (optional) - T2: New weights
__group_966__new_gradients (optional) - T3: New gradients
__group_966__new_moment_1 (optional) - T4: New averaged gradients
__group_966__new_moment_2 (optional) - T4: New averaged squared gradients
__group_966__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_967__new_weights (optional) - T2: New weights
__group_967__new_gradients (optional) - T3: New gradients
__group_967__new_moment_1 (optional) - T4: New averaged gradients
__group_967__new_moment_2 (optional) - T4: New averaged squared gradients
__group_967__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_968__new_weights (optional) - T2: New weights
__group_968__new_gradients (optional) - T3: New gradients
__group_968__new_moment_1 (optional) - T4: New averaged gradients
__group_968__new_moment_2 (optional) - T4: New averaged squared gradients
__group_968__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_969__new_weights (optional) - T2: New weights
__group_969__new_gradients (optional) - T3: New gradients
__group_969__new_moment_1 (optional) - T4: New averaged gradients
__group_969__new_moment_2 (optional) - T4: New averaged squared gradients
__group_969__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_970__new_weights (optional) - T2: New weights
__group_970__new_gradients (optional) - T3: New gradients
__group_970__new_moment_1 (optional) - T4: New averaged gradients
__group_970__new_moment_2 (optional) - T4: New averaged squared gradients
__group_970__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_971__new_weights (optional) - T2: New weights
__group_971__new_gradients (optional) - T3: New gradients
__group_971__new_moment_1 (optional) - T4: New averaged gradients
__group_971__new_moment_2 (optional) - T4: New averaged squared gradients
__group_971__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_972__new_weights (optional) - T2: New weights
__group_972__new_gradients (optional) - T3: New gradients
__group_972__new_moment_1 (optional) - T4: New averaged gradients
__group_972__new_moment_2 (optional) - T4: New averaged squared gradients
__group_972__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_973__new_weights (optional) - T2: New weights
__group_973__new_gradients (optional) - T3: New gradients
__group_973__new_moment_1 (optional) - T4: New averaged gradients
__group_973__new_moment_2 (optional) - T4: New averaged squared gradients
__group_973__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_974__new_weights (optional) - T2: New weights
__group_974__new_gradients (optional) - T3: New gradients
__group_974__new_moment_1 (optional) - T4: New averaged gradients
__group_974__new_moment_2 (optional) - T4: New averaged squared gradients
__group_974__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_975__new_weights (optional) - T2: New weights
__group_975__new_gradients (optional) - T3: New gradients
__group_975__new_moment_1 (optional) - T4: New averaged gradients
__group_975__new_moment_2 (optional) - T4: New averaged squared gradients
__group_975__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_976__new_weights (optional) - T2: New weights
__group_976__new_gradients (optional) - T3: New gradients
__group_976__new_moment_1 (optional) - T4: New averaged gradients
__group_976__new_moment_2 (optional) - T4: New averaged squared gradients
__group_976__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_977__new_weights (optional) - T2: New weights
__group_977__new_gradients (optional) - T3: New gradients
__group_977__new_moment_1 (optional) - T4: New averaged gradients
__group_977__new_moment_2 (optional) - T4: New averaged squared gradients
__group_977__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_978__new_weights (optional) - T2: New weights
__group_978__new_gradients (optional) - T3: New gradients
__group_978__new_moment_1 (optional) - T4: New averaged gradients
__group_978__new_moment_2 (optional) - T4: New averaged squared gradients
__group_978__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_979__new_weights (optional) - T2: New weights
__group_979__new_gradients (optional) - T3: New gradients
__group_979__new_moment_1 (optional) - T4: New averaged gradients
__group_979__new_moment_2 (optional) - T4: New averaged squared gradients
__group_979__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_980__new_weights (optional) - T2: New weights
__group_980__new_gradients (optional) - T3: New gradients
__group_980__new_moment_1 (optional) - T4: New averaged gradients
__group_980__new_moment_2 (optional) - T4: New averaged squared gradients
__group_980__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_981__new_weights (optional) - T2: New weights
__group_981__new_gradients (optional) - T3: New gradients
__group_981__new_moment_1 (optional) - T4: New averaged gradients
__group_981__new_moment_2 (optional) - T4: New averaged squared gradients
__group_981__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_982__new_weights (optional) - T2: New weights
__group_982__new_gradients (optional) - T3: New gradients
__group_982__new_moment_1 (optional) - T4: New averaged gradients
__group_982__new_moment_2 (optional) - T4: New averaged squared gradients
__group_982__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_983__new_weights (optional) - T2: New weights
__group_983__new_gradients (optional) - T3: New gradients
__group_983__new_moment_1 (optional) - T4: New averaged gradients
__group_983__new_moment_2 (optional) - T4: New averaged squared gradients
__group_983__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_984__new_weights (optional) - T2: New weights
__group_984__new_gradients (optional) - T3: New gradients
__group_984__new_moment_1 (optional) - T4: New averaged gradients
__group_984__new_moment_2 (optional) - T4: New averaged squared gradients
__group_984__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_985__new_weights (optional) - T2: New weights
__group_985__new_gradients (optional) - T3: New gradients
__group_985__new_moment_1 (optional) - T4: New averaged gradients
__group_985__new_moment_2 (optional) - T4: New averaged squared gradients
__group_985__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_986__new_weights (optional) - T2: New weights
__group_986__new_gradients (optional) - T3: New gradients
__group_986__new_moment_1 (optional) - T4: New averaged gradients
__group_986__new_moment_2 (optional) - T4: New averaged squared gradients
__group_986__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_987__new_weights (optional) - T2: New weights
__group_987__new_gradients (optional) - T3: New gradients
__group_987__new_moment_1 (optional) - T4: New averaged gradients
__group_987__new_moment_2 (optional) - T4: New averaged squared gradients
__group_987__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_988__new_weights (optional) - T2: New weights
__group_988__new_gradients (optional) - T3: New gradients
__group_988__new_moment_1 (optional) - T4: New averaged gradients
__group_988__new_moment_2 (optional) - T4: New averaged squared gradients
__group_988__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_989__new_weights (optional) - T2: New weights
__group_989__new_gradients (optional) - T3: New gradients
__group_989__new_moment_1 (optional) - T4: New averaged gradients
__group_989__new_moment_2 (optional) - T4: New averaged squared gradients
__group_989__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_990__new_weights (optional) - T2: New weights
__group_990__new_gradients (optional) - T3: New gradients
__group_990__new_moment_1 (optional) - T4: New averaged gradients
__group_990__new_moment_2 (optional) - T4: New averaged squared gradients
__group_990__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_991__new_weights (optional) - T2: New weights
__group_991__new_gradients (optional) - T3: New gradients
__group_991__new_moment_1 (optional) - T4: New averaged gradients
__group_991__new_moment_2 (optional) - T4: New averaged squared gradients
__group_991__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_992__new_weights (optional) - T2: New weights
__group_992__new_gradients (optional) - T3: New gradients
__group_992__new_moment_1 (optional) - T4: New averaged gradients
__group_992__new_moment_2 (optional) - T4: New averaged squared gradients
__group_992__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_993__new_weights (optional) - T2: New weights
__group_993__new_gradients (optional) - T3: New gradients
__group_993__new_moment_1 (optional) - T4: New averaged gradients
__group_993__new_moment_2 (optional) - T4: New averaged squared gradients
__group_993__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_994__new_weights (optional) - T2: New weights
__group_994__new_gradients (optional) - T3: New gradients
__group_994__new_moment_1 (optional) - T4: New averaged gradients
__group_994__new_moment_2 (optional) - T4: New averaged squared gradients
__group_994__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_995__new_weights (optional) - T2: New weights
__group_995__new_gradients (optional) - T3: New gradients
__group_995__new_moment_1 (optional) - T4: New averaged gradients
__group_995__new_moment_2 (optional) - T4: New averaged squared gradients
__group_995__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_996__new_weights (optional) - T2: New weights
__group_996__new_gradients (optional) - T3: New gradients
__group_996__new_moment_1 (optional) - T4: New averaged gradients
__group_996__new_moment_2 (optional) - T4: New averaged squared gradients
__group_996__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_997__new_weights (optional) - T2: New weights
__group_997__new_gradients (optional) - T3: New gradients
__group_997__new_moment_1 (optional) - T4: New averaged gradients
__group_997__new_moment_2 (optional) - T4: New averaged squared gradients
__group_997__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_998__new_weights (optional) - T2: New weights
__group_998__new_gradients (optional) - T3: New gradients
__group_998__new_moment_1 (optional) - T4: New averaged gradients
__group_998__new_moment_2 (optional) - T4: New averaged squared gradients
__group_998__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_999__new_weights (optional) - T2: New weights
__group_999__new_gradients (optional) - T3: New gradients
__group_999__new_moment_1 (optional) - T4: New averaged gradients
__group_999__new_moment_2 (optional) - T4: New averaged squared gradients
__group_999__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1000__new_weights (optional) - T2: New weights
__group_1000__new_gradients (optional) - T3: New gradients
__group_1000__new_moment_1 (optional) - T4: New averaged gradients
__group_1000__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1000__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1001__new_weights (optional) - T2: New weights
__group_1001__new_gradients (optional) - T3: New gradients
__group_1001__new_moment_1 (optional) - T4: New averaged gradients
__group_1001__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1001__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1002__new_weights (optional) - T2: New weights
__group_1002__new_gradients (optional) - T3: New gradients
__group_1002__new_moment_1 (optional) - T4: New averaged gradients
__group_1002__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1002__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1003__new_weights (optional) - T2: New weights
__group_1003__new_gradients (optional) - T3: New gradients
__group_1003__new_moment_1 (optional) - T4: New averaged gradients
__group_1003__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1003__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1004__new_weights (optional) - T2: New weights
__group_1004__new_gradients (optional) - T3: New gradients
__group_1004__new_moment_1 (optional) - T4: New averaged gradients
__group_1004__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1004__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1005__new_weights (optional) - T2: New weights
__group_1005__new_gradients (optional) - T3: New gradients
__group_1005__new_moment_1 (optional) - T4: New averaged gradients
__group_1005__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1005__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1006__new_weights (optional) - T2: New weights
__group_1006__new_gradients (optional) - T3: New gradients
__group_1006__new_moment_1 (optional) - T4: New averaged gradients
__group_1006__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1006__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1007__new_weights (optional) - T2: New weights
__group_1007__new_gradients (optional) - T3: New gradients
__group_1007__new_moment_1 (optional) - T4: New averaged gradients
__group_1007__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1007__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1008__new_weights (optional) - T2: New weights
__group_1008__new_gradients (optional) - T3: New gradients
__group_1008__new_moment_1 (optional) - T4: New averaged gradients
__group_1008__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1008__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1009__new_weights (optional) - T2: New weights
__group_1009__new_gradients (optional) - T3: New gradients
__group_1009__new_moment_1 (optional) - T4: New averaged gradients
__group_1009__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1009__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1010__new_weights (optional) - T2: New weights
__group_1010__new_gradients (optional) - T3: New gradients
__group_1010__new_moment_1 (optional) - T4: New averaged gradients
__group_1010__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1010__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1011__new_weights (optional) - T2: New weights
__group_1011__new_gradients (optional) - T3: New gradients
__group_1011__new_moment_1 (optional) - T4: New averaged gradients
__group_1011__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1011__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1012__new_weights (optional) - T2: New weights
__group_1012__new_gradients (optional) - T3: New gradients
__group_1012__new_moment_1 (optional) - T4: New averaged gradients
__group_1012__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1012__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1013__new_weights (optional) - T2: New weights
__group_1013__new_gradients (optional) - T3: New gradients
__group_1013__new_moment_1 (optional) - T4: New averaged gradients
__group_1013__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1013__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1014__new_weights (optional) - T2: New weights
__group_1014__new_gradients (optional) - T3: New gradients
__group_1014__new_moment_1 (optional) - T4: New averaged gradients
__group_1014__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1014__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1015__new_weights (optional) - T2: New weights
__group_1015__new_gradients (optional) - T3: New gradients
__group_1015__new_moment_1 (optional) - T4: New averaged gradients
__group_1015__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1015__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1016__new_weights (optional) - T2: New weights
__group_1016__new_gradients (optional) - T3: New gradients
__group_1016__new_moment_1 (optional) - T4: New averaged gradients
__group_1016__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1016__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1017__new_weights (optional) - T2: New weights
__group_1017__new_gradients (optional) - T3: New gradients
__group_1017__new_moment_1 (optional) - T4: New averaged gradients
__group_1017__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1017__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1018__new_weights (optional) - T2: New weights
__group_1018__new_gradients (optional) - T3: New gradients
__group_1018__new_moment_1 (optional) - T4: New averaged gradients
__group_1018__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1018__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1019__new_weights (optional) - T2: New weights
__group_1019__new_gradients (optional) - T3: New gradients
__group_1019__new_moment_1 (optional) - T4: New averaged gradients
__group_1019__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1019__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1020__new_weights (optional) - T2: New weights
__group_1020__new_gradients (optional) - T3: New gradients
__group_1020__new_moment_1 (optional) - T4: New averaged gradients
__group_1020__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1020__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1021__new_weights (optional) - T2: New weights
__group_1021__new_gradients (optional) - T3: New gradients
__group_1021__new_moment_1 (optional) - T4: New averaged gradients
__group_1021__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1021__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1022__new_weights (optional) - T2: New weights
__group_1022__new_gradients (optional) - T3: New gradients
__group_1022__new_moment_1 (optional) - T4: New averaged gradients
__group_1022__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1022__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1023__new_weights (optional) - T2: New weights
__group_1023__new_gradients (optional) - T3: New gradients
__group_1023__new_moment_1 (optional) - T4: New averaged gradients
__group_1023__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1023__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
OnnxComMicrosoftLambOptimizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLambOptimizer_1(*args, **kwargs)#
Version
name: LambOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
alpha: Coefficient of previous gradient in running average. Default value is
?.beta: Coefficient of previous squared gradient in running average.The effective learning rate is computed by r = R / (1 + T * decay_factor). Default to 0 so that increasing update counts doesn’t reduce the learning rate. Default value is
?.do_bias_correction: Compute unbiased 1st and 2nd momentums. Default value is
?.epsilon: Small scalar to avoid dividing by zero. Default value is
?.lambda: Regularization coefficient of 0.5 * lambda * ||X||_2^2. Default to 0, which means no regularization. Default value is
?.max_norm_clip: clip threshold of gradients. Default value is
?.ratio_max: Upper bound on confidence ratio. Default value is
?.ratio_min: Lower bound on confidence ratio. Default value is
?.
Inputs
Between 0 and 5125 inputs.
update_signal (optional, heterogeneous) - T_BOOL: This signal indicates if weight tensors should be updated.
loss_scale (optional, heterogeneous) - T2: Loss scale for mixed precision training.
gradient_norm (optional, heterogeneous) - T_GRAD_NORM: Norm of global gradient.
R (optional, heterogeneous) - T1: The initial learning rate.
step (optional, heterogeneous) - TInt64: One-based index of the current training iteration.
__group_0__weights (optional) - T2: weights to optimize.
__group_0__gradients (optional) - T3: gradients computed in this iteration.
__group_0__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_0__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_0__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1__weights (optional) - T2: weights to optimize.
__group_1__gradients (optional) - T3: gradients computed in this iteration.
__group_1__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_2__weights (optional) - T2: weights to optimize.
__group_2__gradients (optional) - T3: gradients computed in this iteration.
__group_2__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_2__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_2__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_3__weights (optional) - T2: weights to optimize.
__group_3__gradients (optional) - T3: gradients computed in this iteration.
__group_3__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_3__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_3__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_4__weights (optional) - T2: weights to optimize.
__group_4__gradients (optional) - T3: gradients computed in this iteration.
__group_4__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_4__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_4__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_5__weights (optional) - T2: weights to optimize.
__group_5__gradients (optional) - T3: gradients computed in this iteration.
__group_5__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_5__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_5__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_6__weights (optional) - T2: weights to optimize.
__group_6__gradients (optional) - T3: gradients computed in this iteration.
__group_6__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_6__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_6__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_7__weights (optional) - T2: weights to optimize.
__group_7__gradients (optional) - T3: gradients computed in this iteration.
__group_7__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_7__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_7__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_8__weights (optional) - T2: weights to optimize.
__group_8__gradients (optional) - T3: gradients computed in this iteration.
__group_8__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_8__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_8__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_9__weights (optional) - T2: weights to optimize.
__group_9__gradients (optional) - T3: gradients computed in this iteration.
__group_9__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_9__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_9__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_10__weights (optional) - T2: weights to optimize.
__group_10__gradients (optional) - T3: gradients computed in this iteration.
__group_10__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_10__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_10__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_11__weights (optional) - T2: weights to optimize.
__group_11__gradients (optional) - T3: gradients computed in this iteration.
__group_11__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_11__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_11__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_12__weights (optional) - T2: weights to optimize.
__group_12__gradients (optional) - T3: gradients computed in this iteration.
__group_12__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_12__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_12__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_13__weights (optional) - T2: weights to optimize.
__group_13__gradients (optional) - T3: gradients computed in this iteration.
__group_13__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_13__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_13__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_14__weights (optional) - T2: weights to optimize.
__group_14__gradients (optional) - T3: gradients computed in this iteration.
__group_14__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_14__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_14__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_15__weights (optional) - T2: weights to optimize.
__group_15__gradients (optional) - T3: gradients computed in this iteration.
__group_15__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_15__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_15__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_16__weights (optional) - T2: weights to optimize.
__group_16__gradients (optional) - T3: gradients computed in this iteration.
__group_16__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_16__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_16__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_17__weights (optional) - T2: weights to optimize.
__group_17__gradients (optional) - T3: gradients computed in this iteration.
__group_17__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_17__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_17__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_18__weights (optional) - T2: weights to optimize.
__group_18__gradients (optional) - T3: gradients computed in this iteration.
__group_18__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_18__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_18__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_19__weights (optional) - T2: weights to optimize.
__group_19__gradients (optional) - T3: gradients computed in this iteration.
__group_19__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_19__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_19__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_20__weights (optional) - T2: weights to optimize.
__group_20__gradients (optional) - T3: gradients computed in this iteration.
__group_20__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_20__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_20__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_21__weights (optional) - T2: weights to optimize.
__group_21__gradients (optional) - T3: gradients computed in this iteration.
__group_21__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_21__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_21__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_22__weights (optional) - T2: weights to optimize.
__group_22__gradients (optional) - T3: gradients computed in this iteration.
__group_22__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_22__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_22__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_23__weights (optional) - T2: weights to optimize.
__group_23__gradients (optional) - T3: gradients computed in this iteration.
__group_23__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_23__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_23__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_24__weights (optional) - T2: weights to optimize.
__group_24__gradients (optional) - T3: gradients computed in this iteration.
__group_24__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_24__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_24__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_25__weights (optional) - T2: weights to optimize.
__group_25__gradients (optional) - T3: gradients computed in this iteration.
__group_25__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_25__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_25__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_26__weights (optional) - T2: weights to optimize.
__group_26__gradients (optional) - T3: gradients computed in this iteration.
__group_26__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_26__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_26__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_27__weights (optional) - T2: weights to optimize.
__group_27__gradients (optional) - T3: gradients computed in this iteration.
__group_27__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_27__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_27__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_28__weights (optional) - T2: weights to optimize.
__group_28__gradients (optional) - T3: gradients computed in this iteration.
__group_28__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_28__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_28__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_29__weights (optional) - T2: weights to optimize.
__group_29__gradients (optional) - T3: gradients computed in this iteration.
__group_29__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_29__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_29__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_30__weights (optional) - T2: weights to optimize.
__group_30__gradients (optional) - T3: gradients computed in this iteration.
__group_30__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_30__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_30__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_31__weights (optional) - T2: weights to optimize.
__group_31__gradients (optional) - T3: gradients computed in this iteration.
__group_31__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_31__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_31__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_32__weights (optional) - T2: weights to optimize.
__group_32__gradients (optional) - T3: gradients computed in this iteration.
__group_32__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_32__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_32__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_33__weights (optional) - T2: weights to optimize.
__group_33__gradients (optional) - T3: gradients computed in this iteration.
__group_33__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_33__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_33__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_34__weights (optional) - T2: weights to optimize.
__group_34__gradients (optional) - T3: gradients computed in this iteration.
__group_34__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_34__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_34__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_35__weights (optional) - T2: weights to optimize.
__group_35__gradients (optional) - T3: gradients computed in this iteration.
__group_35__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_35__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_35__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_36__weights (optional) - T2: weights to optimize.
__group_36__gradients (optional) - T3: gradients computed in this iteration.
__group_36__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_36__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_36__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_37__weights (optional) - T2: weights to optimize.
__group_37__gradients (optional) - T3: gradients computed in this iteration.
__group_37__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_37__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_37__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_38__weights (optional) - T2: weights to optimize.
__group_38__gradients (optional) - T3: gradients computed in this iteration.
__group_38__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_38__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_38__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_39__weights (optional) - T2: weights to optimize.
__group_39__gradients (optional) - T3: gradients computed in this iteration.
__group_39__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_39__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_39__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_40__weights (optional) - T2: weights to optimize.
__group_40__gradients (optional) - T3: gradients computed in this iteration.
__group_40__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_40__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_40__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_41__weights (optional) - T2: weights to optimize.
__group_41__gradients (optional) - T3: gradients computed in this iteration.
__group_41__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_41__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_41__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_42__weights (optional) - T2: weights to optimize.
__group_42__gradients (optional) - T3: gradients computed in this iteration.
__group_42__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_42__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_42__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_43__weights (optional) - T2: weights to optimize.
__group_43__gradients (optional) - T3: gradients computed in this iteration.
__group_43__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_43__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_43__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_44__weights (optional) - T2: weights to optimize.
__group_44__gradients (optional) - T3: gradients computed in this iteration.
__group_44__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_44__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_44__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_45__weights (optional) - T2: weights to optimize.
__group_45__gradients (optional) - T3: gradients computed in this iteration.
__group_45__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_45__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_45__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_46__weights (optional) - T2: weights to optimize.
__group_46__gradients (optional) - T3: gradients computed in this iteration.
__group_46__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_46__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_46__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_47__weights (optional) - T2: weights to optimize.
__group_47__gradients (optional) - T3: gradients computed in this iteration.
__group_47__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_47__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_47__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_48__weights (optional) - T2: weights to optimize.
__group_48__gradients (optional) - T3: gradients computed in this iteration.
__group_48__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_48__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_48__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_49__weights (optional) - T2: weights to optimize.
__group_49__gradients (optional) - T3: gradients computed in this iteration.
__group_49__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_49__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_49__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_50__weights (optional) - T2: weights to optimize.
__group_50__gradients (optional) - T3: gradients computed in this iteration.
__group_50__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_50__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_50__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_51__weights (optional) - T2: weights to optimize.
__group_51__gradients (optional) - T3: gradients computed in this iteration.
__group_51__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_51__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_51__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_52__weights (optional) - T2: weights to optimize.
__group_52__gradients (optional) - T3: gradients computed in this iteration.
__group_52__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_52__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_52__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_53__weights (optional) - T2: weights to optimize.
__group_53__gradients (optional) - T3: gradients computed in this iteration.
__group_53__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_53__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_53__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_54__weights (optional) - T2: weights to optimize.
__group_54__gradients (optional) - T3: gradients computed in this iteration.
__group_54__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_54__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_54__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_55__weights (optional) - T2: weights to optimize.
__group_55__gradients (optional) - T3: gradients computed in this iteration.
__group_55__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_55__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_55__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_56__weights (optional) - T2: weights to optimize.
__group_56__gradients (optional) - T3: gradients computed in this iteration.
__group_56__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_56__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_56__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_57__weights (optional) - T2: weights to optimize.
__group_57__gradients (optional) - T3: gradients computed in this iteration.
__group_57__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_57__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_57__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_58__weights (optional) - T2: weights to optimize.
__group_58__gradients (optional) - T3: gradients computed in this iteration.
__group_58__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_58__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_58__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_59__weights (optional) - T2: weights to optimize.
__group_59__gradients (optional) - T3: gradients computed in this iteration.
__group_59__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_59__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_59__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_60__weights (optional) - T2: weights to optimize.
__group_60__gradients (optional) - T3: gradients computed in this iteration.
__group_60__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_60__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_60__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_61__weights (optional) - T2: weights to optimize.
__group_61__gradients (optional) - T3: gradients computed in this iteration.
__group_61__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_61__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_61__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_62__weights (optional) - T2: weights to optimize.
__group_62__gradients (optional) - T3: gradients computed in this iteration.
__group_62__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_62__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_62__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_63__weights (optional) - T2: weights to optimize.
__group_63__gradients (optional) - T3: gradients computed in this iteration.
__group_63__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_63__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_63__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_64__weights (optional) - T2: weights to optimize.
__group_64__gradients (optional) - T3: gradients computed in this iteration.
__group_64__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_64__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_64__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_65__weights (optional) - T2: weights to optimize.
__group_65__gradients (optional) - T3: gradients computed in this iteration.
__group_65__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_65__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_65__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_66__weights (optional) - T2: weights to optimize.
__group_66__gradients (optional) - T3: gradients computed in this iteration.
__group_66__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_66__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_66__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_67__weights (optional) - T2: weights to optimize.
__group_67__gradients (optional) - T3: gradients computed in this iteration.
__group_67__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_67__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_67__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_68__weights (optional) - T2: weights to optimize.
__group_68__gradients (optional) - T3: gradients computed in this iteration.
__group_68__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_68__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_68__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_69__weights (optional) - T2: weights to optimize.
__group_69__gradients (optional) - T3: gradients computed in this iteration.
__group_69__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_69__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_69__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_70__weights (optional) - T2: weights to optimize.
__group_70__gradients (optional) - T3: gradients computed in this iteration.
__group_70__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_70__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_70__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_71__weights (optional) - T2: weights to optimize.
__group_71__gradients (optional) - T3: gradients computed in this iteration.
__group_71__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_71__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_71__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_72__weights (optional) - T2: weights to optimize.
__group_72__gradients (optional) - T3: gradients computed in this iteration.
__group_72__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_72__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_72__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_73__weights (optional) - T2: weights to optimize.
__group_73__gradients (optional) - T3: gradients computed in this iteration.
__group_73__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_73__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_73__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_74__weights (optional) - T2: weights to optimize.
__group_74__gradients (optional) - T3: gradients computed in this iteration.
__group_74__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_74__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_74__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_75__weights (optional) - T2: weights to optimize.
__group_75__gradients (optional) - T3: gradients computed in this iteration.
__group_75__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_75__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_75__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_76__weights (optional) - T2: weights to optimize.
__group_76__gradients (optional) - T3: gradients computed in this iteration.
__group_76__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_76__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_76__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_77__weights (optional) - T2: weights to optimize.
__group_77__gradients (optional) - T3: gradients computed in this iteration.
__group_77__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_77__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_77__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_78__weights (optional) - T2: weights to optimize.
__group_78__gradients (optional) - T3: gradients computed in this iteration.
__group_78__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_78__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_78__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_79__weights (optional) - T2: weights to optimize.
__group_79__gradients (optional) - T3: gradients computed in this iteration.
__group_79__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_79__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_79__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_80__weights (optional) - T2: weights to optimize.
__group_80__gradients (optional) - T3: gradients computed in this iteration.
__group_80__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_80__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_80__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_81__weights (optional) - T2: weights to optimize.
__group_81__gradients (optional) - T3: gradients computed in this iteration.
__group_81__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_81__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_81__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_82__weights (optional) - T2: weights to optimize.
__group_82__gradients (optional) - T3: gradients computed in this iteration.
__group_82__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_82__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_82__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_83__weights (optional) - T2: weights to optimize.
__group_83__gradients (optional) - T3: gradients computed in this iteration.
__group_83__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_83__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_83__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_84__weights (optional) - T2: weights to optimize.
__group_84__gradients (optional) - T3: gradients computed in this iteration.
__group_84__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_84__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_84__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_85__weights (optional) - T2: weights to optimize.
__group_85__gradients (optional) - T3: gradients computed in this iteration.
__group_85__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_85__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_85__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_86__weights (optional) - T2: weights to optimize.
__group_86__gradients (optional) - T3: gradients computed in this iteration.
__group_86__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_86__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_86__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_87__weights (optional) - T2: weights to optimize.
__group_87__gradients (optional) - T3: gradients computed in this iteration.
__group_87__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_87__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_87__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_88__weights (optional) - T2: weights to optimize.
__group_88__gradients (optional) - T3: gradients computed in this iteration.
__group_88__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_88__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_88__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_89__weights (optional) - T2: weights to optimize.
__group_89__gradients (optional) - T3: gradients computed in this iteration.
__group_89__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_89__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_89__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_90__weights (optional) - T2: weights to optimize.
__group_90__gradients (optional) - T3: gradients computed in this iteration.
__group_90__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_90__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_90__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_91__weights (optional) - T2: weights to optimize.
__group_91__gradients (optional) - T3: gradients computed in this iteration.
__group_91__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_91__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_91__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_92__weights (optional) - T2: weights to optimize.
__group_92__gradients (optional) - T3: gradients computed in this iteration.
__group_92__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_92__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_92__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_93__weights (optional) - T2: weights to optimize.
__group_93__gradients (optional) - T3: gradients computed in this iteration.
__group_93__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_93__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_93__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_94__weights (optional) - T2: weights to optimize.
__group_94__gradients (optional) - T3: gradients computed in this iteration.
__group_94__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_94__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_94__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_95__weights (optional) - T2: weights to optimize.
__group_95__gradients (optional) - T3: gradients computed in this iteration.
__group_95__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_95__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_95__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_96__weights (optional) - T2: weights to optimize.
__group_96__gradients (optional) - T3: gradients computed in this iteration.
__group_96__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_96__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_96__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_97__weights (optional) - T2: weights to optimize.
__group_97__gradients (optional) - T3: gradients computed in this iteration.
__group_97__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_97__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_97__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_98__weights (optional) - T2: weights to optimize.
__group_98__gradients (optional) - T3: gradients computed in this iteration.
__group_98__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_98__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_98__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_99__weights (optional) - T2: weights to optimize.
__group_99__gradients (optional) - T3: gradients computed in this iteration.
__group_99__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_99__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_99__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_100__weights (optional) - T2: weights to optimize.
__group_100__gradients (optional) - T3: gradients computed in this iteration.
__group_100__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_100__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_100__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_101__weights (optional) - T2: weights to optimize.
__group_101__gradients (optional) - T3: gradients computed in this iteration.
__group_101__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_101__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_101__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_102__weights (optional) - T2: weights to optimize.
__group_102__gradients (optional) - T3: gradients computed in this iteration.
__group_102__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_102__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_102__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_103__weights (optional) - T2: weights to optimize.
__group_103__gradients (optional) - T3: gradients computed in this iteration.
__group_103__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_103__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_103__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_104__weights (optional) - T2: weights to optimize.
__group_104__gradients (optional) - T3: gradients computed in this iteration.
__group_104__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_104__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_104__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_105__weights (optional) - T2: weights to optimize.
__group_105__gradients (optional) - T3: gradients computed in this iteration.
__group_105__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_105__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_105__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_106__weights (optional) - T2: weights to optimize.
__group_106__gradients (optional) - T3: gradients computed in this iteration.
__group_106__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_106__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_106__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_107__weights (optional) - T2: weights to optimize.
__group_107__gradients (optional) - T3: gradients computed in this iteration.
__group_107__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_107__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_107__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_108__weights (optional) - T2: weights to optimize.
__group_108__gradients (optional) - T3: gradients computed in this iteration.
__group_108__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_108__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_108__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_109__weights (optional) - T2: weights to optimize.
__group_109__gradients (optional) - T3: gradients computed in this iteration.
__group_109__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_109__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_109__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_110__weights (optional) - T2: weights to optimize.
__group_110__gradients (optional) - T3: gradients computed in this iteration.
__group_110__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_110__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_110__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_111__weights (optional) - T2: weights to optimize.
__group_111__gradients (optional) - T3: gradients computed in this iteration.
__group_111__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_111__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_111__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_112__weights (optional) - T2: weights to optimize.
__group_112__gradients (optional) - T3: gradients computed in this iteration.
__group_112__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_112__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_112__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_113__weights (optional) - T2: weights to optimize.
__group_113__gradients (optional) - T3: gradients computed in this iteration.
__group_113__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_113__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_113__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_114__weights (optional) - T2: weights to optimize.
__group_114__gradients (optional) - T3: gradients computed in this iteration.
__group_114__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_114__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_114__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_115__weights (optional) - T2: weights to optimize.
__group_115__gradients (optional) - T3: gradients computed in this iteration.
__group_115__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_115__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_115__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_116__weights (optional) - T2: weights to optimize.
__group_116__gradients (optional) - T3: gradients computed in this iteration.
__group_116__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_116__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_116__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_117__weights (optional) - T2: weights to optimize.
__group_117__gradients (optional) - T3: gradients computed in this iteration.
__group_117__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_117__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_117__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_118__weights (optional) - T2: weights to optimize.
__group_118__gradients (optional) - T3: gradients computed in this iteration.
__group_118__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_118__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_118__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_119__weights (optional) - T2: weights to optimize.
__group_119__gradients (optional) - T3: gradients computed in this iteration.
__group_119__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_119__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_119__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_120__weights (optional) - T2: weights to optimize.
__group_120__gradients (optional) - T3: gradients computed in this iteration.
__group_120__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_120__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_120__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_121__weights (optional) - T2: weights to optimize.
__group_121__gradients (optional) - T3: gradients computed in this iteration.
__group_121__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_121__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_121__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_122__weights (optional) - T2: weights to optimize.
__group_122__gradients (optional) - T3: gradients computed in this iteration.
__group_122__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_122__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_122__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_123__weights (optional) - T2: weights to optimize.
__group_123__gradients (optional) - T3: gradients computed in this iteration.
__group_123__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_123__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_123__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_124__weights (optional) - T2: weights to optimize.
__group_124__gradients (optional) - T3: gradients computed in this iteration.
__group_124__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_124__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_124__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_125__weights (optional) - T2: weights to optimize.
__group_125__gradients (optional) - T3: gradients computed in this iteration.
__group_125__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_125__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_125__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_126__weights (optional) - T2: weights to optimize.
__group_126__gradients (optional) - T3: gradients computed in this iteration.
__group_126__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_126__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_126__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_127__weights (optional) - T2: weights to optimize.
__group_127__gradients (optional) - T3: gradients computed in this iteration.
__group_127__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_127__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_127__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_128__weights (optional) - T2: weights to optimize.
__group_128__gradients (optional) - T3: gradients computed in this iteration.
__group_128__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_128__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_128__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_129__weights (optional) - T2: weights to optimize.
__group_129__gradients (optional) - T3: gradients computed in this iteration.
__group_129__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_129__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_129__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_130__weights (optional) - T2: weights to optimize.
__group_130__gradients (optional) - T3: gradients computed in this iteration.
__group_130__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_130__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_130__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_131__weights (optional) - T2: weights to optimize.
__group_131__gradients (optional) - T3: gradients computed in this iteration.
__group_131__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_131__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_131__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_132__weights (optional) - T2: weights to optimize.
__group_132__gradients (optional) - T3: gradients computed in this iteration.
__group_132__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_132__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_132__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_133__weights (optional) - T2: weights to optimize.
__group_133__gradients (optional) - T3: gradients computed in this iteration.
__group_133__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_133__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_133__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_134__weights (optional) - T2: weights to optimize.
__group_134__gradients (optional) - T3: gradients computed in this iteration.
__group_134__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_134__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_134__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_135__weights (optional) - T2: weights to optimize.
__group_135__gradients (optional) - T3: gradients computed in this iteration.
__group_135__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_135__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_135__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_136__weights (optional) - T2: weights to optimize.
__group_136__gradients (optional) - T3: gradients computed in this iteration.
__group_136__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_136__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_136__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_137__weights (optional) - T2: weights to optimize.
__group_137__gradients (optional) - T3: gradients computed in this iteration.
__group_137__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_137__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_137__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_138__weights (optional) - T2: weights to optimize.
__group_138__gradients (optional) - T3: gradients computed in this iteration.
__group_138__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_138__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_138__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_139__weights (optional) - T2: weights to optimize.
__group_139__gradients (optional) - T3: gradients computed in this iteration.
__group_139__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_139__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_139__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_140__weights (optional) - T2: weights to optimize.
__group_140__gradients (optional) - T3: gradients computed in this iteration.
__group_140__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_140__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_140__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_141__weights (optional) - T2: weights to optimize.
__group_141__gradients (optional) - T3: gradients computed in this iteration.
__group_141__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_141__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_141__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_142__weights (optional) - T2: weights to optimize.
__group_142__gradients (optional) - T3: gradients computed in this iteration.
__group_142__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_142__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_142__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_143__weights (optional) - T2: weights to optimize.
__group_143__gradients (optional) - T3: gradients computed in this iteration.
__group_143__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_143__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_143__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_144__weights (optional) - T2: weights to optimize.
__group_144__gradients (optional) - T3: gradients computed in this iteration.
__group_144__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_144__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_144__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_145__weights (optional) - T2: weights to optimize.
__group_145__gradients (optional) - T3: gradients computed in this iteration.
__group_145__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_145__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_145__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_146__weights (optional) - T2: weights to optimize.
__group_146__gradients (optional) - T3: gradients computed in this iteration.
__group_146__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_146__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_146__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_147__weights (optional) - T2: weights to optimize.
__group_147__gradients (optional) - T3: gradients computed in this iteration.
__group_147__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_147__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_147__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_148__weights (optional) - T2: weights to optimize.
__group_148__gradients (optional) - T3: gradients computed in this iteration.
__group_148__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_148__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_148__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_149__weights (optional) - T2: weights to optimize.
__group_149__gradients (optional) - T3: gradients computed in this iteration.
__group_149__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_149__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_149__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_150__weights (optional) - T2: weights to optimize.
__group_150__gradients (optional) - T3: gradients computed in this iteration.
__group_150__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_150__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_150__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_151__weights (optional) - T2: weights to optimize.
__group_151__gradients (optional) - T3: gradients computed in this iteration.
__group_151__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_151__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_151__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_152__weights (optional) - T2: weights to optimize.
__group_152__gradients (optional) - T3: gradients computed in this iteration.
__group_152__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_152__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_152__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_153__weights (optional) - T2: weights to optimize.
__group_153__gradients (optional) - T3: gradients computed in this iteration.
__group_153__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_153__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_153__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_154__weights (optional) - T2: weights to optimize.
__group_154__gradients (optional) - T3: gradients computed in this iteration.
__group_154__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_154__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_154__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_155__weights (optional) - T2: weights to optimize.
__group_155__gradients (optional) - T3: gradients computed in this iteration.
__group_155__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_155__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_155__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_156__weights (optional) - T2: weights to optimize.
__group_156__gradients (optional) - T3: gradients computed in this iteration.
__group_156__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_156__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_156__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_157__weights (optional) - T2: weights to optimize.
__group_157__gradients (optional) - T3: gradients computed in this iteration.
__group_157__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_157__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_157__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_158__weights (optional) - T2: weights to optimize.
__group_158__gradients (optional) - T3: gradients computed in this iteration.
__group_158__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_158__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_158__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_159__weights (optional) - T2: weights to optimize.
__group_159__gradients (optional) - T3: gradients computed in this iteration.
__group_159__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_159__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_159__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_160__weights (optional) - T2: weights to optimize.
__group_160__gradients (optional) - T3: gradients computed in this iteration.
__group_160__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_160__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_160__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_161__weights (optional) - T2: weights to optimize.
__group_161__gradients (optional) - T3: gradients computed in this iteration.
__group_161__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_161__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_161__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_162__weights (optional) - T2: weights to optimize.
__group_162__gradients (optional) - T3: gradients computed in this iteration.
__group_162__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_162__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_162__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_163__weights (optional) - T2: weights to optimize.
__group_163__gradients (optional) - T3: gradients computed in this iteration.
__group_163__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_163__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_163__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_164__weights (optional) - T2: weights to optimize.
__group_164__gradients (optional) - T3: gradients computed in this iteration.
__group_164__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_164__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_164__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_165__weights (optional) - T2: weights to optimize.
__group_165__gradients (optional) - T3: gradients computed in this iteration.
__group_165__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_165__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_165__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_166__weights (optional) - T2: weights to optimize.
__group_166__gradients (optional) - T3: gradients computed in this iteration.
__group_166__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_166__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_166__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_167__weights (optional) - T2: weights to optimize.
__group_167__gradients (optional) - T3: gradients computed in this iteration.
__group_167__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_167__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_167__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_168__weights (optional) - T2: weights to optimize.
__group_168__gradients (optional) - T3: gradients computed in this iteration.
__group_168__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_168__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_168__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_169__weights (optional) - T2: weights to optimize.
__group_169__gradients (optional) - T3: gradients computed in this iteration.
__group_169__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_169__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_169__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_170__weights (optional) - T2: weights to optimize.
__group_170__gradients (optional) - T3: gradients computed in this iteration.
__group_170__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_170__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_170__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_171__weights (optional) - T2: weights to optimize.
__group_171__gradients (optional) - T3: gradients computed in this iteration.
__group_171__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_171__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_171__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_172__weights (optional) - T2: weights to optimize.
__group_172__gradients (optional) - T3: gradients computed in this iteration.
__group_172__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_172__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_172__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_173__weights (optional) - T2: weights to optimize.
__group_173__gradients (optional) - T3: gradients computed in this iteration.
__group_173__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_173__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_173__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_174__weights (optional) - T2: weights to optimize.
__group_174__gradients (optional) - T3: gradients computed in this iteration.
__group_174__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_174__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_174__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_175__weights (optional) - T2: weights to optimize.
__group_175__gradients (optional) - T3: gradients computed in this iteration.
__group_175__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_175__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_175__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_176__weights (optional) - T2: weights to optimize.
__group_176__gradients (optional) - T3: gradients computed in this iteration.
__group_176__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_176__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_176__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_177__weights (optional) - T2: weights to optimize.
__group_177__gradients (optional) - T3: gradients computed in this iteration.
__group_177__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_177__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_177__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_178__weights (optional) - T2: weights to optimize.
__group_178__gradients (optional) - T3: gradients computed in this iteration.
__group_178__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_178__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_178__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_179__weights (optional) - T2: weights to optimize.
__group_179__gradients (optional) - T3: gradients computed in this iteration.
__group_179__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_179__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_179__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_180__weights (optional) - T2: weights to optimize.
__group_180__gradients (optional) - T3: gradients computed in this iteration.
__group_180__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_180__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_180__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_181__weights (optional) - T2: weights to optimize.
__group_181__gradients (optional) - T3: gradients computed in this iteration.
__group_181__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_181__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_181__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_182__weights (optional) - T2: weights to optimize.
__group_182__gradients (optional) - T3: gradients computed in this iteration.
__group_182__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_182__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_182__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_183__weights (optional) - T2: weights to optimize.
__group_183__gradients (optional) - T3: gradients computed in this iteration.
__group_183__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_183__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_183__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_184__weights (optional) - T2: weights to optimize.
__group_184__gradients (optional) - T3: gradients computed in this iteration.
__group_184__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_184__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_184__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_185__weights (optional) - T2: weights to optimize.
__group_185__gradients (optional) - T3: gradients computed in this iteration.
__group_185__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_185__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_185__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_186__weights (optional) - T2: weights to optimize.
__group_186__gradients (optional) - T3: gradients computed in this iteration.
__group_186__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_186__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_186__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_187__weights (optional) - T2: weights to optimize.
__group_187__gradients (optional) - T3: gradients computed in this iteration.
__group_187__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_187__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_187__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_188__weights (optional) - T2: weights to optimize.
__group_188__gradients (optional) - T3: gradients computed in this iteration.
__group_188__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_188__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_188__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_189__weights (optional) - T2: weights to optimize.
__group_189__gradients (optional) - T3: gradients computed in this iteration.
__group_189__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_189__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_189__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_190__weights (optional) - T2: weights to optimize.
__group_190__gradients (optional) - T3: gradients computed in this iteration.
__group_190__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_190__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_190__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_191__weights (optional) - T2: weights to optimize.
__group_191__gradients (optional) - T3: gradients computed in this iteration.
__group_191__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_191__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_191__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_192__weights (optional) - T2: weights to optimize.
__group_192__gradients (optional) - T3: gradients computed in this iteration.
__group_192__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_192__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_192__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_193__weights (optional) - T2: weights to optimize.
__group_193__gradients (optional) - T3: gradients computed in this iteration.
__group_193__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_193__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_193__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_194__weights (optional) - T2: weights to optimize.
__group_194__gradients (optional) - T3: gradients computed in this iteration.
__group_194__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_194__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_194__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_195__weights (optional) - T2: weights to optimize.
__group_195__gradients (optional) - T3: gradients computed in this iteration.
__group_195__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_195__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_195__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_196__weights (optional) - T2: weights to optimize.
__group_196__gradients (optional) - T3: gradients computed in this iteration.
__group_196__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_196__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_196__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_197__weights (optional) - T2: weights to optimize.
__group_197__gradients (optional) - T3: gradients computed in this iteration.
__group_197__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_197__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_197__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_198__weights (optional) - T2: weights to optimize.
__group_198__gradients (optional) - T3: gradients computed in this iteration.
__group_198__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_198__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_198__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_199__weights (optional) - T2: weights to optimize.
__group_199__gradients (optional) - T3: gradients computed in this iteration.
__group_199__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_199__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_199__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_200__weights (optional) - T2: weights to optimize.
__group_200__gradients (optional) - T3: gradients computed in this iteration.
__group_200__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_200__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_200__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_201__weights (optional) - T2: weights to optimize.
__group_201__gradients (optional) - T3: gradients computed in this iteration.
__group_201__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_201__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_201__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_202__weights (optional) - T2: weights to optimize.
__group_202__gradients (optional) - T3: gradients computed in this iteration.
__group_202__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_202__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_202__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_203__weights (optional) - T2: weights to optimize.
__group_203__gradients (optional) - T3: gradients computed in this iteration.
__group_203__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_203__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_203__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_204__weights (optional) - T2: weights to optimize.
__group_204__gradients (optional) - T3: gradients computed in this iteration.
__group_204__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_204__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_204__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_205__weights (optional) - T2: weights to optimize.
__group_205__gradients (optional) - T3: gradients computed in this iteration.
__group_205__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_205__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_205__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_206__weights (optional) - T2: weights to optimize.
__group_206__gradients (optional) - T3: gradients computed in this iteration.
__group_206__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_206__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_206__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_207__weights (optional) - T2: weights to optimize.
__group_207__gradients (optional) - T3: gradients computed in this iteration.
__group_207__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_207__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_207__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_208__weights (optional) - T2: weights to optimize.
__group_208__gradients (optional) - T3: gradients computed in this iteration.
__group_208__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_208__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_208__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_209__weights (optional) - T2: weights to optimize.
__group_209__gradients (optional) - T3: gradients computed in this iteration.
__group_209__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_209__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_209__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_210__weights (optional) - T2: weights to optimize.
__group_210__gradients (optional) - T3: gradients computed in this iteration.
__group_210__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_210__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_210__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_211__weights (optional) - T2: weights to optimize.
__group_211__gradients (optional) - T3: gradients computed in this iteration.
__group_211__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_211__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_211__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_212__weights (optional) - T2: weights to optimize.
__group_212__gradients (optional) - T3: gradients computed in this iteration.
__group_212__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_212__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_212__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_213__weights (optional) - T2: weights to optimize.
__group_213__gradients (optional) - T3: gradients computed in this iteration.
__group_213__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_213__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_213__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_214__weights (optional) - T2: weights to optimize.
__group_214__gradients (optional) - T3: gradients computed in this iteration.
__group_214__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_214__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_214__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_215__weights (optional) - T2: weights to optimize.
__group_215__gradients (optional) - T3: gradients computed in this iteration.
__group_215__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_215__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_215__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_216__weights (optional) - T2: weights to optimize.
__group_216__gradients (optional) - T3: gradients computed in this iteration.
__group_216__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_216__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_216__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_217__weights (optional) - T2: weights to optimize.
__group_217__gradients (optional) - T3: gradients computed in this iteration.
__group_217__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_217__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_217__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_218__weights (optional) - T2: weights to optimize.
__group_218__gradients (optional) - T3: gradients computed in this iteration.
__group_218__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_218__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_218__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_219__weights (optional) - T2: weights to optimize.
__group_219__gradients (optional) - T3: gradients computed in this iteration.
__group_219__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_219__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_219__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_220__weights (optional) - T2: weights to optimize.
__group_220__gradients (optional) - T3: gradients computed in this iteration.
__group_220__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_220__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_220__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_221__weights (optional) - T2: weights to optimize.
__group_221__gradients (optional) - T3: gradients computed in this iteration.
__group_221__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_221__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_221__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_222__weights (optional) - T2: weights to optimize.
__group_222__gradients (optional) - T3: gradients computed in this iteration.
__group_222__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_222__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_222__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_223__weights (optional) - T2: weights to optimize.
__group_223__gradients (optional) - T3: gradients computed in this iteration.
__group_223__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_223__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_223__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_224__weights (optional) - T2: weights to optimize.
__group_224__gradients (optional) - T3: gradients computed in this iteration.
__group_224__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_224__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_224__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_225__weights (optional) - T2: weights to optimize.
__group_225__gradients (optional) - T3: gradients computed in this iteration.
__group_225__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_225__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_225__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_226__weights (optional) - T2: weights to optimize.
__group_226__gradients (optional) - T3: gradients computed in this iteration.
__group_226__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_226__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_226__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_227__weights (optional) - T2: weights to optimize.
__group_227__gradients (optional) - T3: gradients computed in this iteration.
__group_227__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_227__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_227__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_228__weights (optional) - T2: weights to optimize.
__group_228__gradients (optional) - T3: gradients computed in this iteration.
__group_228__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_228__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_228__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_229__weights (optional) - T2: weights to optimize.
__group_229__gradients (optional) - T3: gradients computed in this iteration.
__group_229__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_229__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_229__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_230__weights (optional) - T2: weights to optimize.
__group_230__gradients (optional) - T3: gradients computed in this iteration.
__group_230__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_230__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_230__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_231__weights (optional) - T2: weights to optimize.
__group_231__gradients (optional) - T3: gradients computed in this iteration.
__group_231__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_231__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_231__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_232__weights (optional) - T2: weights to optimize.
__group_232__gradients (optional) - T3: gradients computed in this iteration.
__group_232__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_232__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_232__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_233__weights (optional) - T2: weights to optimize.
__group_233__gradients (optional) - T3: gradients computed in this iteration.
__group_233__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_233__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_233__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_234__weights (optional) - T2: weights to optimize.
__group_234__gradients (optional) - T3: gradients computed in this iteration.
__group_234__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_234__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_234__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_235__weights (optional) - T2: weights to optimize.
__group_235__gradients (optional) - T3: gradients computed in this iteration.
__group_235__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_235__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_235__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_236__weights (optional) - T2: weights to optimize.
__group_236__gradients (optional) - T3: gradients computed in this iteration.
__group_236__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_236__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_236__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_237__weights (optional) - T2: weights to optimize.
__group_237__gradients (optional) - T3: gradients computed in this iteration.
__group_237__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_237__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_237__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_238__weights (optional) - T2: weights to optimize.
__group_238__gradients (optional) - T3: gradients computed in this iteration.
__group_238__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_238__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_238__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_239__weights (optional) - T2: weights to optimize.
__group_239__gradients (optional) - T3: gradients computed in this iteration.
__group_239__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_239__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_239__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_240__weights (optional) - T2: weights to optimize.
__group_240__gradients (optional) - T3: gradients computed in this iteration.
__group_240__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_240__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_240__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_241__weights (optional) - T2: weights to optimize.
__group_241__gradients (optional) - T3: gradients computed in this iteration.
__group_241__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_241__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_241__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_242__weights (optional) - T2: weights to optimize.
__group_242__gradients (optional) - T3: gradients computed in this iteration.
__group_242__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_242__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_242__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_243__weights (optional) - T2: weights to optimize.
__group_243__gradients (optional) - T3: gradients computed in this iteration.
__group_243__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_243__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_243__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_244__weights (optional) - T2: weights to optimize.
__group_244__gradients (optional) - T3: gradients computed in this iteration.
__group_244__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_244__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_244__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_245__weights (optional) - T2: weights to optimize.
__group_245__gradients (optional) - T3: gradients computed in this iteration.
__group_245__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_245__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_245__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_246__weights (optional) - T2: weights to optimize.
__group_246__gradients (optional) - T3: gradients computed in this iteration.
__group_246__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_246__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_246__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_247__weights (optional) - T2: weights to optimize.
__group_247__gradients (optional) - T3: gradients computed in this iteration.
__group_247__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_247__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_247__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_248__weights (optional) - T2: weights to optimize.
__group_248__gradients (optional) - T3: gradients computed in this iteration.
__group_248__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_248__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_248__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_249__weights (optional) - T2: weights to optimize.
__group_249__gradients (optional) - T3: gradients computed in this iteration.
__group_249__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_249__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_249__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_250__weights (optional) - T2: weights to optimize.
__group_250__gradients (optional) - T3: gradients computed in this iteration.
__group_250__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_250__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_250__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_251__weights (optional) - T2: weights to optimize.
__group_251__gradients (optional) - T3: gradients computed in this iteration.
__group_251__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_251__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_251__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_252__weights (optional) - T2: weights to optimize.
__group_252__gradients (optional) - T3: gradients computed in this iteration.
__group_252__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_252__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_252__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_253__weights (optional) - T2: weights to optimize.
__group_253__gradients (optional) - T3: gradients computed in this iteration.
__group_253__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_253__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_253__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_254__weights (optional) - T2: weights to optimize.
__group_254__gradients (optional) - T3: gradients computed in this iteration.
__group_254__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_254__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_254__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_255__weights (optional) - T2: weights to optimize.
__group_255__gradients (optional) - T3: gradients computed in this iteration.
__group_255__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_255__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_255__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_256__weights (optional) - T2: weights to optimize.
__group_256__gradients (optional) - T3: gradients computed in this iteration.
__group_256__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_256__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_256__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_257__weights (optional) - T2: weights to optimize.
__group_257__gradients (optional) - T3: gradients computed in this iteration.
__group_257__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_257__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_257__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_258__weights (optional) - T2: weights to optimize.
__group_258__gradients (optional) - T3: gradients computed in this iteration.
__group_258__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_258__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_258__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_259__weights (optional) - T2: weights to optimize.
__group_259__gradients (optional) - T3: gradients computed in this iteration.
__group_259__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_259__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_259__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_260__weights (optional) - T2: weights to optimize.
__group_260__gradients (optional) - T3: gradients computed in this iteration.
__group_260__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_260__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_260__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_261__weights (optional) - T2: weights to optimize.
__group_261__gradients (optional) - T3: gradients computed in this iteration.
__group_261__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_261__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_261__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_262__weights (optional) - T2: weights to optimize.
__group_262__gradients (optional) - T3: gradients computed in this iteration.
__group_262__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_262__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_262__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_263__weights (optional) - T2: weights to optimize.
__group_263__gradients (optional) - T3: gradients computed in this iteration.
__group_263__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_263__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_263__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_264__weights (optional) - T2: weights to optimize.
__group_264__gradients (optional) - T3: gradients computed in this iteration.
__group_264__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_264__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_264__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_265__weights (optional) - T2: weights to optimize.
__group_265__gradients (optional) - T3: gradients computed in this iteration.
__group_265__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_265__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_265__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_266__weights (optional) - T2: weights to optimize.
__group_266__gradients (optional) - T3: gradients computed in this iteration.
__group_266__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_266__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_266__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_267__weights (optional) - T2: weights to optimize.
__group_267__gradients (optional) - T3: gradients computed in this iteration.
__group_267__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_267__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_267__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_268__weights (optional) - T2: weights to optimize.
__group_268__gradients (optional) - T3: gradients computed in this iteration.
__group_268__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_268__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_268__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_269__weights (optional) - T2: weights to optimize.
__group_269__gradients (optional) - T3: gradients computed in this iteration.
__group_269__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_269__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_269__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_270__weights (optional) - T2: weights to optimize.
__group_270__gradients (optional) - T3: gradients computed in this iteration.
__group_270__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_270__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_270__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_271__weights (optional) - T2: weights to optimize.
__group_271__gradients (optional) - T3: gradients computed in this iteration.
__group_271__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_271__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_271__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_272__weights (optional) - T2: weights to optimize.
__group_272__gradients (optional) - T3: gradients computed in this iteration.
__group_272__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_272__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_272__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_273__weights (optional) - T2: weights to optimize.
__group_273__gradients (optional) - T3: gradients computed in this iteration.
__group_273__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_273__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_273__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_274__weights (optional) - T2: weights to optimize.
__group_274__gradients (optional) - T3: gradients computed in this iteration.
__group_274__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_274__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_274__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_275__weights (optional) - T2: weights to optimize.
__group_275__gradients (optional) - T3: gradients computed in this iteration.
__group_275__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_275__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_275__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_276__weights (optional) - T2: weights to optimize.
__group_276__gradients (optional) - T3: gradients computed in this iteration.
__group_276__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_276__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_276__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_277__weights (optional) - T2: weights to optimize.
__group_277__gradients (optional) - T3: gradients computed in this iteration.
__group_277__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_277__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_277__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_278__weights (optional) - T2: weights to optimize.
__group_278__gradients (optional) - T3: gradients computed in this iteration.
__group_278__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_278__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_278__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_279__weights (optional) - T2: weights to optimize.
__group_279__gradients (optional) - T3: gradients computed in this iteration.
__group_279__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_279__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_279__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_280__weights (optional) - T2: weights to optimize.
__group_280__gradients (optional) - T3: gradients computed in this iteration.
__group_280__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_280__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_280__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_281__weights (optional) - T2: weights to optimize.
__group_281__gradients (optional) - T3: gradients computed in this iteration.
__group_281__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_281__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_281__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_282__weights (optional) - T2: weights to optimize.
__group_282__gradients (optional) - T3: gradients computed in this iteration.
__group_282__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_282__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_282__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_283__weights (optional) - T2: weights to optimize.
__group_283__gradients (optional) - T3: gradients computed in this iteration.
__group_283__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_283__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_283__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_284__weights (optional) - T2: weights to optimize.
__group_284__gradients (optional) - T3: gradients computed in this iteration.
__group_284__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_284__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_284__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_285__weights (optional) - T2: weights to optimize.
__group_285__gradients (optional) - T3: gradients computed in this iteration.
__group_285__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_285__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_285__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_286__weights (optional) - T2: weights to optimize.
__group_286__gradients (optional) - T3: gradients computed in this iteration.
__group_286__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_286__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_286__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_287__weights (optional) - T2: weights to optimize.
__group_287__gradients (optional) - T3: gradients computed in this iteration.
__group_287__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_287__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_287__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_288__weights (optional) - T2: weights to optimize.
__group_288__gradients (optional) - T3: gradients computed in this iteration.
__group_288__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_288__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_288__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_289__weights (optional) - T2: weights to optimize.
__group_289__gradients (optional) - T3: gradients computed in this iteration.
__group_289__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_289__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_289__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_290__weights (optional) - T2: weights to optimize.
__group_290__gradients (optional) - T3: gradients computed in this iteration.
__group_290__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_290__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_290__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_291__weights (optional) - T2: weights to optimize.
__group_291__gradients (optional) - T3: gradients computed in this iteration.
__group_291__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_291__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_291__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_292__weights (optional) - T2: weights to optimize.
__group_292__gradients (optional) - T3: gradients computed in this iteration.
__group_292__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_292__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_292__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_293__weights (optional) - T2: weights to optimize.
__group_293__gradients (optional) - T3: gradients computed in this iteration.
__group_293__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_293__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_293__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_294__weights (optional) - T2: weights to optimize.
__group_294__gradients (optional) - T3: gradients computed in this iteration.
__group_294__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_294__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_294__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_295__weights (optional) - T2: weights to optimize.
__group_295__gradients (optional) - T3: gradients computed in this iteration.
__group_295__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_295__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_295__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_296__weights (optional) - T2: weights to optimize.
__group_296__gradients (optional) - T3: gradients computed in this iteration.
__group_296__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_296__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_296__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_297__weights (optional) - T2: weights to optimize.
__group_297__gradients (optional) - T3: gradients computed in this iteration.
__group_297__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_297__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_297__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_298__weights (optional) - T2: weights to optimize.
__group_298__gradients (optional) - T3: gradients computed in this iteration.
__group_298__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_298__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_298__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_299__weights (optional) - T2: weights to optimize.
__group_299__gradients (optional) - T3: gradients computed in this iteration.
__group_299__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_299__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_299__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_300__weights (optional) - T2: weights to optimize.
__group_300__gradients (optional) - T3: gradients computed in this iteration.
__group_300__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_300__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_300__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_301__weights (optional) - T2: weights to optimize.
__group_301__gradients (optional) - T3: gradients computed in this iteration.
__group_301__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_301__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_301__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_302__weights (optional) - T2: weights to optimize.
__group_302__gradients (optional) - T3: gradients computed in this iteration.
__group_302__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_302__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_302__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_303__weights (optional) - T2: weights to optimize.
__group_303__gradients (optional) - T3: gradients computed in this iteration.
__group_303__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_303__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_303__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_304__weights (optional) - T2: weights to optimize.
__group_304__gradients (optional) - T3: gradients computed in this iteration.
__group_304__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_304__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_304__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_305__weights (optional) - T2: weights to optimize.
__group_305__gradients (optional) - T3: gradients computed in this iteration.
__group_305__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_305__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_305__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_306__weights (optional) - T2: weights to optimize.
__group_306__gradients (optional) - T3: gradients computed in this iteration.
__group_306__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_306__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_306__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_307__weights (optional) - T2: weights to optimize.
__group_307__gradients (optional) - T3: gradients computed in this iteration.
__group_307__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_307__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_307__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_308__weights (optional) - T2: weights to optimize.
__group_308__gradients (optional) - T3: gradients computed in this iteration.
__group_308__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_308__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_308__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_309__weights (optional) - T2: weights to optimize.
__group_309__gradients (optional) - T3: gradients computed in this iteration.
__group_309__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_309__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_309__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_310__weights (optional) - T2: weights to optimize.
__group_310__gradients (optional) - T3: gradients computed in this iteration.
__group_310__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_310__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_310__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_311__weights (optional) - T2: weights to optimize.
__group_311__gradients (optional) - T3: gradients computed in this iteration.
__group_311__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_311__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_311__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_312__weights (optional) - T2: weights to optimize.
__group_312__gradients (optional) - T3: gradients computed in this iteration.
__group_312__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_312__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_312__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_313__weights (optional) - T2: weights to optimize.
__group_313__gradients (optional) - T3: gradients computed in this iteration.
__group_313__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_313__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_313__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_314__weights (optional) - T2: weights to optimize.
__group_314__gradients (optional) - T3: gradients computed in this iteration.
__group_314__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_314__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_314__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_315__weights (optional) - T2: weights to optimize.
__group_315__gradients (optional) - T3: gradients computed in this iteration.
__group_315__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_315__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_315__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_316__weights (optional) - T2: weights to optimize.
__group_316__gradients (optional) - T3: gradients computed in this iteration.
__group_316__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_316__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_316__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_317__weights (optional) - T2: weights to optimize.
__group_317__gradients (optional) - T3: gradients computed in this iteration.
__group_317__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_317__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_317__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_318__weights (optional) - T2: weights to optimize.
__group_318__gradients (optional) - T3: gradients computed in this iteration.
__group_318__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_318__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_318__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_319__weights (optional) - T2: weights to optimize.
__group_319__gradients (optional) - T3: gradients computed in this iteration.
__group_319__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_319__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_319__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_320__weights (optional) - T2: weights to optimize.
__group_320__gradients (optional) - T3: gradients computed in this iteration.
__group_320__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_320__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_320__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_321__weights (optional) - T2: weights to optimize.
__group_321__gradients (optional) - T3: gradients computed in this iteration.
__group_321__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_321__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_321__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_322__weights (optional) - T2: weights to optimize.
__group_322__gradients (optional) - T3: gradients computed in this iteration.
__group_322__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_322__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_322__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_323__weights (optional) - T2: weights to optimize.
__group_323__gradients (optional) - T3: gradients computed in this iteration.
__group_323__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_323__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_323__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_324__weights (optional) - T2: weights to optimize.
__group_324__gradients (optional) - T3: gradients computed in this iteration.
__group_324__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_324__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_324__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_325__weights (optional) - T2: weights to optimize.
__group_325__gradients (optional) - T3: gradients computed in this iteration.
__group_325__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_325__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_325__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_326__weights (optional) - T2: weights to optimize.
__group_326__gradients (optional) - T3: gradients computed in this iteration.
__group_326__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_326__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_326__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_327__weights (optional) - T2: weights to optimize.
__group_327__gradients (optional) - T3: gradients computed in this iteration.
__group_327__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_327__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_327__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_328__weights (optional) - T2: weights to optimize.
__group_328__gradients (optional) - T3: gradients computed in this iteration.
__group_328__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_328__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_328__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_329__weights (optional) - T2: weights to optimize.
__group_329__gradients (optional) - T3: gradients computed in this iteration.
__group_329__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_329__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_329__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_330__weights (optional) - T2: weights to optimize.
__group_330__gradients (optional) - T3: gradients computed in this iteration.
__group_330__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_330__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_330__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_331__weights (optional) - T2: weights to optimize.
__group_331__gradients (optional) - T3: gradients computed in this iteration.
__group_331__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_331__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_331__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_332__weights (optional) - T2: weights to optimize.
__group_332__gradients (optional) - T3: gradients computed in this iteration.
__group_332__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_332__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_332__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_333__weights (optional) - T2: weights to optimize.
__group_333__gradients (optional) - T3: gradients computed in this iteration.
__group_333__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_333__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_333__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_334__weights (optional) - T2: weights to optimize.
__group_334__gradients (optional) - T3: gradients computed in this iteration.
__group_334__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_334__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_334__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_335__weights (optional) - T2: weights to optimize.
__group_335__gradients (optional) - T3: gradients computed in this iteration.
__group_335__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_335__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_335__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_336__weights (optional) - T2: weights to optimize.
__group_336__gradients (optional) - T3: gradients computed in this iteration.
__group_336__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_336__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_336__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_337__weights (optional) - T2: weights to optimize.
__group_337__gradients (optional) - T3: gradients computed in this iteration.
__group_337__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_337__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_337__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_338__weights (optional) - T2: weights to optimize.
__group_338__gradients (optional) - T3: gradients computed in this iteration.
__group_338__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_338__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_338__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_339__weights (optional) - T2: weights to optimize.
__group_339__gradients (optional) - T3: gradients computed in this iteration.
__group_339__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_339__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_339__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_340__weights (optional) - T2: weights to optimize.
__group_340__gradients (optional) - T3: gradients computed in this iteration.
__group_340__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_340__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_340__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_341__weights (optional) - T2: weights to optimize.
__group_341__gradients (optional) - T3: gradients computed in this iteration.
__group_341__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_341__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_341__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_342__weights (optional) - T2: weights to optimize.
__group_342__gradients (optional) - T3: gradients computed in this iteration.
__group_342__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_342__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_342__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_343__weights (optional) - T2: weights to optimize.
__group_343__gradients (optional) - T3: gradients computed in this iteration.
__group_343__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_343__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_343__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_344__weights (optional) - T2: weights to optimize.
__group_344__gradients (optional) - T3: gradients computed in this iteration.
__group_344__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_344__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_344__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_345__weights (optional) - T2: weights to optimize.
__group_345__gradients (optional) - T3: gradients computed in this iteration.
__group_345__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_345__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_345__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_346__weights (optional) - T2: weights to optimize.
__group_346__gradients (optional) - T3: gradients computed in this iteration.
__group_346__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_346__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_346__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_347__weights (optional) - T2: weights to optimize.
__group_347__gradients (optional) - T3: gradients computed in this iteration.
__group_347__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_347__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_347__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_348__weights (optional) - T2: weights to optimize.
__group_348__gradients (optional) - T3: gradients computed in this iteration.
__group_348__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_348__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_348__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_349__weights (optional) - T2: weights to optimize.
__group_349__gradients (optional) - T3: gradients computed in this iteration.
__group_349__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_349__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_349__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_350__weights (optional) - T2: weights to optimize.
__group_350__gradients (optional) - T3: gradients computed in this iteration.
__group_350__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_350__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_350__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_351__weights (optional) - T2: weights to optimize.
__group_351__gradients (optional) - T3: gradients computed in this iteration.
__group_351__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_351__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_351__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_352__weights (optional) - T2: weights to optimize.
__group_352__gradients (optional) - T3: gradients computed in this iteration.
__group_352__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_352__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_352__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_353__weights (optional) - T2: weights to optimize.
__group_353__gradients (optional) - T3: gradients computed in this iteration.
__group_353__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_353__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_353__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_354__weights (optional) - T2: weights to optimize.
__group_354__gradients (optional) - T3: gradients computed in this iteration.
__group_354__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_354__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_354__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_355__weights (optional) - T2: weights to optimize.
__group_355__gradients (optional) - T3: gradients computed in this iteration.
__group_355__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_355__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_355__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_356__weights (optional) - T2: weights to optimize.
__group_356__gradients (optional) - T3: gradients computed in this iteration.
__group_356__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_356__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_356__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_357__weights (optional) - T2: weights to optimize.
__group_357__gradients (optional) - T3: gradients computed in this iteration.
__group_357__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_357__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_357__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_358__weights (optional) - T2: weights to optimize.
__group_358__gradients (optional) - T3: gradients computed in this iteration.
__group_358__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_358__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_358__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_359__weights (optional) - T2: weights to optimize.
__group_359__gradients (optional) - T3: gradients computed in this iteration.
__group_359__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_359__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_359__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_360__weights (optional) - T2: weights to optimize.
__group_360__gradients (optional) - T3: gradients computed in this iteration.
__group_360__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_360__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_360__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_361__weights (optional) - T2: weights to optimize.
__group_361__gradients (optional) - T3: gradients computed in this iteration.
__group_361__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_361__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_361__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_362__weights (optional) - T2: weights to optimize.
__group_362__gradients (optional) - T3: gradients computed in this iteration.
__group_362__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_362__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_362__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_363__weights (optional) - T2: weights to optimize.
__group_363__gradients (optional) - T3: gradients computed in this iteration.
__group_363__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_363__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_363__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_364__weights (optional) - T2: weights to optimize.
__group_364__gradients (optional) - T3: gradients computed in this iteration.
__group_364__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_364__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_364__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_365__weights (optional) - T2: weights to optimize.
__group_365__gradients (optional) - T3: gradients computed in this iteration.
__group_365__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_365__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_365__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_366__weights (optional) - T2: weights to optimize.
__group_366__gradients (optional) - T3: gradients computed in this iteration.
__group_366__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_366__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_366__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_367__weights (optional) - T2: weights to optimize.
__group_367__gradients (optional) - T3: gradients computed in this iteration.
__group_367__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_367__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_367__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_368__weights (optional) - T2: weights to optimize.
__group_368__gradients (optional) - T3: gradients computed in this iteration.
__group_368__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_368__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_368__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_369__weights (optional) - T2: weights to optimize.
__group_369__gradients (optional) - T3: gradients computed in this iteration.
__group_369__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_369__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_369__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_370__weights (optional) - T2: weights to optimize.
__group_370__gradients (optional) - T3: gradients computed in this iteration.
__group_370__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_370__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_370__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_371__weights (optional) - T2: weights to optimize.
__group_371__gradients (optional) - T3: gradients computed in this iteration.
__group_371__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_371__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_371__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_372__weights (optional) - T2: weights to optimize.
__group_372__gradients (optional) - T3: gradients computed in this iteration.
__group_372__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_372__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_372__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_373__weights (optional) - T2: weights to optimize.
__group_373__gradients (optional) - T3: gradients computed in this iteration.
__group_373__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_373__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_373__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_374__weights (optional) - T2: weights to optimize.
__group_374__gradients (optional) - T3: gradients computed in this iteration.
__group_374__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_374__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_374__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_375__weights (optional) - T2: weights to optimize.
__group_375__gradients (optional) - T3: gradients computed in this iteration.
__group_375__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_375__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_375__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_376__weights (optional) - T2: weights to optimize.
__group_376__gradients (optional) - T3: gradients computed in this iteration.
__group_376__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_376__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_376__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_377__weights (optional) - T2: weights to optimize.
__group_377__gradients (optional) - T3: gradients computed in this iteration.
__group_377__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_377__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_377__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_378__weights (optional) - T2: weights to optimize.
__group_378__gradients (optional) - T3: gradients computed in this iteration.
__group_378__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_378__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_378__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_379__weights (optional) - T2: weights to optimize.
__group_379__gradients (optional) - T3: gradients computed in this iteration.
__group_379__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_379__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_379__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_380__weights (optional) - T2: weights to optimize.
__group_380__gradients (optional) - T3: gradients computed in this iteration.
__group_380__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_380__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_380__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_381__weights (optional) - T2: weights to optimize.
__group_381__gradients (optional) - T3: gradients computed in this iteration.
__group_381__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_381__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_381__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_382__weights (optional) - T2: weights to optimize.
__group_382__gradients (optional) - T3: gradients computed in this iteration.
__group_382__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_382__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_382__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_383__weights (optional) - T2: weights to optimize.
__group_383__gradients (optional) - T3: gradients computed in this iteration.
__group_383__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_383__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_383__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_384__weights (optional) - T2: weights to optimize.
__group_384__gradients (optional) - T3: gradients computed in this iteration.
__group_384__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_384__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_384__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_385__weights (optional) - T2: weights to optimize.
__group_385__gradients (optional) - T3: gradients computed in this iteration.
__group_385__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_385__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_385__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_386__weights (optional) - T2: weights to optimize.
__group_386__gradients (optional) - T3: gradients computed in this iteration.
__group_386__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_386__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_386__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_387__weights (optional) - T2: weights to optimize.
__group_387__gradients (optional) - T3: gradients computed in this iteration.
__group_387__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_387__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_387__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_388__weights (optional) - T2: weights to optimize.
__group_388__gradients (optional) - T3: gradients computed in this iteration.
__group_388__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_388__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_388__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_389__weights (optional) - T2: weights to optimize.
__group_389__gradients (optional) - T3: gradients computed in this iteration.
__group_389__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_389__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_389__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_390__weights (optional) - T2: weights to optimize.
__group_390__gradients (optional) - T3: gradients computed in this iteration.
__group_390__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_390__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_390__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_391__weights (optional) - T2: weights to optimize.
__group_391__gradients (optional) - T3: gradients computed in this iteration.
__group_391__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_391__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_391__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_392__weights (optional) - T2: weights to optimize.
__group_392__gradients (optional) - T3: gradients computed in this iteration.
__group_392__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_392__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_392__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_393__weights (optional) - T2: weights to optimize.
__group_393__gradients (optional) - T3: gradients computed in this iteration.
__group_393__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_393__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_393__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_394__weights (optional) - T2: weights to optimize.
__group_394__gradients (optional) - T3: gradients computed in this iteration.
__group_394__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_394__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_394__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_395__weights (optional) - T2: weights to optimize.
__group_395__gradients (optional) - T3: gradients computed in this iteration.
__group_395__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_395__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_395__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_396__weights (optional) - T2: weights to optimize.
__group_396__gradients (optional) - T3: gradients computed in this iteration.
__group_396__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_396__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_396__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_397__weights (optional) - T2: weights to optimize.
__group_397__gradients (optional) - T3: gradients computed in this iteration.
__group_397__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_397__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_397__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_398__weights (optional) - T2: weights to optimize.
__group_398__gradients (optional) - T3: gradients computed in this iteration.
__group_398__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_398__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_398__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_399__weights (optional) - T2: weights to optimize.
__group_399__gradients (optional) - T3: gradients computed in this iteration.
__group_399__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_399__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_399__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_400__weights (optional) - T2: weights to optimize.
__group_400__gradients (optional) - T3: gradients computed in this iteration.
__group_400__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_400__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_400__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_401__weights (optional) - T2: weights to optimize.
__group_401__gradients (optional) - T3: gradients computed in this iteration.
__group_401__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_401__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_401__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_402__weights (optional) - T2: weights to optimize.
__group_402__gradients (optional) - T3: gradients computed in this iteration.
__group_402__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_402__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_402__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_403__weights (optional) - T2: weights to optimize.
__group_403__gradients (optional) - T3: gradients computed in this iteration.
__group_403__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_403__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_403__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_404__weights (optional) - T2: weights to optimize.
__group_404__gradients (optional) - T3: gradients computed in this iteration.
__group_404__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_404__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_404__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_405__weights (optional) - T2: weights to optimize.
__group_405__gradients (optional) - T3: gradients computed in this iteration.
__group_405__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_405__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_405__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_406__weights (optional) - T2: weights to optimize.
__group_406__gradients (optional) - T3: gradients computed in this iteration.
__group_406__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_406__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_406__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_407__weights (optional) - T2: weights to optimize.
__group_407__gradients (optional) - T3: gradients computed in this iteration.
__group_407__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_407__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_407__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_408__weights (optional) - T2: weights to optimize.
__group_408__gradients (optional) - T3: gradients computed in this iteration.
__group_408__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_408__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_408__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_409__weights (optional) - T2: weights to optimize.
__group_409__gradients (optional) - T3: gradients computed in this iteration.
__group_409__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_409__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_409__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_410__weights (optional) - T2: weights to optimize.
__group_410__gradients (optional) - T3: gradients computed in this iteration.
__group_410__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_410__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_410__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_411__weights (optional) - T2: weights to optimize.
__group_411__gradients (optional) - T3: gradients computed in this iteration.
__group_411__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_411__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_411__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_412__weights (optional) - T2: weights to optimize.
__group_412__gradients (optional) - T3: gradients computed in this iteration.
__group_412__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_412__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_412__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_413__weights (optional) - T2: weights to optimize.
__group_413__gradients (optional) - T3: gradients computed in this iteration.
__group_413__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_413__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_413__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_414__weights (optional) - T2: weights to optimize.
__group_414__gradients (optional) - T3: gradients computed in this iteration.
__group_414__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_414__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_414__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_415__weights (optional) - T2: weights to optimize.
__group_415__gradients (optional) - T3: gradients computed in this iteration.
__group_415__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_415__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_415__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_416__weights (optional) - T2: weights to optimize.
__group_416__gradients (optional) - T3: gradients computed in this iteration.
__group_416__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_416__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_416__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_417__weights (optional) - T2: weights to optimize.
__group_417__gradients (optional) - T3: gradients computed in this iteration.
__group_417__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_417__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_417__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_418__weights (optional) - T2: weights to optimize.
__group_418__gradients (optional) - T3: gradients computed in this iteration.
__group_418__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_418__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_418__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_419__weights (optional) - T2: weights to optimize.
__group_419__gradients (optional) - T3: gradients computed in this iteration.
__group_419__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_419__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_419__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_420__weights (optional) - T2: weights to optimize.
__group_420__gradients (optional) - T3: gradients computed in this iteration.
__group_420__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_420__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_420__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_421__weights (optional) - T2: weights to optimize.
__group_421__gradients (optional) - T3: gradients computed in this iteration.
__group_421__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_421__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_421__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_422__weights (optional) - T2: weights to optimize.
__group_422__gradients (optional) - T3: gradients computed in this iteration.
__group_422__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_422__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_422__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_423__weights (optional) - T2: weights to optimize.
__group_423__gradients (optional) - T3: gradients computed in this iteration.
__group_423__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_423__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_423__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_424__weights (optional) - T2: weights to optimize.
__group_424__gradients (optional) - T3: gradients computed in this iteration.
__group_424__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_424__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_424__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_425__weights (optional) - T2: weights to optimize.
__group_425__gradients (optional) - T3: gradients computed in this iteration.
__group_425__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_425__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_425__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_426__weights (optional) - T2: weights to optimize.
__group_426__gradients (optional) - T3: gradients computed in this iteration.
__group_426__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_426__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_426__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_427__weights (optional) - T2: weights to optimize.
__group_427__gradients (optional) - T3: gradients computed in this iteration.
__group_427__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_427__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_427__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_428__weights (optional) - T2: weights to optimize.
__group_428__gradients (optional) - T3: gradients computed in this iteration.
__group_428__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_428__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_428__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_429__weights (optional) - T2: weights to optimize.
__group_429__gradients (optional) - T3: gradients computed in this iteration.
__group_429__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_429__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_429__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_430__weights (optional) - T2: weights to optimize.
__group_430__gradients (optional) - T3: gradients computed in this iteration.
__group_430__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_430__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_430__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_431__weights (optional) - T2: weights to optimize.
__group_431__gradients (optional) - T3: gradients computed in this iteration.
__group_431__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_431__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_431__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_432__weights (optional) - T2: weights to optimize.
__group_432__gradients (optional) - T3: gradients computed in this iteration.
__group_432__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_432__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_432__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_433__weights (optional) - T2: weights to optimize.
__group_433__gradients (optional) - T3: gradients computed in this iteration.
__group_433__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_433__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_433__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_434__weights (optional) - T2: weights to optimize.
__group_434__gradients (optional) - T3: gradients computed in this iteration.
__group_434__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_434__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_434__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_435__weights (optional) - T2: weights to optimize.
__group_435__gradients (optional) - T3: gradients computed in this iteration.
__group_435__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_435__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_435__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_436__weights (optional) - T2: weights to optimize.
__group_436__gradients (optional) - T3: gradients computed in this iteration.
__group_436__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_436__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_436__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_437__weights (optional) - T2: weights to optimize.
__group_437__gradients (optional) - T3: gradients computed in this iteration.
__group_437__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_437__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_437__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_438__weights (optional) - T2: weights to optimize.
__group_438__gradients (optional) - T3: gradients computed in this iteration.
__group_438__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_438__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_438__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_439__weights (optional) - T2: weights to optimize.
__group_439__gradients (optional) - T3: gradients computed in this iteration.
__group_439__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_439__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_439__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_440__weights (optional) - T2: weights to optimize.
__group_440__gradients (optional) - T3: gradients computed in this iteration.
__group_440__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_440__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_440__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_441__weights (optional) - T2: weights to optimize.
__group_441__gradients (optional) - T3: gradients computed in this iteration.
__group_441__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_441__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_441__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_442__weights (optional) - T2: weights to optimize.
__group_442__gradients (optional) - T3: gradients computed in this iteration.
__group_442__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_442__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_442__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_443__weights (optional) - T2: weights to optimize.
__group_443__gradients (optional) - T3: gradients computed in this iteration.
__group_443__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_443__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_443__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_444__weights (optional) - T2: weights to optimize.
__group_444__gradients (optional) - T3: gradients computed in this iteration.
__group_444__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_444__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_444__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_445__weights (optional) - T2: weights to optimize.
__group_445__gradients (optional) - T3: gradients computed in this iteration.
__group_445__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_445__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_445__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_446__weights (optional) - T2: weights to optimize.
__group_446__gradients (optional) - T3: gradients computed in this iteration.
__group_446__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_446__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_446__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_447__weights (optional) - T2: weights to optimize.
__group_447__gradients (optional) - T3: gradients computed in this iteration.
__group_447__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_447__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_447__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_448__weights (optional) - T2: weights to optimize.
__group_448__gradients (optional) - T3: gradients computed in this iteration.
__group_448__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_448__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_448__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_449__weights (optional) - T2: weights to optimize.
__group_449__gradients (optional) - T3: gradients computed in this iteration.
__group_449__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_449__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_449__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_450__weights (optional) - T2: weights to optimize.
__group_450__gradients (optional) - T3: gradients computed in this iteration.
__group_450__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_450__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_450__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_451__weights (optional) - T2: weights to optimize.
__group_451__gradients (optional) - T3: gradients computed in this iteration.
__group_451__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_451__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_451__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_452__weights (optional) - T2: weights to optimize.
__group_452__gradients (optional) - T3: gradients computed in this iteration.
__group_452__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_452__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_452__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_453__weights (optional) - T2: weights to optimize.
__group_453__gradients (optional) - T3: gradients computed in this iteration.
__group_453__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_453__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_453__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_454__weights (optional) - T2: weights to optimize.
__group_454__gradients (optional) - T3: gradients computed in this iteration.
__group_454__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_454__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_454__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_455__weights (optional) - T2: weights to optimize.
__group_455__gradients (optional) - T3: gradients computed in this iteration.
__group_455__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_455__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_455__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_456__weights (optional) - T2: weights to optimize.
__group_456__gradients (optional) - T3: gradients computed in this iteration.
__group_456__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_456__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_456__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_457__weights (optional) - T2: weights to optimize.
__group_457__gradients (optional) - T3: gradients computed in this iteration.
__group_457__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_457__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_457__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_458__weights (optional) - T2: weights to optimize.
__group_458__gradients (optional) - T3: gradients computed in this iteration.
__group_458__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_458__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_458__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_459__weights (optional) - T2: weights to optimize.
__group_459__gradients (optional) - T3: gradients computed in this iteration.
__group_459__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_459__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_459__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_460__weights (optional) - T2: weights to optimize.
__group_460__gradients (optional) - T3: gradients computed in this iteration.
__group_460__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_460__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_460__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_461__weights (optional) - T2: weights to optimize.
__group_461__gradients (optional) - T3: gradients computed in this iteration.
__group_461__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_461__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_461__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_462__weights (optional) - T2: weights to optimize.
__group_462__gradients (optional) - T3: gradients computed in this iteration.
__group_462__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_462__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_462__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_463__weights (optional) - T2: weights to optimize.
__group_463__gradients (optional) - T3: gradients computed in this iteration.
__group_463__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_463__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_463__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_464__weights (optional) - T2: weights to optimize.
__group_464__gradients (optional) - T3: gradients computed in this iteration.
__group_464__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_464__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_464__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_465__weights (optional) - T2: weights to optimize.
__group_465__gradients (optional) - T3: gradients computed in this iteration.
__group_465__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_465__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_465__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_466__weights (optional) - T2: weights to optimize.
__group_466__gradients (optional) - T3: gradients computed in this iteration.
__group_466__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_466__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_466__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_467__weights (optional) - T2: weights to optimize.
__group_467__gradients (optional) - T3: gradients computed in this iteration.
__group_467__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_467__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_467__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_468__weights (optional) - T2: weights to optimize.
__group_468__gradients (optional) - T3: gradients computed in this iteration.
__group_468__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_468__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_468__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_469__weights (optional) - T2: weights to optimize.
__group_469__gradients (optional) - T3: gradients computed in this iteration.
__group_469__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_469__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_469__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_470__weights (optional) - T2: weights to optimize.
__group_470__gradients (optional) - T3: gradients computed in this iteration.
__group_470__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_470__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_470__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_471__weights (optional) - T2: weights to optimize.
__group_471__gradients (optional) - T3: gradients computed in this iteration.
__group_471__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_471__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_471__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_472__weights (optional) - T2: weights to optimize.
__group_472__gradients (optional) - T3: gradients computed in this iteration.
__group_472__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_472__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_472__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_473__weights (optional) - T2: weights to optimize.
__group_473__gradients (optional) - T3: gradients computed in this iteration.
__group_473__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_473__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_473__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_474__weights (optional) - T2: weights to optimize.
__group_474__gradients (optional) - T3: gradients computed in this iteration.
__group_474__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_474__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_474__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_475__weights (optional) - T2: weights to optimize.
__group_475__gradients (optional) - T3: gradients computed in this iteration.
__group_475__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_475__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_475__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_476__weights (optional) - T2: weights to optimize.
__group_476__gradients (optional) - T3: gradients computed in this iteration.
__group_476__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_476__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_476__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_477__weights (optional) - T2: weights to optimize.
__group_477__gradients (optional) - T3: gradients computed in this iteration.
__group_477__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_477__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_477__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_478__weights (optional) - T2: weights to optimize.
__group_478__gradients (optional) - T3: gradients computed in this iteration.
__group_478__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_478__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_478__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_479__weights (optional) - T2: weights to optimize.
__group_479__gradients (optional) - T3: gradients computed in this iteration.
__group_479__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_479__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_479__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_480__weights (optional) - T2: weights to optimize.
__group_480__gradients (optional) - T3: gradients computed in this iteration.
__group_480__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_480__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_480__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_481__weights (optional) - T2: weights to optimize.
__group_481__gradients (optional) - T3: gradients computed in this iteration.
__group_481__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_481__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_481__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_482__weights (optional) - T2: weights to optimize.
__group_482__gradients (optional) - T3: gradients computed in this iteration.
__group_482__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_482__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_482__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_483__weights (optional) - T2: weights to optimize.
__group_483__gradients (optional) - T3: gradients computed in this iteration.
__group_483__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_483__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_483__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_484__weights (optional) - T2: weights to optimize.
__group_484__gradients (optional) - T3: gradients computed in this iteration.
__group_484__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_484__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_484__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_485__weights (optional) - T2: weights to optimize.
__group_485__gradients (optional) - T3: gradients computed in this iteration.
__group_485__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_485__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_485__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_486__weights (optional) - T2: weights to optimize.
__group_486__gradients (optional) - T3: gradients computed in this iteration.
__group_486__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_486__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_486__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_487__weights (optional) - T2: weights to optimize.
__group_487__gradients (optional) - T3: gradients computed in this iteration.
__group_487__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_487__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_487__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_488__weights (optional) - T2: weights to optimize.
__group_488__gradients (optional) - T3: gradients computed in this iteration.
__group_488__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_488__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_488__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_489__weights (optional) - T2: weights to optimize.
__group_489__gradients (optional) - T3: gradients computed in this iteration.
__group_489__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_489__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_489__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_490__weights (optional) - T2: weights to optimize.
__group_490__gradients (optional) - T3: gradients computed in this iteration.
__group_490__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_490__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_490__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_491__weights (optional) - T2: weights to optimize.
__group_491__gradients (optional) - T3: gradients computed in this iteration.
__group_491__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_491__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_491__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_492__weights (optional) - T2: weights to optimize.
__group_492__gradients (optional) - T3: gradients computed in this iteration.
__group_492__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_492__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_492__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_493__weights (optional) - T2: weights to optimize.
__group_493__gradients (optional) - T3: gradients computed in this iteration.
__group_493__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_493__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_493__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_494__weights (optional) - T2: weights to optimize.
__group_494__gradients (optional) - T3: gradients computed in this iteration.
__group_494__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_494__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_494__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_495__weights (optional) - T2: weights to optimize.
__group_495__gradients (optional) - T3: gradients computed in this iteration.
__group_495__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_495__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_495__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_496__weights (optional) - T2: weights to optimize.
__group_496__gradients (optional) - T3: gradients computed in this iteration.
__group_496__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_496__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_496__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_497__weights (optional) - T2: weights to optimize.
__group_497__gradients (optional) - T3: gradients computed in this iteration.
__group_497__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_497__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_497__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_498__weights (optional) - T2: weights to optimize.
__group_498__gradients (optional) - T3: gradients computed in this iteration.
__group_498__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_498__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_498__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_499__weights (optional) - T2: weights to optimize.
__group_499__gradients (optional) - T3: gradients computed in this iteration.
__group_499__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_499__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_499__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_500__weights (optional) - T2: weights to optimize.
__group_500__gradients (optional) - T3: gradients computed in this iteration.
__group_500__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_500__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_500__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_501__weights (optional) - T2: weights to optimize.
__group_501__gradients (optional) - T3: gradients computed in this iteration.
__group_501__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_501__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_501__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_502__weights (optional) - T2: weights to optimize.
__group_502__gradients (optional) - T3: gradients computed in this iteration.
__group_502__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_502__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_502__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_503__weights (optional) - T2: weights to optimize.
__group_503__gradients (optional) - T3: gradients computed in this iteration.
__group_503__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_503__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_503__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_504__weights (optional) - T2: weights to optimize.
__group_504__gradients (optional) - T3: gradients computed in this iteration.
__group_504__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_504__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_504__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_505__weights (optional) - T2: weights to optimize.
__group_505__gradients (optional) - T3: gradients computed in this iteration.
__group_505__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_505__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_505__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_506__weights (optional) - T2: weights to optimize.
__group_506__gradients (optional) - T3: gradients computed in this iteration.
__group_506__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_506__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_506__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_507__weights (optional) - T2: weights to optimize.
__group_507__gradients (optional) - T3: gradients computed in this iteration.
__group_507__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_507__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_507__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_508__weights (optional) - T2: weights to optimize.
__group_508__gradients (optional) - T3: gradients computed in this iteration.
__group_508__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_508__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_508__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_509__weights (optional) - T2: weights to optimize.
__group_509__gradients (optional) - T3: gradients computed in this iteration.
__group_509__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_509__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_509__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_510__weights (optional) - T2: weights to optimize.
__group_510__gradients (optional) - T3: gradients computed in this iteration.
__group_510__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_510__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_510__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_511__weights (optional) - T2: weights to optimize.
__group_511__gradients (optional) - T3: gradients computed in this iteration.
__group_511__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_511__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_511__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_512__weights (optional) - T2: weights to optimize.
__group_512__gradients (optional) - T3: gradients computed in this iteration.
__group_512__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_512__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_512__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_513__weights (optional) - T2: weights to optimize.
__group_513__gradients (optional) - T3: gradients computed in this iteration.
__group_513__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_513__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_513__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_514__weights (optional) - T2: weights to optimize.
__group_514__gradients (optional) - T3: gradients computed in this iteration.
__group_514__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_514__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_514__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_515__weights (optional) - T2: weights to optimize.
__group_515__gradients (optional) - T3: gradients computed in this iteration.
__group_515__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_515__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_515__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_516__weights (optional) - T2: weights to optimize.
__group_516__gradients (optional) - T3: gradients computed in this iteration.
__group_516__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_516__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_516__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_517__weights (optional) - T2: weights to optimize.
__group_517__gradients (optional) - T3: gradients computed in this iteration.
__group_517__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_517__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_517__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_518__weights (optional) - T2: weights to optimize.
__group_518__gradients (optional) - T3: gradients computed in this iteration.
__group_518__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_518__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_518__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_519__weights (optional) - T2: weights to optimize.
__group_519__gradients (optional) - T3: gradients computed in this iteration.
__group_519__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_519__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_519__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_520__weights (optional) - T2: weights to optimize.
__group_520__gradients (optional) - T3: gradients computed in this iteration.
__group_520__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_520__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_520__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_521__weights (optional) - T2: weights to optimize.
__group_521__gradients (optional) - T3: gradients computed in this iteration.
__group_521__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_521__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_521__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_522__weights (optional) - T2: weights to optimize.
__group_522__gradients (optional) - T3: gradients computed in this iteration.
__group_522__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_522__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_522__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_523__weights (optional) - T2: weights to optimize.
__group_523__gradients (optional) - T3: gradients computed in this iteration.
__group_523__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_523__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_523__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_524__weights (optional) - T2: weights to optimize.
__group_524__gradients (optional) - T3: gradients computed in this iteration.
__group_524__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_524__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_524__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_525__weights (optional) - T2: weights to optimize.
__group_525__gradients (optional) - T3: gradients computed in this iteration.
__group_525__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_525__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_525__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_526__weights (optional) - T2: weights to optimize.
__group_526__gradients (optional) - T3: gradients computed in this iteration.
__group_526__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_526__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_526__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_527__weights (optional) - T2: weights to optimize.
__group_527__gradients (optional) - T3: gradients computed in this iteration.
__group_527__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_527__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_527__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_528__weights (optional) - T2: weights to optimize.
__group_528__gradients (optional) - T3: gradients computed in this iteration.
__group_528__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_528__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_528__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_529__weights (optional) - T2: weights to optimize.
__group_529__gradients (optional) - T3: gradients computed in this iteration.
__group_529__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_529__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_529__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_530__weights (optional) - T2: weights to optimize.
__group_530__gradients (optional) - T3: gradients computed in this iteration.
__group_530__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_530__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_530__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_531__weights (optional) - T2: weights to optimize.
__group_531__gradients (optional) - T3: gradients computed in this iteration.
__group_531__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_531__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_531__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_532__weights (optional) - T2: weights to optimize.
__group_532__gradients (optional) - T3: gradients computed in this iteration.
__group_532__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_532__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_532__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_533__weights (optional) - T2: weights to optimize.
__group_533__gradients (optional) - T3: gradients computed in this iteration.
__group_533__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_533__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_533__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_534__weights (optional) - T2: weights to optimize.
__group_534__gradients (optional) - T3: gradients computed in this iteration.
__group_534__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_534__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_534__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_535__weights (optional) - T2: weights to optimize.
__group_535__gradients (optional) - T3: gradients computed in this iteration.
__group_535__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_535__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_535__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_536__weights (optional) - T2: weights to optimize.
__group_536__gradients (optional) - T3: gradients computed in this iteration.
__group_536__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_536__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_536__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_537__weights (optional) - T2: weights to optimize.
__group_537__gradients (optional) - T3: gradients computed in this iteration.
__group_537__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_537__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_537__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_538__weights (optional) - T2: weights to optimize.
__group_538__gradients (optional) - T3: gradients computed in this iteration.
__group_538__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_538__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_538__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_539__weights (optional) - T2: weights to optimize.
__group_539__gradients (optional) - T3: gradients computed in this iteration.
__group_539__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_539__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_539__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_540__weights (optional) - T2: weights to optimize.
__group_540__gradients (optional) - T3: gradients computed in this iteration.
__group_540__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_540__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_540__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_541__weights (optional) - T2: weights to optimize.
__group_541__gradients (optional) - T3: gradients computed in this iteration.
__group_541__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_541__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_541__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_542__weights (optional) - T2: weights to optimize.
__group_542__gradients (optional) - T3: gradients computed in this iteration.
__group_542__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_542__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_542__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_543__weights (optional) - T2: weights to optimize.
__group_543__gradients (optional) - T3: gradients computed in this iteration.
__group_543__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_543__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_543__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_544__weights (optional) - T2: weights to optimize.
__group_544__gradients (optional) - T3: gradients computed in this iteration.
__group_544__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_544__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_544__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_545__weights (optional) - T2: weights to optimize.
__group_545__gradients (optional) - T3: gradients computed in this iteration.
__group_545__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_545__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_545__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_546__weights (optional) - T2: weights to optimize.
__group_546__gradients (optional) - T3: gradients computed in this iteration.
__group_546__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_546__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_546__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_547__weights (optional) - T2: weights to optimize.
__group_547__gradients (optional) - T3: gradients computed in this iteration.
__group_547__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_547__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_547__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_548__weights (optional) - T2: weights to optimize.
__group_548__gradients (optional) - T3: gradients computed in this iteration.
__group_548__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_548__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_548__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_549__weights (optional) - T2: weights to optimize.
__group_549__gradients (optional) - T3: gradients computed in this iteration.
__group_549__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_549__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_549__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_550__weights (optional) - T2: weights to optimize.
__group_550__gradients (optional) - T3: gradients computed in this iteration.
__group_550__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_550__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_550__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_551__weights (optional) - T2: weights to optimize.
__group_551__gradients (optional) - T3: gradients computed in this iteration.
__group_551__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_551__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_551__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_552__weights (optional) - T2: weights to optimize.
__group_552__gradients (optional) - T3: gradients computed in this iteration.
__group_552__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_552__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_552__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_553__weights (optional) - T2: weights to optimize.
__group_553__gradients (optional) - T3: gradients computed in this iteration.
__group_553__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_553__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_553__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_554__weights (optional) - T2: weights to optimize.
__group_554__gradients (optional) - T3: gradients computed in this iteration.
__group_554__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_554__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_554__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_555__weights (optional) - T2: weights to optimize.
__group_555__gradients (optional) - T3: gradients computed in this iteration.
__group_555__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_555__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_555__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_556__weights (optional) - T2: weights to optimize.
__group_556__gradients (optional) - T3: gradients computed in this iteration.
__group_556__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_556__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_556__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_557__weights (optional) - T2: weights to optimize.
__group_557__gradients (optional) - T3: gradients computed in this iteration.
__group_557__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_557__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_557__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_558__weights (optional) - T2: weights to optimize.
__group_558__gradients (optional) - T3: gradients computed in this iteration.
__group_558__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_558__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_558__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_559__weights (optional) - T2: weights to optimize.
__group_559__gradients (optional) - T3: gradients computed in this iteration.
__group_559__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_559__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_559__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_560__weights (optional) - T2: weights to optimize.
__group_560__gradients (optional) - T3: gradients computed in this iteration.
__group_560__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_560__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_560__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_561__weights (optional) - T2: weights to optimize.
__group_561__gradients (optional) - T3: gradients computed in this iteration.
__group_561__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_561__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_561__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_562__weights (optional) - T2: weights to optimize.
__group_562__gradients (optional) - T3: gradients computed in this iteration.
__group_562__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_562__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_562__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_563__weights (optional) - T2: weights to optimize.
__group_563__gradients (optional) - T3: gradients computed in this iteration.
__group_563__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_563__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_563__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_564__weights (optional) - T2: weights to optimize.
__group_564__gradients (optional) - T3: gradients computed in this iteration.
__group_564__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_564__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_564__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_565__weights (optional) - T2: weights to optimize.
__group_565__gradients (optional) - T3: gradients computed in this iteration.
__group_565__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_565__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_565__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_566__weights (optional) - T2: weights to optimize.
__group_566__gradients (optional) - T3: gradients computed in this iteration.
__group_566__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_566__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_566__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_567__weights (optional) - T2: weights to optimize.
__group_567__gradients (optional) - T3: gradients computed in this iteration.
__group_567__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_567__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_567__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_568__weights (optional) - T2: weights to optimize.
__group_568__gradients (optional) - T3: gradients computed in this iteration.
__group_568__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_568__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_568__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_569__weights (optional) - T2: weights to optimize.
__group_569__gradients (optional) - T3: gradients computed in this iteration.
__group_569__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_569__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_569__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_570__weights (optional) - T2: weights to optimize.
__group_570__gradients (optional) - T3: gradients computed in this iteration.
__group_570__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_570__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_570__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_571__weights (optional) - T2: weights to optimize.
__group_571__gradients (optional) - T3: gradients computed in this iteration.
__group_571__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_571__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_571__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_572__weights (optional) - T2: weights to optimize.
__group_572__gradients (optional) - T3: gradients computed in this iteration.
__group_572__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_572__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_572__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_573__weights (optional) - T2: weights to optimize.
__group_573__gradients (optional) - T3: gradients computed in this iteration.
__group_573__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_573__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_573__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_574__weights (optional) - T2: weights to optimize.
__group_574__gradients (optional) - T3: gradients computed in this iteration.
__group_574__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_574__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_574__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_575__weights (optional) - T2: weights to optimize.
__group_575__gradients (optional) - T3: gradients computed in this iteration.
__group_575__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_575__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_575__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_576__weights (optional) - T2: weights to optimize.
__group_576__gradients (optional) - T3: gradients computed in this iteration.
__group_576__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_576__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_576__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_577__weights (optional) - T2: weights to optimize.
__group_577__gradients (optional) - T3: gradients computed in this iteration.
__group_577__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_577__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_577__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_578__weights (optional) - T2: weights to optimize.
__group_578__gradients (optional) - T3: gradients computed in this iteration.
__group_578__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_578__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_578__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_579__weights (optional) - T2: weights to optimize.
__group_579__gradients (optional) - T3: gradients computed in this iteration.
__group_579__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_579__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_579__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_580__weights (optional) - T2: weights to optimize.
__group_580__gradients (optional) - T3: gradients computed in this iteration.
__group_580__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_580__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_580__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_581__weights (optional) - T2: weights to optimize.
__group_581__gradients (optional) - T3: gradients computed in this iteration.
__group_581__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_581__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_581__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_582__weights (optional) - T2: weights to optimize.
__group_582__gradients (optional) - T3: gradients computed in this iteration.
__group_582__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_582__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_582__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_583__weights (optional) - T2: weights to optimize.
__group_583__gradients (optional) - T3: gradients computed in this iteration.
__group_583__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_583__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_583__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_584__weights (optional) - T2: weights to optimize.
__group_584__gradients (optional) - T3: gradients computed in this iteration.
__group_584__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_584__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_584__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_585__weights (optional) - T2: weights to optimize.
__group_585__gradients (optional) - T3: gradients computed in this iteration.
__group_585__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_585__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_585__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_586__weights (optional) - T2: weights to optimize.
__group_586__gradients (optional) - T3: gradients computed in this iteration.
__group_586__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_586__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_586__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_587__weights (optional) - T2: weights to optimize.
__group_587__gradients (optional) - T3: gradients computed in this iteration.
__group_587__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_587__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_587__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_588__weights (optional) - T2: weights to optimize.
__group_588__gradients (optional) - T3: gradients computed in this iteration.
__group_588__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_588__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_588__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_589__weights (optional) - T2: weights to optimize.
__group_589__gradients (optional) - T3: gradients computed in this iteration.
__group_589__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_589__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_589__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_590__weights (optional) - T2: weights to optimize.
__group_590__gradients (optional) - T3: gradients computed in this iteration.
__group_590__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_590__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_590__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_591__weights (optional) - T2: weights to optimize.
__group_591__gradients (optional) - T3: gradients computed in this iteration.
__group_591__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_591__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_591__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_592__weights (optional) - T2: weights to optimize.
__group_592__gradients (optional) - T3: gradients computed in this iteration.
__group_592__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_592__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_592__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_593__weights (optional) - T2: weights to optimize.
__group_593__gradients (optional) - T3: gradients computed in this iteration.
__group_593__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_593__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_593__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_594__weights (optional) - T2: weights to optimize.
__group_594__gradients (optional) - T3: gradients computed in this iteration.
__group_594__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_594__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_594__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_595__weights (optional) - T2: weights to optimize.
__group_595__gradients (optional) - T3: gradients computed in this iteration.
__group_595__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_595__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_595__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_596__weights (optional) - T2: weights to optimize.
__group_596__gradients (optional) - T3: gradients computed in this iteration.
__group_596__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_596__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_596__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_597__weights (optional) - T2: weights to optimize.
__group_597__gradients (optional) - T3: gradients computed in this iteration.
__group_597__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_597__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_597__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_598__weights (optional) - T2: weights to optimize.
__group_598__gradients (optional) - T3: gradients computed in this iteration.
__group_598__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_598__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_598__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_599__weights (optional) - T2: weights to optimize.
__group_599__gradients (optional) - T3: gradients computed in this iteration.
__group_599__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_599__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_599__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_600__weights (optional) - T2: weights to optimize.
__group_600__gradients (optional) - T3: gradients computed in this iteration.
__group_600__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_600__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_600__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_601__weights (optional) - T2: weights to optimize.
__group_601__gradients (optional) - T3: gradients computed in this iteration.
__group_601__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_601__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_601__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_602__weights (optional) - T2: weights to optimize.
__group_602__gradients (optional) - T3: gradients computed in this iteration.
__group_602__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_602__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_602__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_603__weights (optional) - T2: weights to optimize.
__group_603__gradients (optional) - T3: gradients computed in this iteration.
__group_603__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_603__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_603__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_604__weights (optional) - T2: weights to optimize.
__group_604__gradients (optional) - T3: gradients computed in this iteration.
__group_604__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_604__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_604__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_605__weights (optional) - T2: weights to optimize.
__group_605__gradients (optional) - T3: gradients computed in this iteration.
__group_605__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_605__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_605__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_606__weights (optional) - T2: weights to optimize.
__group_606__gradients (optional) - T3: gradients computed in this iteration.
__group_606__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_606__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_606__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_607__weights (optional) - T2: weights to optimize.
__group_607__gradients (optional) - T3: gradients computed in this iteration.
__group_607__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_607__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_607__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_608__weights (optional) - T2: weights to optimize.
__group_608__gradients (optional) - T3: gradients computed in this iteration.
__group_608__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_608__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_608__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_609__weights (optional) - T2: weights to optimize.
__group_609__gradients (optional) - T3: gradients computed in this iteration.
__group_609__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_609__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_609__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_610__weights (optional) - T2: weights to optimize.
__group_610__gradients (optional) - T3: gradients computed in this iteration.
__group_610__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_610__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_610__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_611__weights (optional) - T2: weights to optimize.
__group_611__gradients (optional) - T3: gradients computed in this iteration.
__group_611__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_611__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_611__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_612__weights (optional) - T2: weights to optimize.
__group_612__gradients (optional) - T3: gradients computed in this iteration.
__group_612__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_612__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_612__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_613__weights (optional) - T2: weights to optimize.
__group_613__gradients (optional) - T3: gradients computed in this iteration.
__group_613__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_613__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_613__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_614__weights (optional) - T2: weights to optimize.
__group_614__gradients (optional) - T3: gradients computed in this iteration.
__group_614__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_614__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_614__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_615__weights (optional) - T2: weights to optimize.
__group_615__gradients (optional) - T3: gradients computed in this iteration.
__group_615__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_615__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_615__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_616__weights (optional) - T2: weights to optimize.
__group_616__gradients (optional) - T3: gradients computed in this iteration.
__group_616__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_616__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_616__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_617__weights (optional) - T2: weights to optimize.
__group_617__gradients (optional) - T3: gradients computed in this iteration.
__group_617__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_617__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_617__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_618__weights (optional) - T2: weights to optimize.
__group_618__gradients (optional) - T3: gradients computed in this iteration.
__group_618__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_618__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_618__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_619__weights (optional) - T2: weights to optimize.
__group_619__gradients (optional) - T3: gradients computed in this iteration.
__group_619__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_619__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_619__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_620__weights (optional) - T2: weights to optimize.
__group_620__gradients (optional) - T3: gradients computed in this iteration.
__group_620__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_620__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_620__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_621__weights (optional) - T2: weights to optimize.
__group_621__gradients (optional) - T3: gradients computed in this iteration.
__group_621__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_621__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_621__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_622__weights (optional) - T2: weights to optimize.
__group_622__gradients (optional) - T3: gradients computed in this iteration.
__group_622__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_622__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_622__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_623__weights (optional) - T2: weights to optimize.
__group_623__gradients (optional) - T3: gradients computed in this iteration.
__group_623__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_623__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_623__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_624__weights (optional) - T2: weights to optimize.
__group_624__gradients (optional) - T3: gradients computed in this iteration.
__group_624__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_624__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_624__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_625__weights (optional) - T2: weights to optimize.
__group_625__gradients (optional) - T3: gradients computed in this iteration.
__group_625__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_625__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_625__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_626__weights (optional) - T2: weights to optimize.
__group_626__gradients (optional) - T3: gradients computed in this iteration.
__group_626__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_626__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_626__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_627__weights (optional) - T2: weights to optimize.
__group_627__gradients (optional) - T3: gradients computed in this iteration.
__group_627__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_627__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_627__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_628__weights (optional) - T2: weights to optimize.
__group_628__gradients (optional) - T3: gradients computed in this iteration.
__group_628__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_628__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_628__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_629__weights (optional) - T2: weights to optimize.
__group_629__gradients (optional) - T3: gradients computed in this iteration.
__group_629__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_629__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_629__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_630__weights (optional) - T2: weights to optimize.
__group_630__gradients (optional) - T3: gradients computed in this iteration.
__group_630__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_630__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_630__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_631__weights (optional) - T2: weights to optimize.
__group_631__gradients (optional) - T3: gradients computed in this iteration.
__group_631__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_631__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_631__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_632__weights (optional) - T2: weights to optimize.
__group_632__gradients (optional) - T3: gradients computed in this iteration.
__group_632__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_632__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_632__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_633__weights (optional) - T2: weights to optimize.
__group_633__gradients (optional) - T3: gradients computed in this iteration.
__group_633__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_633__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_633__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_634__weights (optional) - T2: weights to optimize.
__group_634__gradients (optional) - T3: gradients computed in this iteration.
__group_634__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_634__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_634__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_635__weights (optional) - T2: weights to optimize.
__group_635__gradients (optional) - T3: gradients computed in this iteration.
__group_635__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_635__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_635__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_636__weights (optional) - T2: weights to optimize.
__group_636__gradients (optional) - T3: gradients computed in this iteration.
__group_636__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_636__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_636__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_637__weights (optional) - T2: weights to optimize.
__group_637__gradients (optional) - T3: gradients computed in this iteration.
__group_637__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_637__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_637__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_638__weights (optional) - T2: weights to optimize.
__group_638__gradients (optional) - T3: gradients computed in this iteration.
__group_638__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_638__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_638__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_639__weights (optional) - T2: weights to optimize.
__group_639__gradients (optional) - T3: gradients computed in this iteration.
__group_639__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_639__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_639__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_640__weights (optional) - T2: weights to optimize.
__group_640__gradients (optional) - T3: gradients computed in this iteration.
__group_640__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_640__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_640__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_641__weights (optional) - T2: weights to optimize.
__group_641__gradients (optional) - T3: gradients computed in this iteration.
__group_641__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_641__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_641__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_642__weights (optional) - T2: weights to optimize.
__group_642__gradients (optional) - T3: gradients computed in this iteration.
__group_642__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_642__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_642__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_643__weights (optional) - T2: weights to optimize.
__group_643__gradients (optional) - T3: gradients computed in this iteration.
__group_643__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_643__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_643__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_644__weights (optional) - T2: weights to optimize.
__group_644__gradients (optional) - T3: gradients computed in this iteration.
__group_644__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_644__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_644__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_645__weights (optional) - T2: weights to optimize.
__group_645__gradients (optional) - T3: gradients computed in this iteration.
__group_645__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_645__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_645__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_646__weights (optional) - T2: weights to optimize.
__group_646__gradients (optional) - T3: gradients computed in this iteration.
__group_646__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_646__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_646__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_647__weights (optional) - T2: weights to optimize.
__group_647__gradients (optional) - T3: gradients computed in this iteration.
__group_647__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_647__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_647__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_648__weights (optional) - T2: weights to optimize.
__group_648__gradients (optional) - T3: gradients computed in this iteration.
__group_648__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_648__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_648__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_649__weights (optional) - T2: weights to optimize.
__group_649__gradients (optional) - T3: gradients computed in this iteration.
__group_649__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_649__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_649__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_650__weights (optional) - T2: weights to optimize.
__group_650__gradients (optional) - T3: gradients computed in this iteration.
__group_650__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_650__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_650__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_651__weights (optional) - T2: weights to optimize.
__group_651__gradients (optional) - T3: gradients computed in this iteration.
__group_651__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_651__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_651__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_652__weights (optional) - T2: weights to optimize.
__group_652__gradients (optional) - T3: gradients computed in this iteration.
__group_652__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_652__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_652__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_653__weights (optional) - T2: weights to optimize.
__group_653__gradients (optional) - T3: gradients computed in this iteration.
__group_653__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_653__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_653__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_654__weights (optional) - T2: weights to optimize.
__group_654__gradients (optional) - T3: gradients computed in this iteration.
__group_654__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_654__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_654__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_655__weights (optional) - T2: weights to optimize.
__group_655__gradients (optional) - T3: gradients computed in this iteration.
__group_655__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_655__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_655__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_656__weights (optional) - T2: weights to optimize.
__group_656__gradients (optional) - T3: gradients computed in this iteration.
__group_656__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_656__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_656__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_657__weights (optional) - T2: weights to optimize.
__group_657__gradients (optional) - T3: gradients computed in this iteration.
__group_657__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_657__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_657__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_658__weights (optional) - T2: weights to optimize.
__group_658__gradients (optional) - T3: gradients computed in this iteration.
__group_658__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_658__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_658__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_659__weights (optional) - T2: weights to optimize.
__group_659__gradients (optional) - T3: gradients computed in this iteration.
__group_659__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_659__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_659__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_660__weights (optional) - T2: weights to optimize.
__group_660__gradients (optional) - T3: gradients computed in this iteration.
__group_660__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_660__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_660__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_661__weights (optional) - T2: weights to optimize.
__group_661__gradients (optional) - T3: gradients computed in this iteration.
__group_661__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_661__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_661__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_662__weights (optional) - T2: weights to optimize.
__group_662__gradients (optional) - T3: gradients computed in this iteration.
__group_662__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_662__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_662__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_663__weights (optional) - T2: weights to optimize.
__group_663__gradients (optional) - T3: gradients computed in this iteration.
__group_663__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_663__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_663__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_664__weights (optional) - T2: weights to optimize.
__group_664__gradients (optional) - T3: gradients computed in this iteration.
__group_664__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_664__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_664__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_665__weights (optional) - T2: weights to optimize.
__group_665__gradients (optional) - T3: gradients computed in this iteration.
__group_665__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_665__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_665__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_666__weights (optional) - T2: weights to optimize.
__group_666__gradients (optional) - T3: gradients computed in this iteration.
__group_666__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_666__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_666__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_667__weights (optional) - T2: weights to optimize.
__group_667__gradients (optional) - T3: gradients computed in this iteration.
__group_667__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_667__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_667__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_668__weights (optional) - T2: weights to optimize.
__group_668__gradients (optional) - T3: gradients computed in this iteration.
__group_668__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_668__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_668__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_669__weights (optional) - T2: weights to optimize.
__group_669__gradients (optional) - T3: gradients computed in this iteration.
__group_669__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_669__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_669__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_670__weights (optional) - T2: weights to optimize.
__group_670__gradients (optional) - T3: gradients computed in this iteration.
__group_670__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_670__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_670__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_671__weights (optional) - T2: weights to optimize.
__group_671__gradients (optional) - T3: gradients computed in this iteration.
__group_671__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_671__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_671__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_672__weights (optional) - T2: weights to optimize.
__group_672__gradients (optional) - T3: gradients computed in this iteration.
__group_672__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_672__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_672__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_673__weights (optional) - T2: weights to optimize.
__group_673__gradients (optional) - T3: gradients computed in this iteration.
__group_673__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_673__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_673__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_674__weights (optional) - T2: weights to optimize.
__group_674__gradients (optional) - T3: gradients computed in this iteration.
__group_674__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_674__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_674__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_675__weights (optional) - T2: weights to optimize.
__group_675__gradients (optional) - T3: gradients computed in this iteration.
__group_675__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_675__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_675__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_676__weights (optional) - T2: weights to optimize.
__group_676__gradients (optional) - T3: gradients computed in this iteration.
__group_676__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_676__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_676__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_677__weights (optional) - T2: weights to optimize.
__group_677__gradients (optional) - T3: gradients computed in this iteration.
__group_677__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_677__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_677__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_678__weights (optional) - T2: weights to optimize.
__group_678__gradients (optional) - T3: gradients computed in this iteration.
__group_678__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_678__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_678__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_679__weights (optional) - T2: weights to optimize.
__group_679__gradients (optional) - T3: gradients computed in this iteration.
__group_679__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_679__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_679__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_680__weights (optional) - T2: weights to optimize.
__group_680__gradients (optional) - T3: gradients computed in this iteration.
__group_680__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_680__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_680__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_681__weights (optional) - T2: weights to optimize.
__group_681__gradients (optional) - T3: gradients computed in this iteration.
__group_681__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_681__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_681__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_682__weights (optional) - T2: weights to optimize.
__group_682__gradients (optional) - T3: gradients computed in this iteration.
__group_682__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_682__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_682__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_683__weights (optional) - T2: weights to optimize.
__group_683__gradients (optional) - T3: gradients computed in this iteration.
__group_683__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_683__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_683__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_684__weights (optional) - T2: weights to optimize.
__group_684__gradients (optional) - T3: gradients computed in this iteration.
__group_684__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_684__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_684__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_685__weights (optional) - T2: weights to optimize.
__group_685__gradients (optional) - T3: gradients computed in this iteration.
__group_685__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_685__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_685__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_686__weights (optional) - T2: weights to optimize.
__group_686__gradients (optional) - T3: gradients computed in this iteration.
__group_686__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_686__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_686__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_687__weights (optional) - T2: weights to optimize.
__group_687__gradients (optional) - T3: gradients computed in this iteration.
__group_687__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_687__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_687__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_688__weights (optional) - T2: weights to optimize.
__group_688__gradients (optional) - T3: gradients computed in this iteration.
__group_688__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_688__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_688__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_689__weights (optional) - T2: weights to optimize.
__group_689__gradients (optional) - T3: gradients computed in this iteration.
__group_689__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_689__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_689__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_690__weights (optional) - T2: weights to optimize.
__group_690__gradients (optional) - T3: gradients computed in this iteration.
__group_690__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_690__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_690__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_691__weights (optional) - T2: weights to optimize.
__group_691__gradients (optional) - T3: gradients computed in this iteration.
__group_691__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_691__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_691__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_692__weights (optional) - T2: weights to optimize.
__group_692__gradients (optional) - T3: gradients computed in this iteration.
__group_692__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_692__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_692__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_693__weights (optional) - T2: weights to optimize.
__group_693__gradients (optional) - T3: gradients computed in this iteration.
__group_693__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_693__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_693__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_694__weights (optional) - T2: weights to optimize.
__group_694__gradients (optional) - T3: gradients computed in this iteration.
__group_694__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_694__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_694__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_695__weights (optional) - T2: weights to optimize.
__group_695__gradients (optional) - T3: gradients computed in this iteration.
__group_695__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_695__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_695__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_696__weights (optional) - T2: weights to optimize.
__group_696__gradients (optional) - T3: gradients computed in this iteration.
__group_696__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_696__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_696__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_697__weights (optional) - T2: weights to optimize.
__group_697__gradients (optional) - T3: gradients computed in this iteration.
__group_697__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_697__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_697__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_698__weights (optional) - T2: weights to optimize.
__group_698__gradients (optional) - T3: gradients computed in this iteration.
__group_698__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_698__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_698__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_699__weights (optional) - T2: weights to optimize.
__group_699__gradients (optional) - T3: gradients computed in this iteration.
__group_699__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_699__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_699__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_700__weights (optional) - T2: weights to optimize.
__group_700__gradients (optional) - T3: gradients computed in this iteration.
__group_700__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_700__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_700__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_701__weights (optional) - T2: weights to optimize.
__group_701__gradients (optional) - T3: gradients computed in this iteration.
__group_701__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_701__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_701__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_702__weights (optional) - T2: weights to optimize.
__group_702__gradients (optional) - T3: gradients computed in this iteration.
__group_702__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_702__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_702__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_703__weights (optional) - T2: weights to optimize.
__group_703__gradients (optional) - T3: gradients computed in this iteration.
__group_703__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_703__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_703__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_704__weights (optional) - T2: weights to optimize.
__group_704__gradients (optional) - T3: gradients computed in this iteration.
__group_704__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_704__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_704__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_705__weights (optional) - T2: weights to optimize.
__group_705__gradients (optional) - T3: gradients computed in this iteration.
__group_705__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_705__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_705__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_706__weights (optional) - T2: weights to optimize.
__group_706__gradients (optional) - T3: gradients computed in this iteration.
__group_706__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_706__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_706__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_707__weights (optional) - T2: weights to optimize.
__group_707__gradients (optional) - T3: gradients computed in this iteration.
__group_707__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_707__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_707__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_708__weights (optional) - T2: weights to optimize.
__group_708__gradients (optional) - T3: gradients computed in this iteration.
__group_708__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_708__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_708__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_709__weights (optional) - T2: weights to optimize.
__group_709__gradients (optional) - T3: gradients computed in this iteration.
__group_709__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_709__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_709__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_710__weights (optional) - T2: weights to optimize.
__group_710__gradients (optional) - T3: gradients computed in this iteration.
__group_710__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_710__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_710__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_711__weights (optional) - T2: weights to optimize.
__group_711__gradients (optional) - T3: gradients computed in this iteration.
__group_711__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_711__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_711__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_712__weights (optional) - T2: weights to optimize.
__group_712__gradients (optional) - T3: gradients computed in this iteration.
__group_712__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_712__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_712__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_713__weights (optional) - T2: weights to optimize.
__group_713__gradients (optional) - T3: gradients computed in this iteration.
__group_713__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_713__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_713__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_714__weights (optional) - T2: weights to optimize.
__group_714__gradients (optional) - T3: gradients computed in this iteration.
__group_714__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_714__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_714__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_715__weights (optional) - T2: weights to optimize.
__group_715__gradients (optional) - T3: gradients computed in this iteration.
__group_715__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_715__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_715__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_716__weights (optional) - T2: weights to optimize.
__group_716__gradients (optional) - T3: gradients computed in this iteration.
__group_716__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_716__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_716__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_717__weights (optional) - T2: weights to optimize.
__group_717__gradients (optional) - T3: gradients computed in this iteration.
__group_717__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_717__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_717__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_718__weights (optional) - T2: weights to optimize.
__group_718__gradients (optional) - T3: gradients computed in this iteration.
__group_718__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_718__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_718__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_719__weights (optional) - T2: weights to optimize.
__group_719__gradients (optional) - T3: gradients computed in this iteration.
__group_719__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_719__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_719__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_720__weights (optional) - T2: weights to optimize.
__group_720__gradients (optional) - T3: gradients computed in this iteration.
__group_720__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_720__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_720__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_721__weights (optional) - T2: weights to optimize.
__group_721__gradients (optional) - T3: gradients computed in this iteration.
__group_721__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_721__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_721__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_722__weights (optional) - T2: weights to optimize.
__group_722__gradients (optional) - T3: gradients computed in this iteration.
__group_722__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_722__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_722__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_723__weights (optional) - T2: weights to optimize.
__group_723__gradients (optional) - T3: gradients computed in this iteration.
__group_723__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_723__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_723__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_724__weights (optional) - T2: weights to optimize.
__group_724__gradients (optional) - T3: gradients computed in this iteration.
__group_724__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_724__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_724__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_725__weights (optional) - T2: weights to optimize.
__group_725__gradients (optional) - T3: gradients computed in this iteration.
__group_725__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_725__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_725__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_726__weights (optional) - T2: weights to optimize.
__group_726__gradients (optional) - T3: gradients computed in this iteration.
__group_726__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_726__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_726__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_727__weights (optional) - T2: weights to optimize.
__group_727__gradients (optional) - T3: gradients computed in this iteration.
__group_727__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_727__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_727__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_728__weights (optional) - T2: weights to optimize.
__group_728__gradients (optional) - T3: gradients computed in this iteration.
__group_728__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_728__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_728__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_729__weights (optional) - T2: weights to optimize.
__group_729__gradients (optional) - T3: gradients computed in this iteration.
__group_729__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_729__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_729__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_730__weights (optional) - T2: weights to optimize.
__group_730__gradients (optional) - T3: gradients computed in this iteration.
__group_730__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_730__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_730__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_731__weights (optional) - T2: weights to optimize.
__group_731__gradients (optional) - T3: gradients computed in this iteration.
__group_731__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_731__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_731__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_732__weights (optional) - T2: weights to optimize.
__group_732__gradients (optional) - T3: gradients computed in this iteration.
__group_732__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_732__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_732__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_733__weights (optional) - T2: weights to optimize.
__group_733__gradients (optional) - T3: gradients computed in this iteration.
__group_733__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_733__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_733__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_734__weights (optional) - T2: weights to optimize.
__group_734__gradients (optional) - T3: gradients computed in this iteration.
__group_734__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_734__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_734__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_735__weights (optional) - T2: weights to optimize.
__group_735__gradients (optional) - T3: gradients computed in this iteration.
__group_735__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_735__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_735__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_736__weights (optional) - T2: weights to optimize.
__group_736__gradients (optional) - T3: gradients computed in this iteration.
__group_736__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_736__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_736__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_737__weights (optional) - T2: weights to optimize.
__group_737__gradients (optional) - T3: gradients computed in this iteration.
__group_737__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_737__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_737__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_738__weights (optional) - T2: weights to optimize.
__group_738__gradients (optional) - T3: gradients computed in this iteration.
__group_738__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_738__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_738__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_739__weights (optional) - T2: weights to optimize.
__group_739__gradients (optional) - T3: gradients computed in this iteration.
__group_739__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_739__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_739__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_740__weights (optional) - T2: weights to optimize.
__group_740__gradients (optional) - T3: gradients computed in this iteration.
__group_740__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_740__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_740__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_741__weights (optional) - T2: weights to optimize.
__group_741__gradients (optional) - T3: gradients computed in this iteration.
__group_741__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_741__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_741__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_742__weights (optional) - T2: weights to optimize.
__group_742__gradients (optional) - T3: gradients computed in this iteration.
__group_742__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_742__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_742__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_743__weights (optional) - T2: weights to optimize.
__group_743__gradients (optional) - T3: gradients computed in this iteration.
__group_743__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_743__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_743__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_744__weights (optional) - T2: weights to optimize.
__group_744__gradients (optional) - T3: gradients computed in this iteration.
__group_744__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_744__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_744__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_745__weights (optional) - T2: weights to optimize.
__group_745__gradients (optional) - T3: gradients computed in this iteration.
__group_745__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_745__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_745__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_746__weights (optional) - T2: weights to optimize.
__group_746__gradients (optional) - T3: gradients computed in this iteration.
__group_746__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_746__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_746__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_747__weights (optional) - T2: weights to optimize.
__group_747__gradients (optional) - T3: gradients computed in this iteration.
__group_747__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_747__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_747__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_748__weights (optional) - T2: weights to optimize.
__group_748__gradients (optional) - T3: gradients computed in this iteration.
__group_748__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_748__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_748__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_749__weights (optional) - T2: weights to optimize.
__group_749__gradients (optional) - T3: gradients computed in this iteration.
__group_749__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_749__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_749__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_750__weights (optional) - T2: weights to optimize.
__group_750__gradients (optional) - T3: gradients computed in this iteration.
__group_750__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_750__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_750__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_751__weights (optional) - T2: weights to optimize.
__group_751__gradients (optional) - T3: gradients computed in this iteration.
__group_751__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_751__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_751__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_752__weights (optional) - T2: weights to optimize.
__group_752__gradients (optional) - T3: gradients computed in this iteration.
__group_752__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_752__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_752__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_753__weights (optional) - T2: weights to optimize.
__group_753__gradients (optional) - T3: gradients computed in this iteration.
__group_753__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_753__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_753__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_754__weights (optional) - T2: weights to optimize.
__group_754__gradients (optional) - T3: gradients computed in this iteration.
__group_754__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_754__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_754__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_755__weights (optional) - T2: weights to optimize.
__group_755__gradients (optional) - T3: gradients computed in this iteration.
__group_755__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_755__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_755__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_756__weights (optional) - T2: weights to optimize.
__group_756__gradients (optional) - T3: gradients computed in this iteration.
__group_756__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_756__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_756__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_757__weights (optional) - T2: weights to optimize.
__group_757__gradients (optional) - T3: gradients computed in this iteration.
__group_757__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_757__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_757__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_758__weights (optional) - T2: weights to optimize.
__group_758__gradients (optional) - T3: gradients computed in this iteration.
__group_758__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_758__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_758__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_759__weights (optional) - T2: weights to optimize.
__group_759__gradients (optional) - T3: gradients computed in this iteration.
__group_759__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_759__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_759__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_760__weights (optional) - T2: weights to optimize.
__group_760__gradients (optional) - T3: gradients computed in this iteration.
__group_760__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_760__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_760__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_761__weights (optional) - T2: weights to optimize.
__group_761__gradients (optional) - T3: gradients computed in this iteration.
__group_761__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_761__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_761__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_762__weights (optional) - T2: weights to optimize.
__group_762__gradients (optional) - T3: gradients computed in this iteration.
__group_762__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_762__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_762__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_763__weights (optional) - T2: weights to optimize.
__group_763__gradients (optional) - T3: gradients computed in this iteration.
__group_763__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_763__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_763__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_764__weights (optional) - T2: weights to optimize.
__group_764__gradients (optional) - T3: gradients computed in this iteration.
__group_764__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_764__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_764__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_765__weights (optional) - T2: weights to optimize.
__group_765__gradients (optional) - T3: gradients computed in this iteration.
__group_765__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_765__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_765__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_766__weights (optional) - T2: weights to optimize.
__group_766__gradients (optional) - T3: gradients computed in this iteration.
__group_766__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_766__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_766__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_767__weights (optional) - T2: weights to optimize.
__group_767__gradients (optional) - T3: gradients computed in this iteration.
__group_767__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_767__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_767__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_768__weights (optional) - T2: weights to optimize.
__group_768__gradients (optional) - T3: gradients computed in this iteration.
__group_768__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_768__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_768__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_769__weights (optional) - T2: weights to optimize.
__group_769__gradients (optional) - T3: gradients computed in this iteration.
__group_769__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_769__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_769__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_770__weights (optional) - T2: weights to optimize.
__group_770__gradients (optional) - T3: gradients computed in this iteration.
__group_770__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_770__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_770__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_771__weights (optional) - T2: weights to optimize.
__group_771__gradients (optional) - T3: gradients computed in this iteration.
__group_771__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_771__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_771__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_772__weights (optional) - T2: weights to optimize.
__group_772__gradients (optional) - T3: gradients computed in this iteration.
__group_772__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_772__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_772__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_773__weights (optional) - T2: weights to optimize.
__group_773__gradients (optional) - T3: gradients computed in this iteration.
__group_773__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_773__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_773__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_774__weights (optional) - T2: weights to optimize.
__group_774__gradients (optional) - T3: gradients computed in this iteration.
__group_774__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_774__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_774__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_775__weights (optional) - T2: weights to optimize.
__group_775__gradients (optional) - T3: gradients computed in this iteration.
__group_775__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_775__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_775__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_776__weights (optional) - T2: weights to optimize.
__group_776__gradients (optional) - T3: gradients computed in this iteration.
__group_776__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_776__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_776__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_777__weights (optional) - T2: weights to optimize.
__group_777__gradients (optional) - T3: gradients computed in this iteration.
__group_777__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_777__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_777__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_778__weights (optional) - T2: weights to optimize.
__group_778__gradients (optional) - T3: gradients computed in this iteration.
__group_778__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_778__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_778__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_779__weights (optional) - T2: weights to optimize.
__group_779__gradients (optional) - T3: gradients computed in this iteration.
__group_779__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_779__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_779__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_780__weights (optional) - T2: weights to optimize.
__group_780__gradients (optional) - T3: gradients computed in this iteration.
__group_780__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_780__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_780__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_781__weights (optional) - T2: weights to optimize.
__group_781__gradients (optional) - T3: gradients computed in this iteration.
__group_781__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_781__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_781__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_782__weights (optional) - T2: weights to optimize.
__group_782__gradients (optional) - T3: gradients computed in this iteration.
__group_782__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_782__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_782__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_783__weights (optional) - T2: weights to optimize.
__group_783__gradients (optional) - T3: gradients computed in this iteration.
__group_783__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_783__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_783__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_784__weights (optional) - T2: weights to optimize.
__group_784__gradients (optional) - T3: gradients computed in this iteration.
__group_784__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_784__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_784__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_785__weights (optional) - T2: weights to optimize.
__group_785__gradients (optional) - T3: gradients computed in this iteration.
__group_785__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_785__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_785__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_786__weights (optional) - T2: weights to optimize.
__group_786__gradients (optional) - T3: gradients computed in this iteration.
__group_786__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_786__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_786__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_787__weights (optional) - T2: weights to optimize.
__group_787__gradients (optional) - T3: gradients computed in this iteration.
__group_787__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_787__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_787__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_788__weights (optional) - T2: weights to optimize.
__group_788__gradients (optional) - T3: gradients computed in this iteration.
__group_788__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_788__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_788__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_789__weights (optional) - T2: weights to optimize.
__group_789__gradients (optional) - T3: gradients computed in this iteration.
__group_789__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_789__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_789__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_790__weights (optional) - T2: weights to optimize.
__group_790__gradients (optional) - T3: gradients computed in this iteration.
__group_790__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_790__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_790__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_791__weights (optional) - T2: weights to optimize.
__group_791__gradients (optional) - T3: gradients computed in this iteration.
__group_791__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_791__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_791__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_792__weights (optional) - T2: weights to optimize.
__group_792__gradients (optional) - T3: gradients computed in this iteration.
__group_792__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_792__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_792__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_793__weights (optional) - T2: weights to optimize.
__group_793__gradients (optional) - T3: gradients computed in this iteration.
__group_793__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_793__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_793__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_794__weights (optional) - T2: weights to optimize.
__group_794__gradients (optional) - T3: gradients computed in this iteration.
__group_794__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_794__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_794__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_795__weights (optional) - T2: weights to optimize.
__group_795__gradients (optional) - T3: gradients computed in this iteration.
__group_795__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_795__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_795__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_796__weights (optional) - T2: weights to optimize.
__group_796__gradients (optional) - T3: gradients computed in this iteration.
__group_796__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_796__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_796__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_797__weights (optional) - T2: weights to optimize.
__group_797__gradients (optional) - T3: gradients computed in this iteration.
__group_797__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_797__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_797__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_798__weights (optional) - T2: weights to optimize.
__group_798__gradients (optional) - T3: gradients computed in this iteration.
__group_798__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_798__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_798__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_799__weights (optional) - T2: weights to optimize.
__group_799__gradients (optional) - T3: gradients computed in this iteration.
__group_799__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_799__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_799__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_800__weights (optional) - T2: weights to optimize.
__group_800__gradients (optional) - T3: gradients computed in this iteration.
__group_800__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_800__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_800__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_801__weights (optional) - T2: weights to optimize.
__group_801__gradients (optional) - T3: gradients computed in this iteration.
__group_801__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_801__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_801__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_802__weights (optional) - T2: weights to optimize.
__group_802__gradients (optional) - T3: gradients computed in this iteration.
__group_802__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_802__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_802__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_803__weights (optional) - T2: weights to optimize.
__group_803__gradients (optional) - T3: gradients computed in this iteration.
__group_803__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_803__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_803__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_804__weights (optional) - T2: weights to optimize.
__group_804__gradients (optional) - T3: gradients computed in this iteration.
__group_804__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_804__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_804__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_805__weights (optional) - T2: weights to optimize.
__group_805__gradients (optional) - T3: gradients computed in this iteration.
__group_805__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_805__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_805__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_806__weights (optional) - T2: weights to optimize.
__group_806__gradients (optional) - T3: gradients computed in this iteration.
__group_806__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_806__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_806__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_807__weights (optional) - T2: weights to optimize.
__group_807__gradients (optional) - T3: gradients computed in this iteration.
__group_807__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_807__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_807__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_808__weights (optional) - T2: weights to optimize.
__group_808__gradients (optional) - T3: gradients computed in this iteration.
__group_808__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_808__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_808__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_809__weights (optional) - T2: weights to optimize.
__group_809__gradients (optional) - T3: gradients computed in this iteration.
__group_809__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_809__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_809__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_810__weights (optional) - T2: weights to optimize.
__group_810__gradients (optional) - T3: gradients computed in this iteration.
__group_810__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_810__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_810__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_811__weights (optional) - T2: weights to optimize.
__group_811__gradients (optional) - T3: gradients computed in this iteration.
__group_811__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_811__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_811__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_812__weights (optional) - T2: weights to optimize.
__group_812__gradients (optional) - T3: gradients computed in this iteration.
__group_812__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_812__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_812__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_813__weights (optional) - T2: weights to optimize.
__group_813__gradients (optional) - T3: gradients computed in this iteration.
__group_813__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_813__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_813__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_814__weights (optional) - T2: weights to optimize.
__group_814__gradients (optional) - T3: gradients computed in this iteration.
__group_814__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_814__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_814__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_815__weights (optional) - T2: weights to optimize.
__group_815__gradients (optional) - T3: gradients computed in this iteration.
__group_815__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_815__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_815__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_816__weights (optional) - T2: weights to optimize.
__group_816__gradients (optional) - T3: gradients computed in this iteration.
__group_816__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_816__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_816__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_817__weights (optional) - T2: weights to optimize.
__group_817__gradients (optional) - T3: gradients computed in this iteration.
__group_817__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_817__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_817__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_818__weights (optional) - T2: weights to optimize.
__group_818__gradients (optional) - T3: gradients computed in this iteration.
__group_818__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_818__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_818__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_819__weights (optional) - T2: weights to optimize.
__group_819__gradients (optional) - T3: gradients computed in this iteration.
__group_819__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_819__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_819__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_820__weights (optional) - T2: weights to optimize.
__group_820__gradients (optional) - T3: gradients computed in this iteration.
__group_820__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_820__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_820__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_821__weights (optional) - T2: weights to optimize.
__group_821__gradients (optional) - T3: gradients computed in this iteration.
__group_821__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_821__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_821__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_822__weights (optional) - T2: weights to optimize.
__group_822__gradients (optional) - T3: gradients computed in this iteration.
__group_822__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_822__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_822__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_823__weights (optional) - T2: weights to optimize.
__group_823__gradients (optional) - T3: gradients computed in this iteration.
__group_823__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_823__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_823__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_824__weights (optional) - T2: weights to optimize.
__group_824__gradients (optional) - T3: gradients computed in this iteration.
__group_824__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_824__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_824__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_825__weights (optional) - T2: weights to optimize.
__group_825__gradients (optional) - T3: gradients computed in this iteration.
__group_825__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_825__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_825__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_826__weights (optional) - T2: weights to optimize.
__group_826__gradients (optional) - T3: gradients computed in this iteration.
__group_826__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_826__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_826__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_827__weights (optional) - T2: weights to optimize.
__group_827__gradients (optional) - T3: gradients computed in this iteration.
__group_827__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_827__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_827__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_828__weights (optional) - T2: weights to optimize.
__group_828__gradients (optional) - T3: gradients computed in this iteration.
__group_828__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_828__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_828__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_829__weights (optional) - T2: weights to optimize.
__group_829__gradients (optional) - T3: gradients computed in this iteration.
__group_829__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_829__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_829__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_830__weights (optional) - T2: weights to optimize.
__group_830__gradients (optional) - T3: gradients computed in this iteration.
__group_830__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_830__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_830__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_831__weights (optional) - T2: weights to optimize.
__group_831__gradients (optional) - T3: gradients computed in this iteration.
__group_831__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_831__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_831__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_832__weights (optional) - T2: weights to optimize.
__group_832__gradients (optional) - T3: gradients computed in this iteration.
__group_832__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_832__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_832__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_833__weights (optional) - T2: weights to optimize.
__group_833__gradients (optional) - T3: gradients computed in this iteration.
__group_833__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_833__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_833__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_834__weights (optional) - T2: weights to optimize.
__group_834__gradients (optional) - T3: gradients computed in this iteration.
__group_834__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_834__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_834__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_835__weights (optional) - T2: weights to optimize.
__group_835__gradients (optional) - T3: gradients computed in this iteration.
__group_835__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_835__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_835__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_836__weights (optional) - T2: weights to optimize.
__group_836__gradients (optional) - T3: gradients computed in this iteration.
__group_836__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_836__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_836__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_837__weights (optional) - T2: weights to optimize.
__group_837__gradients (optional) - T3: gradients computed in this iteration.
__group_837__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_837__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_837__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_838__weights (optional) - T2: weights to optimize.
__group_838__gradients (optional) - T3: gradients computed in this iteration.
__group_838__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_838__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_838__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_839__weights (optional) - T2: weights to optimize.
__group_839__gradients (optional) - T3: gradients computed in this iteration.
__group_839__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_839__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_839__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_840__weights (optional) - T2: weights to optimize.
__group_840__gradients (optional) - T3: gradients computed in this iteration.
__group_840__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_840__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_840__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_841__weights (optional) - T2: weights to optimize.
__group_841__gradients (optional) - T3: gradients computed in this iteration.
__group_841__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_841__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_841__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_842__weights (optional) - T2: weights to optimize.
__group_842__gradients (optional) - T3: gradients computed in this iteration.
__group_842__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_842__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_842__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_843__weights (optional) - T2: weights to optimize.
__group_843__gradients (optional) - T3: gradients computed in this iteration.
__group_843__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_843__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_843__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_844__weights (optional) - T2: weights to optimize.
__group_844__gradients (optional) - T3: gradients computed in this iteration.
__group_844__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_844__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_844__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_845__weights (optional) - T2: weights to optimize.
__group_845__gradients (optional) - T3: gradients computed in this iteration.
__group_845__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_845__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_845__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_846__weights (optional) - T2: weights to optimize.
__group_846__gradients (optional) - T3: gradients computed in this iteration.
__group_846__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_846__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_846__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_847__weights (optional) - T2: weights to optimize.
__group_847__gradients (optional) - T3: gradients computed in this iteration.
__group_847__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_847__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_847__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_848__weights (optional) - T2: weights to optimize.
__group_848__gradients (optional) - T3: gradients computed in this iteration.
__group_848__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_848__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_848__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_849__weights (optional) - T2: weights to optimize.
__group_849__gradients (optional) - T3: gradients computed in this iteration.
__group_849__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_849__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_849__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_850__weights (optional) - T2: weights to optimize.
__group_850__gradients (optional) - T3: gradients computed in this iteration.
__group_850__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_850__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_850__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_851__weights (optional) - T2: weights to optimize.
__group_851__gradients (optional) - T3: gradients computed in this iteration.
__group_851__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_851__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_851__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_852__weights (optional) - T2: weights to optimize.
__group_852__gradients (optional) - T3: gradients computed in this iteration.
__group_852__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_852__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_852__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_853__weights (optional) - T2: weights to optimize.
__group_853__gradients (optional) - T3: gradients computed in this iteration.
__group_853__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_853__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_853__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_854__weights (optional) - T2: weights to optimize.
__group_854__gradients (optional) - T3: gradients computed in this iteration.
__group_854__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_854__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_854__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_855__weights (optional) - T2: weights to optimize.
__group_855__gradients (optional) - T3: gradients computed in this iteration.
__group_855__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_855__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_855__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_856__weights (optional) - T2: weights to optimize.
__group_856__gradients (optional) - T3: gradients computed in this iteration.
__group_856__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_856__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_856__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_857__weights (optional) - T2: weights to optimize.
__group_857__gradients (optional) - T3: gradients computed in this iteration.
__group_857__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_857__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_857__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_858__weights (optional) - T2: weights to optimize.
__group_858__gradients (optional) - T3: gradients computed in this iteration.
__group_858__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_858__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_858__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_859__weights (optional) - T2: weights to optimize.
__group_859__gradients (optional) - T3: gradients computed in this iteration.
__group_859__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_859__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_859__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_860__weights (optional) - T2: weights to optimize.
__group_860__gradients (optional) - T3: gradients computed in this iteration.
__group_860__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_860__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_860__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_861__weights (optional) - T2: weights to optimize.
__group_861__gradients (optional) - T3: gradients computed in this iteration.
__group_861__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_861__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_861__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_862__weights (optional) - T2: weights to optimize.
__group_862__gradients (optional) - T3: gradients computed in this iteration.
__group_862__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_862__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_862__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_863__weights (optional) - T2: weights to optimize.
__group_863__gradients (optional) - T3: gradients computed in this iteration.
__group_863__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_863__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_863__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_864__weights (optional) - T2: weights to optimize.
__group_864__gradients (optional) - T3: gradients computed in this iteration.
__group_864__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_864__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_864__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_865__weights (optional) - T2: weights to optimize.
__group_865__gradients (optional) - T3: gradients computed in this iteration.
__group_865__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_865__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_865__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_866__weights (optional) - T2: weights to optimize.
__group_866__gradients (optional) - T3: gradients computed in this iteration.
__group_866__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_866__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_866__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_867__weights (optional) - T2: weights to optimize.
__group_867__gradients (optional) - T3: gradients computed in this iteration.
__group_867__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_867__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_867__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_868__weights (optional) - T2: weights to optimize.
__group_868__gradients (optional) - T3: gradients computed in this iteration.
__group_868__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_868__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_868__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_869__weights (optional) - T2: weights to optimize.
__group_869__gradients (optional) - T3: gradients computed in this iteration.
__group_869__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_869__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_869__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_870__weights (optional) - T2: weights to optimize.
__group_870__gradients (optional) - T3: gradients computed in this iteration.
__group_870__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_870__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_870__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_871__weights (optional) - T2: weights to optimize.
__group_871__gradients (optional) - T3: gradients computed in this iteration.
__group_871__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_871__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_871__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_872__weights (optional) - T2: weights to optimize.
__group_872__gradients (optional) - T3: gradients computed in this iteration.
__group_872__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_872__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_872__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_873__weights (optional) - T2: weights to optimize.
__group_873__gradients (optional) - T3: gradients computed in this iteration.
__group_873__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_873__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_873__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_874__weights (optional) - T2: weights to optimize.
__group_874__gradients (optional) - T3: gradients computed in this iteration.
__group_874__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_874__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_874__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_875__weights (optional) - T2: weights to optimize.
__group_875__gradients (optional) - T3: gradients computed in this iteration.
__group_875__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_875__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_875__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_876__weights (optional) - T2: weights to optimize.
__group_876__gradients (optional) - T3: gradients computed in this iteration.
__group_876__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_876__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_876__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_877__weights (optional) - T2: weights to optimize.
__group_877__gradients (optional) - T3: gradients computed in this iteration.
__group_877__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_877__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_877__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_878__weights (optional) - T2: weights to optimize.
__group_878__gradients (optional) - T3: gradients computed in this iteration.
__group_878__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_878__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_878__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_879__weights (optional) - T2: weights to optimize.
__group_879__gradients (optional) - T3: gradients computed in this iteration.
__group_879__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_879__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_879__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_880__weights (optional) - T2: weights to optimize.
__group_880__gradients (optional) - T3: gradients computed in this iteration.
__group_880__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_880__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_880__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_881__weights (optional) - T2: weights to optimize.
__group_881__gradients (optional) - T3: gradients computed in this iteration.
__group_881__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_881__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_881__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_882__weights (optional) - T2: weights to optimize.
__group_882__gradients (optional) - T3: gradients computed in this iteration.
__group_882__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_882__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_882__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_883__weights (optional) - T2: weights to optimize.
__group_883__gradients (optional) - T3: gradients computed in this iteration.
__group_883__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_883__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_883__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_884__weights (optional) - T2: weights to optimize.
__group_884__gradients (optional) - T3: gradients computed in this iteration.
__group_884__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_884__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_884__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_885__weights (optional) - T2: weights to optimize.
__group_885__gradients (optional) - T3: gradients computed in this iteration.
__group_885__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_885__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_885__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_886__weights (optional) - T2: weights to optimize.
__group_886__gradients (optional) - T3: gradients computed in this iteration.
__group_886__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_886__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_886__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_887__weights (optional) - T2: weights to optimize.
__group_887__gradients (optional) - T3: gradients computed in this iteration.
__group_887__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_887__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_887__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_888__weights (optional) - T2: weights to optimize.
__group_888__gradients (optional) - T3: gradients computed in this iteration.
__group_888__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_888__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_888__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_889__weights (optional) - T2: weights to optimize.
__group_889__gradients (optional) - T3: gradients computed in this iteration.
__group_889__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_889__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_889__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_890__weights (optional) - T2: weights to optimize.
__group_890__gradients (optional) - T3: gradients computed in this iteration.
__group_890__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_890__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_890__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_891__weights (optional) - T2: weights to optimize.
__group_891__gradients (optional) - T3: gradients computed in this iteration.
__group_891__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_891__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_891__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_892__weights (optional) - T2: weights to optimize.
__group_892__gradients (optional) - T3: gradients computed in this iteration.
__group_892__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_892__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_892__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_893__weights (optional) - T2: weights to optimize.
__group_893__gradients (optional) - T3: gradients computed in this iteration.
__group_893__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_893__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_893__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_894__weights (optional) - T2: weights to optimize.
__group_894__gradients (optional) - T3: gradients computed in this iteration.
__group_894__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_894__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_894__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_895__weights (optional) - T2: weights to optimize.
__group_895__gradients (optional) - T3: gradients computed in this iteration.
__group_895__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_895__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_895__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_896__weights (optional) - T2: weights to optimize.
__group_896__gradients (optional) - T3: gradients computed in this iteration.
__group_896__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_896__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_896__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_897__weights (optional) - T2: weights to optimize.
__group_897__gradients (optional) - T3: gradients computed in this iteration.
__group_897__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_897__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_897__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_898__weights (optional) - T2: weights to optimize.
__group_898__gradients (optional) - T3: gradients computed in this iteration.
__group_898__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_898__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_898__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_899__weights (optional) - T2: weights to optimize.
__group_899__gradients (optional) - T3: gradients computed in this iteration.
__group_899__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_899__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_899__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_900__weights (optional) - T2: weights to optimize.
__group_900__gradients (optional) - T3: gradients computed in this iteration.
__group_900__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_900__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_900__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_901__weights (optional) - T2: weights to optimize.
__group_901__gradients (optional) - T3: gradients computed in this iteration.
__group_901__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_901__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_901__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_902__weights (optional) - T2: weights to optimize.
__group_902__gradients (optional) - T3: gradients computed in this iteration.
__group_902__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_902__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_902__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_903__weights (optional) - T2: weights to optimize.
__group_903__gradients (optional) - T3: gradients computed in this iteration.
__group_903__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_903__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_903__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_904__weights (optional) - T2: weights to optimize.
__group_904__gradients (optional) - T3: gradients computed in this iteration.
__group_904__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_904__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_904__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_905__weights (optional) - T2: weights to optimize.
__group_905__gradients (optional) - T3: gradients computed in this iteration.
__group_905__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_905__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_905__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_906__weights (optional) - T2: weights to optimize.
__group_906__gradients (optional) - T3: gradients computed in this iteration.
__group_906__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_906__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_906__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_907__weights (optional) - T2: weights to optimize.
__group_907__gradients (optional) - T3: gradients computed in this iteration.
__group_907__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_907__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_907__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_908__weights (optional) - T2: weights to optimize.
__group_908__gradients (optional) - T3: gradients computed in this iteration.
__group_908__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_908__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_908__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_909__weights (optional) - T2: weights to optimize.
__group_909__gradients (optional) - T3: gradients computed in this iteration.
__group_909__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_909__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_909__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_910__weights (optional) - T2: weights to optimize.
__group_910__gradients (optional) - T3: gradients computed in this iteration.
__group_910__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_910__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_910__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_911__weights (optional) - T2: weights to optimize.
__group_911__gradients (optional) - T3: gradients computed in this iteration.
__group_911__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_911__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_911__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_912__weights (optional) - T2: weights to optimize.
__group_912__gradients (optional) - T3: gradients computed in this iteration.
__group_912__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_912__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_912__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_913__weights (optional) - T2: weights to optimize.
__group_913__gradients (optional) - T3: gradients computed in this iteration.
__group_913__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_913__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_913__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_914__weights (optional) - T2: weights to optimize.
__group_914__gradients (optional) - T3: gradients computed in this iteration.
__group_914__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_914__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_914__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_915__weights (optional) - T2: weights to optimize.
__group_915__gradients (optional) - T3: gradients computed in this iteration.
__group_915__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_915__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_915__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_916__weights (optional) - T2: weights to optimize.
__group_916__gradients (optional) - T3: gradients computed in this iteration.
__group_916__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_916__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_916__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_917__weights (optional) - T2: weights to optimize.
__group_917__gradients (optional) - T3: gradients computed in this iteration.
__group_917__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_917__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_917__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_918__weights (optional) - T2: weights to optimize.
__group_918__gradients (optional) - T3: gradients computed in this iteration.
__group_918__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_918__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_918__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_919__weights (optional) - T2: weights to optimize.
__group_919__gradients (optional) - T3: gradients computed in this iteration.
__group_919__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_919__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_919__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_920__weights (optional) - T2: weights to optimize.
__group_920__gradients (optional) - T3: gradients computed in this iteration.
__group_920__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_920__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_920__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_921__weights (optional) - T2: weights to optimize.
__group_921__gradients (optional) - T3: gradients computed in this iteration.
__group_921__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_921__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_921__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_922__weights (optional) - T2: weights to optimize.
__group_922__gradients (optional) - T3: gradients computed in this iteration.
__group_922__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_922__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_922__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_923__weights (optional) - T2: weights to optimize.
__group_923__gradients (optional) - T3: gradients computed in this iteration.
__group_923__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_923__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_923__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_924__weights (optional) - T2: weights to optimize.
__group_924__gradients (optional) - T3: gradients computed in this iteration.
__group_924__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_924__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_924__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_925__weights (optional) - T2: weights to optimize.
__group_925__gradients (optional) - T3: gradients computed in this iteration.
__group_925__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_925__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_925__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_926__weights (optional) - T2: weights to optimize.
__group_926__gradients (optional) - T3: gradients computed in this iteration.
__group_926__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_926__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_926__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_927__weights (optional) - T2: weights to optimize.
__group_927__gradients (optional) - T3: gradients computed in this iteration.
__group_927__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_927__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_927__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_928__weights (optional) - T2: weights to optimize.
__group_928__gradients (optional) - T3: gradients computed in this iteration.
__group_928__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_928__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_928__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_929__weights (optional) - T2: weights to optimize.
__group_929__gradients (optional) - T3: gradients computed in this iteration.
__group_929__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_929__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_929__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_930__weights (optional) - T2: weights to optimize.
__group_930__gradients (optional) - T3: gradients computed in this iteration.
__group_930__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_930__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_930__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_931__weights (optional) - T2: weights to optimize.
__group_931__gradients (optional) - T3: gradients computed in this iteration.
__group_931__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_931__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_931__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_932__weights (optional) - T2: weights to optimize.
__group_932__gradients (optional) - T3: gradients computed in this iteration.
__group_932__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_932__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_932__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_933__weights (optional) - T2: weights to optimize.
__group_933__gradients (optional) - T3: gradients computed in this iteration.
__group_933__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_933__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_933__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_934__weights (optional) - T2: weights to optimize.
__group_934__gradients (optional) - T3: gradients computed in this iteration.
__group_934__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_934__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_934__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_935__weights (optional) - T2: weights to optimize.
__group_935__gradients (optional) - T3: gradients computed in this iteration.
__group_935__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_935__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_935__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_936__weights (optional) - T2: weights to optimize.
__group_936__gradients (optional) - T3: gradients computed in this iteration.
__group_936__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_936__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_936__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_937__weights (optional) - T2: weights to optimize.
__group_937__gradients (optional) - T3: gradients computed in this iteration.
__group_937__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_937__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_937__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_938__weights (optional) - T2: weights to optimize.
__group_938__gradients (optional) - T3: gradients computed in this iteration.
__group_938__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_938__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_938__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_939__weights (optional) - T2: weights to optimize.
__group_939__gradients (optional) - T3: gradients computed in this iteration.
__group_939__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_939__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_939__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_940__weights (optional) - T2: weights to optimize.
__group_940__gradients (optional) - T3: gradients computed in this iteration.
__group_940__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_940__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_940__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_941__weights (optional) - T2: weights to optimize.
__group_941__gradients (optional) - T3: gradients computed in this iteration.
__group_941__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_941__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_941__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_942__weights (optional) - T2: weights to optimize.
__group_942__gradients (optional) - T3: gradients computed in this iteration.
__group_942__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_942__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_942__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_943__weights (optional) - T2: weights to optimize.
__group_943__gradients (optional) - T3: gradients computed in this iteration.
__group_943__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_943__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_943__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_944__weights (optional) - T2: weights to optimize.
__group_944__gradients (optional) - T3: gradients computed in this iteration.
__group_944__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_944__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_944__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_945__weights (optional) - T2: weights to optimize.
__group_945__gradients (optional) - T3: gradients computed in this iteration.
__group_945__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_945__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_945__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_946__weights (optional) - T2: weights to optimize.
__group_946__gradients (optional) - T3: gradients computed in this iteration.
__group_946__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_946__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_946__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_947__weights (optional) - T2: weights to optimize.
__group_947__gradients (optional) - T3: gradients computed in this iteration.
__group_947__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_947__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_947__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_948__weights (optional) - T2: weights to optimize.
__group_948__gradients (optional) - T3: gradients computed in this iteration.
__group_948__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_948__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_948__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_949__weights (optional) - T2: weights to optimize.
__group_949__gradients (optional) - T3: gradients computed in this iteration.
__group_949__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_949__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_949__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_950__weights (optional) - T2: weights to optimize.
__group_950__gradients (optional) - T3: gradients computed in this iteration.
__group_950__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_950__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_950__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_951__weights (optional) - T2: weights to optimize.
__group_951__gradients (optional) - T3: gradients computed in this iteration.
__group_951__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_951__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_951__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_952__weights (optional) - T2: weights to optimize.
__group_952__gradients (optional) - T3: gradients computed in this iteration.
__group_952__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_952__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_952__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_953__weights (optional) - T2: weights to optimize.
__group_953__gradients (optional) - T3: gradients computed in this iteration.
__group_953__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_953__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_953__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_954__weights (optional) - T2: weights to optimize.
__group_954__gradients (optional) - T3: gradients computed in this iteration.
__group_954__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_954__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_954__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_955__weights (optional) - T2: weights to optimize.
__group_955__gradients (optional) - T3: gradients computed in this iteration.
__group_955__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_955__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_955__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_956__weights (optional) - T2: weights to optimize.
__group_956__gradients (optional) - T3: gradients computed in this iteration.
__group_956__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_956__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_956__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_957__weights (optional) - T2: weights to optimize.
__group_957__gradients (optional) - T3: gradients computed in this iteration.
__group_957__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_957__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_957__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_958__weights (optional) - T2: weights to optimize.
__group_958__gradients (optional) - T3: gradients computed in this iteration.
__group_958__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_958__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_958__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_959__weights (optional) - T2: weights to optimize.
__group_959__gradients (optional) - T3: gradients computed in this iteration.
__group_959__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_959__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_959__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_960__weights (optional) - T2: weights to optimize.
__group_960__gradients (optional) - T3: gradients computed in this iteration.
__group_960__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_960__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_960__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_961__weights (optional) - T2: weights to optimize.
__group_961__gradients (optional) - T3: gradients computed in this iteration.
__group_961__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_961__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_961__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_962__weights (optional) - T2: weights to optimize.
__group_962__gradients (optional) - T3: gradients computed in this iteration.
__group_962__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_962__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_962__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_963__weights (optional) - T2: weights to optimize.
__group_963__gradients (optional) - T3: gradients computed in this iteration.
__group_963__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_963__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_963__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_964__weights (optional) - T2: weights to optimize.
__group_964__gradients (optional) - T3: gradients computed in this iteration.
__group_964__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_964__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_964__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_965__weights (optional) - T2: weights to optimize.
__group_965__gradients (optional) - T3: gradients computed in this iteration.
__group_965__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_965__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_965__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_966__weights (optional) - T2: weights to optimize.
__group_966__gradients (optional) - T3: gradients computed in this iteration.
__group_966__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_966__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_966__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_967__weights (optional) - T2: weights to optimize.
__group_967__gradients (optional) - T3: gradients computed in this iteration.
__group_967__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_967__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_967__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_968__weights (optional) - T2: weights to optimize.
__group_968__gradients (optional) - T3: gradients computed in this iteration.
__group_968__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_968__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_968__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_969__weights (optional) - T2: weights to optimize.
__group_969__gradients (optional) - T3: gradients computed in this iteration.
__group_969__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_969__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_969__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_970__weights (optional) - T2: weights to optimize.
__group_970__gradients (optional) - T3: gradients computed in this iteration.
__group_970__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_970__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_970__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_971__weights (optional) - T2: weights to optimize.
__group_971__gradients (optional) - T3: gradients computed in this iteration.
__group_971__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_971__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_971__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_972__weights (optional) - T2: weights to optimize.
__group_972__gradients (optional) - T3: gradients computed in this iteration.
__group_972__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_972__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_972__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_973__weights (optional) - T2: weights to optimize.
__group_973__gradients (optional) - T3: gradients computed in this iteration.
__group_973__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_973__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_973__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_974__weights (optional) - T2: weights to optimize.
__group_974__gradients (optional) - T3: gradients computed in this iteration.
__group_974__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_974__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_974__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_975__weights (optional) - T2: weights to optimize.
__group_975__gradients (optional) - T3: gradients computed in this iteration.
__group_975__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_975__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_975__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_976__weights (optional) - T2: weights to optimize.
__group_976__gradients (optional) - T3: gradients computed in this iteration.
__group_976__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_976__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_976__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_977__weights (optional) - T2: weights to optimize.
__group_977__gradients (optional) - T3: gradients computed in this iteration.
__group_977__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_977__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_977__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_978__weights (optional) - T2: weights to optimize.
__group_978__gradients (optional) - T3: gradients computed in this iteration.
__group_978__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_978__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_978__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_979__weights (optional) - T2: weights to optimize.
__group_979__gradients (optional) - T3: gradients computed in this iteration.
__group_979__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_979__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_979__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_980__weights (optional) - T2: weights to optimize.
__group_980__gradients (optional) - T3: gradients computed in this iteration.
__group_980__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_980__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_980__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_981__weights (optional) - T2: weights to optimize.
__group_981__gradients (optional) - T3: gradients computed in this iteration.
__group_981__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_981__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_981__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_982__weights (optional) - T2: weights to optimize.
__group_982__gradients (optional) - T3: gradients computed in this iteration.
__group_982__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_982__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_982__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_983__weights (optional) - T2: weights to optimize.
__group_983__gradients (optional) - T3: gradients computed in this iteration.
__group_983__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_983__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_983__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_984__weights (optional) - T2: weights to optimize.
__group_984__gradients (optional) - T3: gradients computed in this iteration.
__group_984__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_984__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_984__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_985__weights (optional) - T2: weights to optimize.
__group_985__gradients (optional) - T3: gradients computed in this iteration.
__group_985__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_985__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_985__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_986__weights (optional) - T2: weights to optimize.
__group_986__gradients (optional) - T3: gradients computed in this iteration.
__group_986__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_986__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_986__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_987__weights (optional) - T2: weights to optimize.
__group_987__gradients (optional) - T3: gradients computed in this iteration.
__group_987__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_987__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_987__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_988__weights (optional) - T2: weights to optimize.
__group_988__gradients (optional) - T3: gradients computed in this iteration.
__group_988__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_988__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_988__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_989__weights (optional) - T2: weights to optimize.
__group_989__gradients (optional) - T3: gradients computed in this iteration.
__group_989__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_989__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_989__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_990__weights (optional) - T2: weights to optimize.
__group_990__gradients (optional) - T3: gradients computed in this iteration.
__group_990__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_990__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_990__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_991__weights (optional) - T2: weights to optimize.
__group_991__gradients (optional) - T3: gradients computed in this iteration.
__group_991__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_991__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_991__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_992__weights (optional) - T2: weights to optimize.
__group_992__gradients (optional) - T3: gradients computed in this iteration.
__group_992__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_992__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_992__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_993__weights (optional) - T2: weights to optimize.
__group_993__gradients (optional) - T3: gradients computed in this iteration.
__group_993__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_993__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_993__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_994__weights (optional) - T2: weights to optimize.
__group_994__gradients (optional) - T3: gradients computed in this iteration.
__group_994__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_994__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_994__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_995__weights (optional) - T2: weights to optimize.
__group_995__gradients (optional) - T3: gradients computed in this iteration.
__group_995__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_995__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_995__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_996__weights (optional) - T2: weights to optimize.
__group_996__gradients (optional) - T3: gradients computed in this iteration.
__group_996__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_996__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_996__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_997__weights (optional) - T2: weights to optimize.
__group_997__gradients (optional) - T3: gradients computed in this iteration.
__group_997__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_997__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_997__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_998__weights (optional) - T2: weights to optimize.
__group_998__gradients (optional) - T3: gradients computed in this iteration.
__group_998__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_998__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_998__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_999__weights (optional) - T2: weights to optimize.
__group_999__gradients (optional) - T3: gradients computed in this iteration.
__group_999__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_999__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_999__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1000__weights (optional) - T2: weights to optimize.
__group_1000__gradients (optional) - T3: gradients computed in this iteration.
__group_1000__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1000__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1000__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1001__weights (optional) - T2: weights to optimize.
__group_1001__gradients (optional) - T3: gradients computed in this iteration.
__group_1001__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1001__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1001__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1002__weights (optional) - T2: weights to optimize.
__group_1002__gradients (optional) - T3: gradients computed in this iteration.
__group_1002__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1002__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1002__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1003__weights (optional) - T2: weights to optimize.
__group_1003__gradients (optional) - T3: gradients computed in this iteration.
__group_1003__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1003__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1003__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1004__weights (optional) - T2: weights to optimize.
__group_1004__gradients (optional) - T3: gradients computed in this iteration.
__group_1004__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1004__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1004__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1005__weights (optional) - T2: weights to optimize.
__group_1005__gradients (optional) - T3: gradients computed in this iteration.
__group_1005__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1005__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1005__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1006__weights (optional) - T2: weights to optimize.
__group_1006__gradients (optional) - T3: gradients computed in this iteration.
__group_1006__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1006__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1006__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1007__weights (optional) - T2: weights to optimize.
__group_1007__gradients (optional) - T3: gradients computed in this iteration.
__group_1007__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1007__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1007__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1008__weights (optional) - T2: weights to optimize.
__group_1008__gradients (optional) - T3: gradients computed in this iteration.
__group_1008__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1008__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1008__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1009__weights (optional) - T2: weights to optimize.
__group_1009__gradients (optional) - T3: gradients computed in this iteration.
__group_1009__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1009__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1009__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1010__weights (optional) - T2: weights to optimize.
__group_1010__gradients (optional) - T3: gradients computed in this iteration.
__group_1010__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1010__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1010__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1011__weights (optional) - T2: weights to optimize.
__group_1011__gradients (optional) - T3: gradients computed in this iteration.
__group_1011__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1011__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1011__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1012__weights (optional) - T2: weights to optimize.
__group_1012__gradients (optional) - T3: gradients computed in this iteration.
__group_1012__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1012__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1012__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1013__weights (optional) - T2: weights to optimize.
__group_1013__gradients (optional) - T3: gradients computed in this iteration.
__group_1013__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1013__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1013__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1014__weights (optional) - T2: weights to optimize.
__group_1014__gradients (optional) - T3: gradients computed in this iteration.
__group_1014__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1014__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1014__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1015__weights (optional) - T2: weights to optimize.
__group_1015__gradients (optional) - T3: gradients computed in this iteration.
__group_1015__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1015__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1015__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1016__weights (optional) - T2: weights to optimize.
__group_1016__gradients (optional) - T3: gradients computed in this iteration.
__group_1016__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1016__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1016__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1017__weights (optional) - T2: weights to optimize.
__group_1017__gradients (optional) - T3: gradients computed in this iteration.
__group_1017__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1017__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1017__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1018__weights (optional) - T2: weights to optimize.
__group_1018__gradients (optional) - T3: gradients computed in this iteration.
__group_1018__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1018__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1018__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1019__weights (optional) - T2: weights to optimize.
__group_1019__gradients (optional) - T3: gradients computed in this iteration.
__group_1019__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1019__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1019__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1020__weights (optional) - T2: weights to optimize.
__group_1020__gradients (optional) - T3: gradients computed in this iteration.
__group_1020__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1020__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1020__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1021__weights (optional) - T2: weights to optimize.
__group_1021__gradients (optional) - T3: gradients computed in this iteration.
__group_1021__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1021__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1021__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1022__weights (optional) - T2: weights to optimize.
__group_1022__gradients (optional) - T3: gradients computed in this iteration.
__group_1022__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1022__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1022__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
__group_1023__weights (optional) - T2: weights to optimize.
__group_1023__gradients (optional) - T3: gradients computed in this iteration.
__group_1023__moment1 (optional) - T4: exponentially averaged historical gradients.
__group_1023__moment2 (optional) - T4: exponentially averaged historical squared gradients.
__group_1023__mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: FP16 or BF16 weights to optimize.
Outputs
Between 0 and 5121 outputs.
new_step (optional, heterogeneous) - TInt64: One-based index of the next training iteration.
__group_0__new_weights (optional) - T2: New weights
__group_0__new_gradients (optional) - T3: New gradients
__group_0__new_moment_1 (optional) - T4: New averaged gradients
__group_0__new_moment_2 (optional) - T4: New averaged squared gradients
__group_0__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1__new_weights (optional) - T2: New weights
__group_1__new_gradients (optional) - T3: New gradients
__group_1__new_moment_1 (optional) - T4: New averaged gradients
__group_1__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_2__new_weights (optional) - T2: New weights
__group_2__new_gradients (optional) - T3: New gradients
__group_2__new_moment_1 (optional) - T4: New averaged gradients
__group_2__new_moment_2 (optional) - T4: New averaged squared gradients
__group_2__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_3__new_weights (optional) - T2: New weights
__group_3__new_gradients (optional) - T3: New gradients
__group_3__new_moment_1 (optional) - T4: New averaged gradients
__group_3__new_moment_2 (optional) - T4: New averaged squared gradients
__group_3__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_4__new_weights (optional) - T2: New weights
__group_4__new_gradients (optional) - T3: New gradients
__group_4__new_moment_1 (optional) - T4: New averaged gradients
__group_4__new_moment_2 (optional) - T4: New averaged squared gradients
__group_4__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_5__new_weights (optional) - T2: New weights
__group_5__new_gradients (optional) - T3: New gradients
__group_5__new_moment_1 (optional) - T4: New averaged gradients
__group_5__new_moment_2 (optional) - T4: New averaged squared gradients
__group_5__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_6__new_weights (optional) - T2: New weights
__group_6__new_gradients (optional) - T3: New gradients
__group_6__new_moment_1 (optional) - T4: New averaged gradients
__group_6__new_moment_2 (optional) - T4: New averaged squared gradients
__group_6__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_7__new_weights (optional) - T2: New weights
__group_7__new_gradients (optional) - T3: New gradients
__group_7__new_moment_1 (optional) - T4: New averaged gradients
__group_7__new_moment_2 (optional) - T4: New averaged squared gradients
__group_7__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_8__new_weights (optional) - T2: New weights
__group_8__new_gradients (optional) - T3: New gradients
__group_8__new_moment_1 (optional) - T4: New averaged gradients
__group_8__new_moment_2 (optional) - T4: New averaged squared gradients
__group_8__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_9__new_weights (optional) - T2: New weights
__group_9__new_gradients (optional) - T3: New gradients
__group_9__new_moment_1 (optional) - T4: New averaged gradients
__group_9__new_moment_2 (optional) - T4: New averaged squared gradients
__group_9__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_10__new_weights (optional) - T2: New weights
__group_10__new_gradients (optional) - T3: New gradients
__group_10__new_moment_1 (optional) - T4: New averaged gradients
__group_10__new_moment_2 (optional) - T4: New averaged squared gradients
__group_10__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_11__new_weights (optional) - T2: New weights
__group_11__new_gradients (optional) - T3: New gradients
__group_11__new_moment_1 (optional) - T4: New averaged gradients
__group_11__new_moment_2 (optional) - T4: New averaged squared gradients
__group_11__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_12__new_weights (optional) - T2: New weights
__group_12__new_gradients (optional) - T3: New gradients
__group_12__new_moment_1 (optional) - T4: New averaged gradients
__group_12__new_moment_2 (optional) - T4: New averaged squared gradients
__group_12__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_13__new_weights (optional) - T2: New weights
__group_13__new_gradients (optional) - T3: New gradients
__group_13__new_moment_1 (optional) - T4: New averaged gradients
__group_13__new_moment_2 (optional) - T4: New averaged squared gradients
__group_13__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_14__new_weights (optional) - T2: New weights
__group_14__new_gradients (optional) - T3: New gradients
__group_14__new_moment_1 (optional) - T4: New averaged gradients
__group_14__new_moment_2 (optional) - T4: New averaged squared gradients
__group_14__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_15__new_weights (optional) - T2: New weights
__group_15__new_gradients (optional) - T3: New gradients
__group_15__new_moment_1 (optional) - T4: New averaged gradients
__group_15__new_moment_2 (optional) - T4: New averaged squared gradients
__group_15__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_16__new_weights (optional) - T2: New weights
__group_16__new_gradients (optional) - T3: New gradients
__group_16__new_moment_1 (optional) - T4: New averaged gradients
__group_16__new_moment_2 (optional) - T4: New averaged squared gradients
__group_16__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_17__new_weights (optional) - T2: New weights
__group_17__new_gradients (optional) - T3: New gradients
__group_17__new_moment_1 (optional) - T4: New averaged gradients
__group_17__new_moment_2 (optional) - T4: New averaged squared gradients
__group_17__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_18__new_weights (optional) - T2: New weights
__group_18__new_gradients (optional) - T3: New gradients
__group_18__new_moment_1 (optional) - T4: New averaged gradients
__group_18__new_moment_2 (optional) - T4: New averaged squared gradients
__group_18__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_19__new_weights (optional) - T2: New weights
__group_19__new_gradients (optional) - T3: New gradients
__group_19__new_moment_1 (optional) - T4: New averaged gradients
__group_19__new_moment_2 (optional) - T4: New averaged squared gradients
__group_19__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_20__new_weights (optional) - T2: New weights
__group_20__new_gradients (optional) - T3: New gradients
__group_20__new_moment_1 (optional) - T4: New averaged gradients
__group_20__new_moment_2 (optional) - T4: New averaged squared gradients
__group_20__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_21__new_weights (optional) - T2: New weights
__group_21__new_gradients (optional) - T3: New gradients
__group_21__new_moment_1 (optional) - T4: New averaged gradients
__group_21__new_moment_2 (optional) - T4: New averaged squared gradients
__group_21__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_22__new_weights (optional) - T2: New weights
__group_22__new_gradients (optional) - T3: New gradients
__group_22__new_moment_1 (optional) - T4: New averaged gradients
__group_22__new_moment_2 (optional) - T4: New averaged squared gradients
__group_22__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_23__new_weights (optional) - T2: New weights
__group_23__new_gradients (optional) - T3: New gradients
__group_23__new_moment_1 (optional) - T4: New averaged gradients
__group_23__new_moment_2 (optional) - T4: New averaged squared gradients
__group_23__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_24__new_weights (optional) - T2: New weights
__group_24__new_gradients (optional) - T3: New gradients
__group_24__new_moment_1 (optional) - T4: New averaged gradients
__group_24__new_moment_2 (optional) - T4: New averaged squared gradients
__group_24__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_25__new_weights (optional) - T2: New weights
__group_25__new_gradients (optional) - T3: New gradients
__group_25__new_moment_1 (optional) - T4: New averaged gradients
__group_25__new_moment_2 (optional) - T4: New averaged squared gradients
__group_25__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_26__new_weights (optional) - T2: New weights
__group_26__new_gradients (optional) - T3: New gradients
__group_26__new_moment_1 (optional) - T4: New averaged gradients
__group_26__new_moment_2 (optional) - T4: New averaged squared gradients
__group_26__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_27__new_weights (optional) - T2: New weights
__group_27__new_gradients (optional) - T3: New gradients
__group_27__new_moment_1 (optional) - T4: New averaged gradients
__group_27__new_moment_2 (optional) - T4: New averaged squared gradients
__group_27__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_28__new_weights (optional) - T2: New weights
__group_28__new_gradients (optional) - T3: New gradients
__group_28__new_moment_1 (optional) - T4: New averaged gradients
__group_28__new_moment_2 (optional) - T4: New averaged squared gradients
__group_28__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_29__new_weights (optional) - T2: New weights
__group_29__new_gradients (optional) - T3: New gradients
__group_29__new_moment_1 (optional) - T4: New averaged gradients
__group_29__new_moment_2 (optional) - T4: New averaged squared gradients
__group_29__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_30__new_weights (optional) - T2: New weights
__group_30__new_gradients (optional) - T3: New gradients
__group_30__new_moment_1 (optional) - T4: New averaged gradients
__group_30__new_moment_2 (optional) - T4: New averaged squared gradients
__group_30__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_31__new_weights (optional) - T2: New weights
__group_31__new_gradients (optional) - T3: New gradients
__group_31__new_moment_1 (optional) - T4: New averaged gradients
__group_31__new_moment_2 (optional) - T4: New averaged squared gradients
__group_31__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_32__new_weights (optional) - T2: New weights
__group_32__new_gradients (optional) - T3: New gradients
__group_32__new_moment_1 (optional) - T4: New averaged gradients
__group_32__new_moment_2 (optional) - T4: New averaged squared gradients
__group_32__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_33__new_weights (optional) - T2: New weights
__group_33__new_gradients (optional) - T3: New gradients
__group_33__new_moment_1 (optional) - T4: New averaged gradients
__group_33__new_moment_2 (optional) - T4: New averaged squared gradients
__group_33__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_34__new_weights (optional) - T2: New weights
__group_34__new_gradients (optional) - T3: New gradients
__group_34__new_moment_1 (optional) - T4: New averaged gradients
__group_34__new_moment_2 (optional) - T4: New averaged squared gradients
__group_34__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_35__new_weights (optional) - T2: New weights
__group_35__new_gradients (optional) - T3: New gradients
__group_35__new_moment_1 (optional) - T4: New averaged gradients
__group_35__new_moment_2 (optional) - T4: New averaged squared gradients
__group_35__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_36__new_weights (optional) - T2: New weights
__group_36__new_gradients (optional) - T3: New gradients
__group_36__new_moment_1 (optional) - T4: New averaged gradients
__group_36__new_moment_2 (optional) - T4: New averaged squared gradients
__group_36__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_37__new_weights (optional) - T2: New weights
__group_37__new_gradients (optional) - T3: New gradients
__group_37__new_moment_1 (optional) - T4: New averaged gradients
__group_37__new_moment_2 (optional) - T4: New averaged squared gradients
__group_37__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_38__new_weights (optional) - T2: New weights
__group_38__new_gradients (optional) - T3: New gradients
__group_38__new_moment_1 (optional) - T4: New averaged gradients
__group_38__new_moment_2 (optional) - T4: New averaged squared gradients
__group_38__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_39__new_weights (optional) - T2: New weights
__group_39__new_gradients (optional) - T3: New gradients
__group_39__new_moment_1 (optional) - T4: New averaged gradients
__group_39__new_moment_2 (optional) - T4: New averaged squared gradients
__group_39__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_40__new_weights (optional) - T2: New weights
__group_40__new_gradients (optional) - T3: New gradients
__group_40__new_moment_1 (optional) - T4: New averaged gradients
__group_40__new_moment_2 (optional) - T4: New averaged squared gradients
__group_40__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_41__new_weights (optional) - T2: New weights
__group_41__new_gradients (optional) - T3: New gradients
__group_41__new_moment_1 (optional) - T4: New averaged gradients
__group_41__new_moment_2 (optional) - T4: New averaged squared gradients
__group_41__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_42__new_weights (optional) - T2: New weights
__group_42__new_gradients (optional) - T3: New gradients
__group_42__new_moment_1 (optional) - T4: New averaged gradients
__group_42__new_moment_2 (optional) - T4: New averaged squared gradients
__group_42__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_43__new_weights (optional) - T2: New weights
__group_43__new_gradients (optional) - T3: New gradients
__group_43__new_moment_1 (optional) - T4: New averaged gradients
__group_43__new_moment_2 (optional) - T4: New averaged squared gradients
__group_43__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_44__new_weights (optional) - T2: New weights
__group_44__new_gradients (optional) - T3: New gradients
__group_44__new_moment_1 (optional) - T4: New averaged gradients
__group_44__new_moment_2 (optional) - T4: New averaged squared gradients
__group_44__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_45__new_weights (optional) - T2: New weights
__group_45__new_gradients (optional) - T3: New gradients
__group_45__new_moment_1 (optional) - T4: New averaged gradients
__group_45__new_moment_2 (optional) - T4: New averaged squared gradients
__group_45__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_46__new_weights (optional) - T2: New weights
__group_46__new_gradients (optional) - T3: New gradients
__group_46__new_moment_1 (optional) - T4: New averaged gradients
__group_46__new_moment_2 (optional) - T4: New averaged squared gradients
__group_46__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_47__new_weights (optional) - T2: New weights
__group_47__new_gradients (optional) - T3: New gradients
__group_47__new_moment_1 (optional) - T4: New averaged gradients
__group_47__new_moment_2 (optional) - T4: New averaged squared gradients
__group_47__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_48__new_weights (optional) - T2: New weights
__group_48__new_gradients (optional) - T3: New gradients
__group_48__new_moment_1 (optional) - T4: New averaged gradients
__group_48__new_moment_2 (optional) - T4: New averaged squared gradients
__group_48__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_49__new_weights (optional) - T2: New weights
__group_49__new_gradients (optional) - T3: New gradients
__group_49__new_moment_1 (optional) - T4: New averaged gradients
__group_49__new_moment_2 (optional) - T4: New averaged squared gradients
__group_49__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_50__new_weights (optional) - T2: New weights
__group_50__new_gradients (optional) - T3: New gradients
__group_50__new_moment_1 (optional) - T4: New averaged gradients
__group_50__new_moment_2 (optional) - T4: New averaged squared gradients
__group_50__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_51__new_weights (optional) - T2: New weights
__group_51__new_gradients (optional) - T3: New gradients
__group_51__new_moment_1 (optional) - T4: New averaged gradients
__group_51__new_moment_2 (optional) - T4: New averaged squared gradients
__group_51__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_52__new_weights (optional) - T2: New weights
__group_52__new_gradients (optional) - T3: New gradients
__group_52__new_moment_1 (optional) - T4: New averaged gradients
__group_52__new_moment_2 (optional) - T4: New averaged squared gradients
__group_52__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_53__new_weights (optional) - T2: New weights
__group_53__new_gradients (optional) - T3: New gradients
__group_53__new_moment_1 (optional) - T4: New averaged gradients
__group_53__new_moment_2 (optional) - T4: New averaged squared gradients
__group_53__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_54__new_weights (optional) - T2: New weights
__group_54__new_gradients (optional) - T3: New gradients
__group_54__new_moment_1 (optional) - T4: New averaged gradients
__group_54__new_moment_2 (optional) - T4: New averaged squared gradients
__group_54__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_55__new_weights (optional) - T2: New weights
__group_55__new_gradients (optional) - T3: New gradients
__group_55__new_moment_1 (optional) - T4: New averaged gradients
__group_55__new_moment_2 (optional) - T4: New averaged squared gradients
__group_55__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_56__new_weights (optional) - T2: New weights
__group_56__new_gradients (optional) - T3: New gradients
__group_56__new_moment_1 (optional) - T4: New averaged gradients
__group_56__new_moment_2 (optional) - T4: New averaged squared gradients
__group_56__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_57__new_weights (optional) - T2: New weights
__group_57__new_gradients (optional) - T3: New gradients
__group_57__new_moment_1 (optional) - T4: New averaged gradients
__group_57__new_moment_2 (optional) - T4: New averaged squared gradients
__group_57__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_58__new_weights (optional) - T2: New weights
__group_58__new_gradients (optional) - T3: New gradients
__group_58__new_moment_1 (optional) - T4: New averaged gradients
__group_58__new_moment_2 (optional) - T4: New averaged squared gradients
__group_58__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_59__new_weights (optional) - T2: New weights
__group_59__new_gradients (optional) - T3: New gradients
__group_59__new_moment_1 (optional) - T4: New averaged gradients
__group_59__new_moment_2 (optional) - T4: New averaged squared gradients
__group_59__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_60__new_weights (optional) - T2: New weights
__group_60__new_gradients (optional) - T3: New gradients
__group_60__new_moment_1 (optional) - T4: New averaged gradients
__group_60__new_moment_2 (optional) - T4: New averaged squared gradients
__group_60__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_61__new_weights (optional) - T2: New weights
__group_61__new_gradients (optional) - T3: New gradients
__group_61__new_moment_1 (optional) - T4: New averaged gradients
__group_61__new_moment_2 (optional) - T4: New averaged squared gradients
__group_61__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_62__new_weights (optional) - T2: New weights
__group_62__new_gradients (optional) - T3: New gradients
__group_62__new_moment_1 (optional) - T4: New averaged gradients
__group_62__new_moment_2 (optional) - T4: New averaged squared gradients
__group_62__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_63__new_weights (optional) - T2: New weights
__group_63__new_gradients (optional) - T3: New gradients
__group_63__new_moment_1 (optional) - T4: New averaged gradients
__group_63__new_moment_2 (optional) - T4: New averaged squared gradients
__group_63__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_64__new_weights (optional) - T2: New weights
__group_64__new_gradients (optional) - T3: New gradients
__group_64__new_moment_1 (optional) - T4: New averaged gradients
__group_64__new_moment_2 (optional) - T4: New averaged squared gradients
__group_64__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_65__new_weights (optional) - T2: New weights
__group_65__new_gradients (optional) - T3: New gradients
__group_65__new_moment_1 (optional) - T4: New averaged gradients
__group_65__new_moment_2 (optional) - T4: New averaged squared gradients
__group_65__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_66__new_weights (optional) - T2: New weights
__group_66__new_gradients (optional) - T3: New gradients
__group_66__new_moment_1 (optional) - T4: New averaged gradients
__group_66__new_moment_2 (optional) - T4: New averaged squared gradients
__group_66__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_67__new_weights (optional) - T2: New weights
__group_67__new_gradients (optional) - T3: New gradients
__group_67__new_moment_1 (optional) - T4: New averaged gradients
__group_67__new_moment_2 (optional) - T4: New averaged squared gradients
__group_67__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_68__new_weights (optional) - T2: New weights
__group_68__new_gradients (optional) - T3: New gradients
__group_68__new_moment_1 (optional) - T4: New averaged gradients
__group_68__new_moment_2 (optional) - T4: New averaged squared gradients
__group_68__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_69__new_weights (optional) - T2: New weights
__group_69__new_gradients (optional) - T3: New gradients
__group_69__new_moment_1 (optional) - T4: New averaged gradients
__group_69__new_moment_2 (optional) - T4: New averaged squared gradients
__group_69__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_70__new_weights (optional) - T2: New weights
__group_70__new_gradients (optional) - T3: New gradients
__group_70__new_moment_1 (optional) - T4: New averaged gradients
__group_70__new_moment_2 (optional) - T4: New averaged squared gradients
__group_70__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_71__new_weights (optional) - T2: New weights
__group_71__new_gradients (optional) - T3: New gradients
__group_71__new_moment_1 (optional) - T4: New averaged gradients
__group_71__new_moment_2 (optional) - T4: New averaged squared gradients
__group_71__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_72__new_weights (optional) - T2: New weights
__group_72__new_gradients (optional) - T3: New gradients
__group_72__new_moment_1 (optional) - T4: New averaged gradients
__group_72__new_moment_2 (optional) - T4: New averaged squared gradients
__group_72__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_73__new_weights (optional) - T2: New weights
__group_73__new_gradients (optional) - T3: New gradients
__group_73__new_moment_1 (optional) - T4: New averaged gradients
__group_73__new_moment_2 (optional) - T4: New averaged squared gradients
__group_73__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_74__new_weights (optional) - T2: New weights
__group_74__new_gradients (optional) - T3: New gradients
__group_74__new_moment_1 (optional) - T4: New averaged gradients
__group_74__new_moment_2 (optional) - T4: New averaged squared gradients
__group_74__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_75__new_weights (optional) - T2: New weights
__group_75__new_gradients (optional) - T3: New gradients
__group_75__new_moment_1 (optional) - T4: New averaged gradients
__group_75__new_moment_2 (optional) - T4: New averaged squared gradients
__group_75__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_76__new_weights (optional) - T2: New weights
__group_76__new_gradients (optional) - T3: New gradients
__group_76__new_moment_1 (optional) - T4: New averaged gradients
__group_76__new_moment_2 (optional) - T4: New averaged squared gradients
__group_76__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_77__new_weights (optional) - T2: New weights
__group_77__new_gradients (optional) - T3: New gradients
__group_77__new_moment_1 (optional) - T4: New averaged gradients
__group_77__new_moment_2 (optional) - T4: New averaged squared gradients
__group_77__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_78__new_weights (optional) - T2: New weights
__group_78__new_gradients (optional) - T3: New gradients
__group_78__new_moment_1 (optional) - T4: New averaged gradients
__group_78__new_moment_2 (optional) - T4: New averaged squared gradients
__group_78__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_79__new_weights (optional) - T2: New weights
__group_79__new_gradients (optional) - T3: New gradients
__group_79__new_moment_1 (optional) - T4: New averaged gradients
__group_79__new_moment_2 (optional) - T4: New averaged squared gradients
__group_79__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_80__new_weights (optional) - T2: New weights
__group_80__new_gradients (optional) - T3: New gradients
__group_80__new_moment_1 (optional) - T4: New averaged gradients
__group_80__new_moment_2 (optional) - T4: New averaged squared gradients
__group_80__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_81__new_weights (optional) - T2: New weights
__group_81__new_gradients (optional) - T3: New gradients
__group_81__new_moment_1 (optional) - T4: New averaged gradients
__group_81__new_moment_2 (optional) - T4: New averaged squared gradients
__group_81__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_82__new_weights (optional) - T2: New weights
__group_82__new_gradients (optional) - T3: New gradients
__group_82__new_moment_1 (optional) - T4: New averaged gradients
__group_82__new_moment_2 (optional) - T4: New averaged squared gradients
__group_82__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_83__new_weights (optional) - T2: New weights
__group_83__new_gradients (optional) - T3: New gradients
__group_83__new_moment_1 (optional) - T4: New averaged gradients
__group_83__new_moment_2 (optional) - T4: New averaged squared gradients
__group_83__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_84__new_weights (optional) - T2: New weights
__group_84__new_gradients (optional) - T3: New gradients
__group_84__new_moment_1 (optional) - T4: New averaged gradients
__group_84__new_moment_2 (optional) - T4: New averaged squared gradients
__group_84__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_85__new_weights (optional) - T2: New weights
__group_85__new_gradients (optional) - T3: New gradients
__group_85__new_moment_1 (optional) - T4: New averaged gradients
__group_85__new_moment_2 (optional) - T4: New averaged squared gradients
__group_85__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_86__new_weights (optional) - T2: New weights
__group_86__new_gradients (optional) - T3: New gradients
__group_86__new_moment_1 (optional) - T4: New averaged gradients
__group_86__new_moment_2 (optional) - T4: New averaged squared gradients
__group_86__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_87__new_weights (optional) - T2: New weights
__group_87__new_gradients (optional) - T3: New gradients
__group_87__new_moment_1 (optional) - T4: New averaged gradients
__group_87__new_moment_2 (optional) - T4: New averaged squared gradients
__group_87__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_88__new_weights (optional) - T2: New weights
__group_88__new_gradients (optional) - T3: New gradients
__group_88__new_moment_1 (optional) - T4: New averaged gradients
__group_88__new_moment_2 (optional) - T4: New averaged squared gradients
__group_88__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_89__new_weights (optional) - T2: New weights
__group_89__new_gradients (optional) - T3: New gradients
__group_89__new_moment_1 (optional) - T4: New averaged gradients
__group_89__new_moment_2 (optional) - T4: New averaged squared gradients
__group_89__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_90__new_weights (optional) - T2: New weights
__group_90__new_gradients (optional) - T3: New gradients
__group_90__new_moment_1 (optional) - T4: New averaged gradients
__group_90__new_moment_2 (optional) - T4: New averaged squared gradients
__group_90__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_91__new_weights (optional) - T2: New weights
__group_91__new_gradients (optional) - T3: New gradients
__group_91__new_moment_1 (optional) - T4: New averaged gradients
__group_91__new_moment_2 (optional) - T4: New averaged squared gradients
__group_91__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_92__new_weights (optional) - T2: New weights
__group_92__new_gradients (optional) - T3: New gradients
__group_92__new_moment_1 (optional) - T4: New averaged gradients
__group_92__new_moment_2 (optional) - T4: New averaged squared gradients
__group_92__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_93__new_weights (optional) - T2: New weights
__group_93__new_gradients (optional) - T3: New gradients
__group_93__new_moment_1 (optional) - T4: New averaged gradients
__group_93__new_moment_2 (optional) - T4: New averaged squared gradients
__group_93__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_94__new_weights (optional) - T2: New weights
__group_94__new_gradients (optional) - T3: New gradients
__group_94__new_moment_1 (optional) - T4: New averaged gradients
__group_94__new_moment_2 (optional) - T4: New averaged squared gradients
__group_94__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_95__new_weights (optional) - T2: New weights
__group_95__new_gradients (optional) - T3: New gradients
__group_95__new_moment_1 (optional) - T4: New averaged gradients
__group_95__new_moment_2 (optional) - T4: New averaged squared gradients
__group_95__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_96__new_weights (optional) - T2: New weights
__group_96__new_gradients (optional) - T3: New gradients
__group_96__new_moment_1 (optional) - T4: New averaged gradients
__group_96__new_moment_2 (optional) - T4: New averaged squared gradients
__group_96__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_97__new_weights (optional) - T2: New weights
__group_97__new_gradients (optional) - T3: New gradients
__group_97__new_moment_1 (optional) - T4: New averaged gradients
__group_97__new_moment_2 (optional) - T4: New averaged squared gradients
__group_97__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_98__new_weights (optional) - T2: New weights
__group_98__new_gradients (optional) - T3: New gradients
__group_98__new_moment_1 (optional) - T4: New averaged gradients
__group_98__new_moment_2 (optional) - T4: New averaged squared gradients
__group_98__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_99__new_weights (optional) - T2: New weights
__group_99__new_gradients (optional) - T3: New gradients
__group_99__new_moment_1 (optional) - T4: New averaged gradients
__group_99__new_moment_2 (optional) - T4: New averaged squared gradients
__group_99__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_100__new_weights (optional) - T2: New weights
__group_100__new_gradients (optional) - T3: New gradients
__group_100__new_moment_1 (optional) - T4: New averaged gradients
__group_100__new_moment_2 (optional) - T4: New averaged squared gradients
__group_100__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_101__new_weights (optional) - T2: New weights
__group_101__new_gradients (optional) - T3: New gradients
__group_101__new_moment_1 (optional) - T4: New averaged gradients
__group_101__new_moment_2 (optional) - T4: New averaged squared gradients
__group_101__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_102__new_weights (optional) - T2: New weights
__group_102__new_gradients (optional) - T3: New gradients
__group_102__new_moment_1 (optional) - T4: New averaged gradients
__group_102__new_moment_2 (optional) - T4: New averaged squared gradients
__group_102__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_103__new_weights (optional) - T2: New weights
__group_103__new_gradients (optional) - T3: New gradients
__group_103__new_moment_1 (optional) - T4: New averaged gradients
__group_103__new_moment_2 (optional) - T4: New averaged squared gradients
__group_103__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_104__new_weights (optional) - T2: New weights
__group_104__new_gradients (optional) - T3: New gradients
__group_104__new_moment_1 (optional) - T4: New averaged gradients
__group_104__new_moment_2 (optional) - T4: New averaged squared gradients
__group_104__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_105__new_weights (optional) - T2: New weights
__group_105__new_gradients (optional) - T3: New gradients
__group_105__new_moment_1 (optional) - T4: New averaged gradients
__group_105__new_moment_2 (optional) - T4: New averaged squared gradients
__group_105__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_106__new_weights (optional) - T2: New weights
__group_106__new_gradients (optional) - T3: New gradients
__group_106__new_moment_1 (optional) - T4: New averaged gradients
__group_106__new_moment_2 (optional) - T4: New averaged squared gradients
__group_106__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_107__new_weights (optional) - T2: New weights
__group_107__new_gradients (optional) - T3: New gradients
__group_107__new_moment_1 (optional) - T4: New averaged gradients
__group_107__new_moment_2 (optional) - T4: New averaged squared gradients
__group_107__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_108__new_weights (optional) - T2: New weights
__group_108__new_gradients (optional) - T3: New gradients
__group_108__new_moment_1 (optional) - T4: New averaged gradients
__group_108__new_moment_2 (optional) - T4: New averaged squared gradients
__group_108__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_109__new_weights (optional) - T2: New weights
__group_109__new_gradients (optional) - T3: New gradients
__group_109__new_moment_1 (optional) - T4: New averaged gradients
__group_109__new_moment_2 (optional) - T4: New averaged squared gradients
__group_109__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_110__new_weights (optional) - T2: New weights
__group_110__new_gradients (optional) - T3: New gradients
__group_110__new_moment_1 (optional) - T4: New averaged gradients
__group_110__new_moment_2 (optional) - T4: New averaged squared gradients
__group_110__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_111__new_weights (optional) - T2: New weights
__group_111__new_gradients (optional) - T3: New gradients
__group_111__new_moment_1 (optional) - T4: New averaged gradients
__group_111__new_moment_2 (optional) - T4: New averaged squared gradients
__group_111__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_112__new_weights (optional) - T2: New weights
__group_112__new_gradients (optional) - T3: New gradients
__group_112__new_moment_1 (optional) - T4: New averaged gradients
__group_112__new_moment_2 (optional) - T4: New averaged squared gradients
__group_112__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_113__new_weights (optional) - T2: New weights
__group_113__new_gradients (optional) - T3: New gradients
__group_113__new_moment_1 (optional) - T4: New averaged gradients
__group_113__new_moment_2 (optional) - T4: New averaged squared gradients
__group_113__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_114__new_weights (optional) - T2: New weights
__group_114__new_gradients (optional) - T3: New gradients
__group_114__new_moment_1 (optional) - T4: New averaged gradients
__group_114__new_moment_2 (optional) - T4: New averaged squared gradients
__group_114__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_115__new_weights (optional) - T2: New weights
__group_115__new_gradients (optional) - T3: New gradients
__group_115__new_moment_1 (optional) - T4: New averaged gradients
__group_115__new_moment_2 (optional) - T4: New averaged squared gradients
__group_115__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_116__new_weights (optional) - T2: New weights
__group_116__new_gradients (optional) - T3: New gradients
__group_116__new_moment_1 (optional) - T4: New averaged gradients
__group_116__new_moment_2 (optional) - T4: New averaged squared gradients
__group_116__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_117__new_weights (optional) - T2: New weights
__group_117__new_gradients (optional) - T3: New gradients
__group_117__new_moment_1 (optional) - T4: New averaged gradients
__group_117__new_moment_2 (optional) - T4: New averaged squared gradients
__group_117__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_118__new_weights (optional) - T2: New weights
__group_118__new_gradients (optional) - T3: New gradients
__group_118__new_moment_1 (optional) - T4: New averaged gradients
__group_118__new_moment_2 (optional) - T4: New averaged squared gradients
__group_118__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_119__new_weights (optional) - T2: New weights
__group_119__new_gradients (optional) - T3: New gradients
__group_119__new_moment_1 (optional) - T4: New averaged gradients
__group_119__new_moment_2 (optional) - T4: New averaged squared gradients
__group_119__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_120__new_weights (optional) - T2: New weights
__group_120__new_gradients (optional) - T3: New gradients
__group_120__new_moment_1 (optional) - T4: New averaged gradients
__group_120__new_moment_2 (optional) - T4: New averaged squared gradients
__group_120__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_121__new_weights (optional) - T2: New weights
__group_121__new_gradients (optional) - T3: New gradients
__group_121__new_moment_1 (optional) - T4: New averaged gradients
__group_121__new_moment_2 (optional) - T4: New averaged squared gradients
__group_121__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_122__new_weights (optional) - T2: New weights
__group_122__new_gradients (optional) - T3: New gradients
__group_122__new_moment_1 (optional) - T4: New averaged gradients
__group_122__new_moment_2 (optional) - T4: New averaged squared gradients
__group_122__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_123__new_weights (optional) - T2: New weights
__group_123__new_gradients (optional) - T3: New gradients
__group_123__new_moment_1 (optional) - T4: New averaged gradients
__group_123__new_moment_2 (optional) - T4: New averaged squared gradients
__group_123__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_124__new_weights (optional) - T2: New weights
__group_124__new_gradients (optional) - T3: New gradients
__group_124__new_moment_1 (optional) - T4: New averaged gradients
__group_124__new_moment_2 (optional) - T4: New averaged squared gradients
__group_124__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_125__new_weights (optional) - T2: New weights
__group_125__new_gradients (optional) - T3: New gradients
__group_125__new_moment_1 (optional) - T4: New averaged gradients
__group_125__new_moment_2 (optional) - T4: New averaged squared gradients
__group_125__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_126__new_weights (optional) - T2: New weights
__group_126__new_gradients (optional) - T3: New gradients
__group_126__new_moment_1 (optional) - T4: New averaged gradients
__group_126__new_moment_2 (optional) - T4: New averaged squared gradients
__group_126__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_127__new_weights (optional) - T2: New weights
__group_127__new_gradients (optional) - T3: New gradients
__group_127__new_moment_1 (optional) - T4: New averaged gradients
__group_127__new_moment_2 (optional) - T4: New averaged squared gradients
__group_127__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_128__new_weights (optional) - T2: New weights
__group_128__new_gradients (optional) - T3: New gradients
__group_128__new_moment_1 (optional) - T4: New averaged gradients
__group_128__new_moment_2 (optional) - T4: New averaged squared gradients
__group_128__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_129__new_weights (optional) - T2: New weights
__group_129__new_gradients (optional) - T3: New gradients
__group_129__new_moment_1 (optional) - T4: New averaged gradients
__group_129__new_moment_2 (optional) - T4: New averaged squared gradients
__group_129__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_130__new_weights (optional) - T2: New weights
__group_130__new_gradients (optional) - T3: New gradients
__group_130__new_moment_1 (optional) - T4: New averaged gradients
__group_130__new_moment_2 (optional) - T4: New averaged squared gradients
__group_130__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_131__new_weights (optional) - T2: New weights
__group_131__new_gradients (optional) - T3: New gradients
__group_131__new_moment_1 (optional) - T4: New averaged gradients
__group_131__new_moment_2 (optional) - T4: New averaged squared gradients
__group_131__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_132__new_weights (optional) - T2: New weights
__group_132__new_gradients (optional) - T3: New gradients
__group_132__new_moment_1 (optional) - T4: New averaged gradients
__group_132__new_moment_2 (optional) - T4: New averaged squared gradients
__group_132__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_133__new_weights (optional) - T2: New weights
__group_133__new_gradients (optional) - T3: New gradients
__group_133__new_moment_1 (optional) - T4: New averaged gradients
__group_133__new_moment_2 (optional) - T4: New averaged squared gradients
__group_133__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_134__new_weights (optional) - T2: New weights
__group_134__new_gradients (optional) - T3: New gradients
__group_134__new_moment_1 (optional) - T4: New averaged gradients
__group_134__new_moment_2 (optional) - T4: New averaged squared gradients
__group_134__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_135__new_weights (optional) - T2: New weights
__group_135__new_gradients (optional) - T3: New gradients
__group_135__new_moment_1 (optional) - T4: New averaged gradients
__group_135__new_moment_2 (optional) - T4: New averaged squared gradients
__group_135__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_136__new_weights (optional) - T2: New weights
__group_136__new_gradients (optional) - T3: New gradients
__group_136__new_moment_1 (optional) - T4: New averaged gradients
__group_136__new_moment_2 (optional) - T4: New averaged squared gradients
__group_136__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_137__new_weights (optional) - T2: New weights
__group_137__new_gradients (optional) - T3: New gradients
__group_137__new_moment_1 (optional) - T4: New averaged gradients
__group_137__new_moment_2 (optional) - T4: New averaged squared gradients
__group_137__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_138__new_weights (optional) - T2: New weights
__group_138__new_gradients (optional) - T3: New gradients
__group_138__new_moment_1 (optional) - T4: New averaged gradients
__group_138__new_moment_2 (optional) - T4: New averaged squared gradients
__group_138__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_139__new_weights (optional) - T2: New weights
__group_139__new_gradients (optional) - T3: New gradients
__group_139__new_moment_1 (optional) - T4: New averaged gradients
__group_139__new_moment_2 (optional) - T4: New averaged squared gradients
__group_139__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_140__new_weights (optional) - T2: New weights
__group_140__new_gradients (optional) - T3: New gradients
__group_140__new_moment_1 (optional) - T4: New averaged gradients
__group_140__new_moment_2 (optional) - T4: New averaged squared gradients
__group_140__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_141__new_weights (optional) - T2: New weights
__group_141__new_gradients (optional) - T3: New gradients
__group_141__new_moment_1 (optional) - T4: New averaged gradients
__group_141__new_moment_2 (optional) - T4: New averaged squared gradients
__group_141__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_142__new_weights (optional) - T2: New weights
__group_142__new_gradients (optional) - T3: New gradients
__group_142__new_moment_1 (optional) - T4: New averaged gradients
__group_142__new_moment_2 (optional) - T4: New averaged squared gradients
__group_142__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_143__new_weights (optional) - T2: New weights
__group_143__new_gradients (optional) - T3: New gradients
__group_143__new_moment_1 (optional) - T4: New averaged gradients
__group_143__new_moment_2 (optional) - T4: New averaged squared gradients
__group_143__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_144__new_weights (optional) - T2: New weights
__group_144__new_gradients (optional) - T3: New gradients
__group_144__new_moment_1 (optional) - T4: New averaged gradients
__group_144__new_moment_2 (optional) - T4: New averaged squared gradients
__group_144__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_145__new_weights (optional) - T2: New weights
__group_145__new_gradients (optional) - T3: New gradients
__group_145__new_moment_1 (optional) - T4: New averaged gradients
__group_145__new_moment_2 (optional) - T4: New averaged squared gradients
__group_145__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_146__new_weights (optional) - T2: New weights
__group_146__new_gradients (optional) - T3: New gradients
__group_146__new_moment_1 (optional) - T4: New averaged gradients
__group_146__new_moment_2 (optional) - T4: New averaged squared gradients
__group_146__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_147__new_weights (optional) - T2: New weights
__group_147__new_gradients (optional) - T3: New gradients
__group_147__new_moment_1 (optional) - T4: New averaged gradients
__group_147__new_moment_2 (optional) - T4: New averaged squared gradients
__group_147__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_148__new_weights (optional) - T2: New weights
__group_148__new_gradients (optional) - T3: New gradients
__group_148__new_moment_1 (optional) - T4: New averaged gradients
__group_148__new_moment_2 (optional) - T4: New averaged squared gradients
__group_148__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_149__new_weights (optional) - T2: New weights
__group_149__new_gradients (optional) - T3: New gradients
__group_149__new_moment_1 (optional) - T4: New averaged gradients
__group_149__new_moment_2 (optional) - T4: New averaged squared gradients
__group_149__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_150__new_weights (optional) - T2: New weights
__group_150__new_gradients (optional) - T3: New gradients
__group_150__new_moment_1 (optional) - T4: New averaged gradients
__group_150__new_moment_2 (optional) - T4: New averaged squared gradients
__group_150__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_151__new_weights (optional) - T2: New weights
__group_151__new_gradients (optional) - T3: New gradients
__group_151__new_moment_1 (optional) - T4: New averaged gradients
__group_151__new_moment_2 (optional) - T4: New averaged squared gradients
__group_151__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_152__new_weights (optional) - T2: New weights
__group_152__new_gradients (optional) - T3: New gradients
__group_152__new_moment_1 (optional) - T4: New averaged gradients
__group_152__new_moment_2 (optional) - T4: New averaged squared gradients
__group_152__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_153__new_weights (optional) - T2: New weights
__group_153__new_gradients (optional) - T3: New gradients
__group_153__new_moment_1 (optional) - T4: New averaged gradients
__group_153__new_moment_2 (optional) - T4: New averaged squared gradients
__group_153__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_154__new_weights (optional) - T2: New weights
__group_154__new_gradients (optional) - T3: New gradients
__group_154__new_moment_1 (optional) - T4: New averaged gradients
__group_154__new_moment_2 (optional) - T4: New averaged squared gradients
__group_154__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_155__new_weights (optional) - T2: New weights
__group_155__new_gradients (optional) - T3: New gradients
__group_155__new_moment_1 (optional) - T4: New averaged gradients
__group_155__new_moment_2 (optional) - T4: New averaged squared gradients
__group_155__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_156__new_weights (optional) - T2: New weights
__group_156__new_gradients (optional) - T3: New gradients
__group_156__new_moment_1 (optional) - T4: New averaged gradients
__group_156__new_moment_2 (optional) - T4: New averaged squared gradients
__group_156__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_157__new_weights (optional) - T2: New weights
__group_157__new_gradients (optional) - T3: New gradients
__group_157__new_moment_1 (optional) - T4: New averaged gradients
__group_157__new_moment_2 (optional) - T4: New averaged squared gradients
__group_157__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_158__new_weights (optional) - T2: New weights
__group_158__new_gradients (optional) - T3: New gradients
__group_158__new_moment_1 (optional) - T4: New averaged gradients
__group_158__new_moment_2 (optional) - T4: New averaged squared gradients
__group_158__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_159__new_weights (optional) - T2: New weights
__group_159__new_gradients (optional) - T3: New gradients
__group_159__new_moment_1 (optional) - T4: New averaged gradients
__group_159__new_moment_2 (optional) - T4: New averaged squared gradients
__group_159__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_160__new_weights (optional) - T2: New weights
__group_160__new_gradients (optional) - T3: New gradients
__group_160__new_moment_1 (optional) - T4: New averaged gradients
__group_160__new_moment_2 (optional) - T4: New averaged squared gradients
__group_160__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_161__new_weights (optional) - T2: New weights
__group_161__new_gradients (optional) - T3: New gradients
__group_161__new_moment_1 (optional) - T4: New averaged gradients
__group_161__new_moment_2 (optional) - T4: New averaged squared gradients
__group_161__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_162__new_weights (optional) - T2: New weights
__group_162__new_gradients (optional) - T3: New gradients
__group_162__new_moment_1 (optional) - T4: New averaged gradients
__group_162__new_moment_2 (optional) - T4: New averaged squared gradients
__group_162__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_163__new_weights (optional) - T2: New weights
__group_163__new_gradients (optional) - T3: New gradients
__group_163__new_moment_1 (optional) - T4: New averaged gradients
__group_163__new_moment_2 (optional) - T4: New averaged squared gradients
__group_163__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_164__new_weights (optional) - T2: New weights
__group_164__new_gradients (optional) - T3: New gradients
__group_164__new_moment_1 (optional) - T4: New averaged gradients
__group_164__new_moment_2 (optional) - T4: New averaged squared gradients
__group_164__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_165__new_weights (optional) - T2: New weights
__group_165__new_gradients (optional) - T3: New gradients
__group_165__new_moment_1 (optional) - T4: New averaged gradients
__group_165__new_moment_2 (optional) - T4: New averaged squared gradients
__group_165__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_166__new_weights (optional) - T2: New weights
__group_166__new_gradients (optional) - T3: New gradients
__group_166__new_moment_1 (optional) - T4: New averaged gradients
__group_166__new_moment_2 (optional) - T4: New averaged squared gradients
__group_166__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_167__new_weights (optional) - T2: New weights
__group_167__new_gradients (optional) - T3: New gradients
__group_167__new_moment_1 (optional) - T4: New averaged gradients
__group_167__new_moment_2 (optional) - T4: New averaged squared gradients
__group_167__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_168__new_weights (optional) - T2: New weights
__group_168__new_gradients (optional) - T3: New gradients
__group_168__new_moment_1 (optional) - T4: New averaged gradients
__group_168__new_moment_2 (optional) - T4: New averaged squared gradients
__group_168__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_169__new_weights (optional) - T2: New weights
__group_169__new_gradients (optional) - T3: New gradients
__group_169__new_moment_1 (optional) - T4: New averaged gradients
__group_169__new_moment_2 (optional) - T4: New averaged squared gradients
__group_169__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_170__new_weights (optional) - T2: New weights
__group_170__new_gradients (optional) - T3: New gradients
__group_170__new_moment_1 (optional) - T4: New averaged gradients
__group_170__new_moment_2 (optional) - T4: New averaged squared gradients
__group_170__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_171__new_weights (optional) - T2: New weights
__group_171__new_gradients (optional) - T3: New gradients
__group_171__new_moment_1 (optional) - T4: New averaged gradients
__group_171__new_moment_2 (optional) - T4: New averaged squared gradients
__group_171__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_172__new_weights (optional) - T2: New weights
__group_172__new_gradients (optional) - T3: New gradients
__group_172__new_moment_1 (optional) - T4: New averaged gradients
__group_172__new_moment_2 (optional) - T4: New averaged squared gradients
__group_172__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_173__new_weights (optional) - T2: New weights
__group_173__new_gradients (optional) - T3: New gradients
__group_173__new_moment_1 (optional) - T4: New averaged gradients
__group_173__new_moment_2 (optional) - T4: New averaged squared gradients
__group_173__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_174__new_weights (optional) - T2: New weights
__group_174__new_gradients (optional) - T3: New gradients
__group_174__new_moment_1 (optional) - T4: New averaged gradients
__group_174__new_moment_2 (optional) - T4: New averaged squared gradients
__group_174__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_175__new_weights (optional) - T2: New weights
__group_175__new_gradients (optional) - T3: New gradients
__group_175__new_moment_1 (optional) - T4: New averaged gradients
__group_175__new_moment_2 (optional) - T4: New averaged squared gradients
__group_175__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_176__new_weights (optional) - T2: New weights
__group_176__new_gradients (optional) - T3: New gradients
__group_176__new_moment_1 (optional) - T4: New averaged gradients
__group_176__new_moment_2 (optional) - T4: New averaged squared gradients
__group_176__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_177__new_weights (optional) - T2: New weights
__group_177__new_gradients (optional) - T3: New gradients
__group_177__new_moment_1 (optional) - T4: New averaged gradients
__group_177__new_moment_2 (optional) - T4: New averaged squared gradients
__group_177__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_178__new_weights (optional) - T2: New weights
__group_178__new_gradients (optional) - T3: New gradients
__group_178__new_moment_1 (optional) - T4: New averaged gradients
__group_178__new_moment_2 (optional) - T4: New averaged squared gradients
__group_178__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_179__new_weights (optional) - T2: New weights
__group_179__new_gradients (optional) - T3: New gradients
__group_179__new_moment_1 (optional) - T4: New averaged gradients
__group_179__new_moment_2 (optional) - T4: New averaged squared gradients
__group_179__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_180__new_weights (optional) - T2: New weights
__group_180__new_gradients (optional) - T3: New gradients
__group_180__new_moment_1 (optional) - T4: New averaged gradients
__group_180__new_moment_2 (optional) - T4: New averaged squared gradients
__group_180__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_181__new_weights (optional) - T2: New weights
__group_181__new_gradients (optional) - T3: New gradients
__group_181__new_moment_1 (optional) - T4: New averaged gradients
__group_181__new_moment_2 (optional) - T4: New averaged squared gradients
__group_181__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_182__new_weights (optional) - T2: New weights
__group_182__new_gradients (optional) - T3: New gradients
__group_182__new_moment_1 (optional) - T4: New averaged gradients
__group_182__new_moment_2 (optional) - T4: New averaged squared gradients
__group_182__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_183__new_weights (optional) - T2: New weights
__group_183__new_gradients (optional) - T3: New gradients
__group_183__new_moment_1 (optional) - T4: New averaged gradients
__group_183__new_moment_2 (optional) - T4: New averaged squared gradients
__group_183__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_184__new_weights (optional) - T2: New weights
__group_184__new_gradients (optional) - T3: New gradients
__group_184__new_moment_1 (optional) - T4: New averaged gradients
__group_184__new_moment_2 (optional) - T4: New averaged squared gradients
__group_184__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_185__new_weights (optional) - T2: New weights
__group_185__new_gradients (optional) - T3: New gradients
__group_185__new_moment_1 (optional) - T4: New averaged gradients
__group_185__new_moment_2 (optional) - T4: New averaged squared gradients
__group_185__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_186__new_weights (optional) - T2: New weights
__group_186__new_gradients (optional) - T3: New gradients
__group_186__new_moment_1 (optional) - T4: New averaged gradients
__group_186__new_moment_2 (optional) - T4: New averaged squared gradients
__group_186__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_187__new_weights (optional) - T2: New weights
__group_187__new_gradients (optional) - T3: New gradients
__group_187__new_moment_1 (optional) - T4: New averaged gradients
__group_187__new_moment_2 (optional) - T4: New averaged squared gradients
__group_187__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_188__new_weights (optional) - T2: New weights
__group_188__new_gradients (optional) - T3: New gradients
__group_188__new_moment_1 (optional) - T4: New averaged gradients
__group_188__new_moment_2 (optional) - T4: New averaged squared gradients
__group_188__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_189__new_weights (optional) - T2: New weights
__group_189__new_gradients (optional) - T3: New gradients
__group_189__new_moment_1 (optional) - T4: New averaged gradients
__group_189__new_moment_2 (optional) - T4: New averaged squared gradients
__group_189__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_190__new_weights (optional) - T2: New weights
__group_190__new_gradients (optional) - T3: New gradients
__group_190__new_moment_1 (optional) - T4: New averaged gradients
__group_190__new_moment_2 (optional) - T4: New averaged squared gradients
__group_190__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_191__new_weights (optional) - T2: New weights
__group_191__new_gradients (optional) - T3: New gradients
__group_191__new_moment_1 (optional) - T4: New averaged gradients
__group_191__new_moment_2 (optional) - T4: New averaged squared gradients
__group_191__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_192__new_weights (optional) - T2: New weights
__group_192__new_gradients (optional) - T3: New gradients
__group_192__new_moment_1 (optional) - T4: New averaged gradients
__group_192__new_moment_2 (optional) - T4: New averaged squared gradients
__group_192__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_193__new_weights (optional) - T2: New weights
__group_193__new_gradients (optional) - T3: New gradients
__group_193__new_moment_1 (optional) - T4: New averaged gradients
__group_193__new_moment_2 (optional) - T4: New averaged squared gradients
__group_193__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_194__new_weights (optional) - T2: New weights
__group_194__new_gradients (optional) - T3: New gradients
__group_194__new_moment_1 (optional) - T4: New averaged gradients
__group_194__new_moment_2 (optional) - T4: New averaged squared gradients
__group_194__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_195__new_weights (optional) - T2: New weights
__group_195__new_gradients (optional) - T3: New gradients
__group_195__new_moment_1 (optional) - T4: New averaged gradients
__group_195__new_moment_2 (optional) - T4: New averaged squared gradients
__group_195__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_196__new_weights (optional) - T2: New weights
__group_196__new_gradients (optional) - T3: New gradients
__group_196__new_moment_1 (optional) - T4: New averaged gradients
__group_196__new_moment_2 (optional) - T4: New averaged squared gradients
__group_196__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_197__new_weights (optional) - T2: New weights
__group_197__new_gradients (optional) - T3: New gradients
__group_197__new_moment_1 (optional) - T4: New averaged gradients
__group_197__new_moment_2 (optional) - T4: New averaged squared gradients
__group_197__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_198__new_weights (optional) - T2: New weights
__group_198__new_gradients (optional) - T3: New gradients
__group_198__new_moment_1 (optional) - T4: New averaged gradients
__group_198__new_moment_2 (optional) - T4: New averaged squared gradients
__group_198__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_199__new_weights (optional) - T2: New weights
__group_199__new_gradients (optional) - T3: New gradients
__group_199__new_moment_1 (optional) - T4: New averaged gradients
__group_199__new_moment_2 (optional) - T4: New averaged squared gradients
__group_199__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_200__new_weights (optional) - T2: New weights
__group_200__new_gradients (optional) - T3: New gradients
__group_200__new_moment_1 (optional) - T4: New averaged gradients
__group_200__new_moment_2 (optional) - T4: New averaged squared gradients
__group_200__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_201__new_weights (optional) - T2: New weights
__group_201__new_gradients (optional) - T3: New gradients
__group_201__new_moment_1 (optional) - T4: New averaged gradients
__group_201__new_moment_2 (optional) - T4: New averaged squared gradients
__group_201__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_202__new_weights (optional) - T2: New weights
__group_202__new_gradients (optional) - T3: New gradients
__group_202__new_moment_1 (optional) - T4: New averaged gradients
__group_202__new_moment_2 (optional) - T4: New averaged squared gradients
__group_202__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_203__new_weights (optional) - T2: New weights
__group_203__new_gradients (optional) - T3: New gradients
__group_203__new_moment_1 (optional) - T4: New averaged gradients
__group_203__new_moment_2 (optional) - T4: New averaged squared gradients
__group_203__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_204__new_weights (optional) - T2: New weights
__group_204__new_gradients (optional) - T3: New gradients
__group_204__new_moment_1 (optional) - T4: New averaged gradients
__group_204__new_moment_2 (optional) - T4: New averaged squared gradients
__group_204__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_205__new_weights (optional) - T2: New weights
__group_205__new_gradients (optional) - T3: New gradients
__group_205__new_moment_1 (optional) - T4: New averaged gradients
__group_205__new_moment_2 (optional) - T4: New averaged squared gradients
__group_205__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_206__new_weights (optional) - T2: New weights
__group_206__new_gradients (optional) - T3: New gradients
__group_206__new_moment_1 (optional) - T4: New averaged gradients
__group_206__new_moment_2 (optional) - T4: New averaged squared gradients
__group_206__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_207__new_weights (optional) - T2: New weights
__group_207__new_gradients (optional) - T3: New gradients
__group_207__new_moment_1 (optional) - T4: New averaged gradients
__group_207__new_moment_2 (optional) - T4: New averaged squared gradients
__group_207__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_208__new_weights (optional) - T2: New weights
__group_208__new_gradients (optional) - T3: New gradients
__group_208__new_moment_1 (optional) - T4: New averaged gradients
__group_208__new_moment_2 (optional) - T4: New averaged squared gradients
__group_208__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_209__new_weights (optional) - T2: New weights
__group_209__new_gradients (optional) - T3: New gradients
__group_209__new_moment_1 (optional) - T4: New averaged gradients
__group_209__new_moment_2 (optional) - T4: New averaged squared gradients
__group_209__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_210__new_weights (optional) - T2: New weights
__group_210__new_gradients (optional) - T3: New gradients
__group_210__new_moment_1 (optional) - T4: New averaged gradients
__group_210__new_moment_2 (optional) - T4: New averaged squared gradients
__group_210__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_211__new_weights (optional) - T2: New weights
__group_211__new_gradients (optional) - T3: New gradients
__group_211__new_moment_1 (optional) - T4: New averaged gradients
__group_211__new_moment_2 (optional) - T4: New averaged squared gradients
__group_211__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_212__new_weights (optional) - T2: New weights
__group_212__new_gradients (optional) - T3: New gradients
__group_212__new_moment_1 (optional) - T4: New averaged gradients
__group_212__new_moment_2 (optional) - T4: New averaged squared gradients
__group_212__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_213__new_weights (optional) - T2: New weights
__group_213__new_gradients (optional) - T3: New gradients
__group_213__new_moment_1 (optional) - T4: New averaged gradients
__group_213__new_moment_2 (optional) - T4: New averaged squared gradients
__group_213__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_214__new_weights (optional) - T2: New weights
__group_214__new_gradients (optional) - T3: New gradients
__group_214__new_moment_1 (optional) - T4: New averaged gradients
__group_214__new_moment_2 (optional) - T4: New averaged squared gradients
__group_214__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_215__new_weights (optional) - T2: New weights
__group_215__new_gradients (optional) - T3: New gradients
__group_215__new_moment_1 (optional) - T4: New averaged gradients
__group_215__new_moment_2 (optional) - T4: New averaged squared gradients
__group_215__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_216__new_weights (optional) - T2: New weights
__group_216__new_gradients (optional) - T3: New gradients
__group_216__new_moment_1 (optional) - T4: New averaged gradients
__group_216__new_moment_2 (optional) - T4: New averaged squared gradients
__group_216__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_217__new_weights (optional) - T2: New weights
__group_217__new_gradients (optional) - T3: New gradients
__group_217__new_moment_1 (optional) - T4: New averaged gradients
__group_217__new_moment_2 (optional) - T4: New averaged squared gradients
__group_217__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_218__new_weights (optional) - T2: New weights
__group_218__new_gradients (optional) - T3: New gradients
__group_218__new_moment_1 (optional) - T4: New averaged gradients
__group_218__new_moment_2 (optional) - T4: New averaged squared gradients
__group_218__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_219__new_weights (optional) - T2: New weights
__group_219__new_gradients (optional) - T3: New gradients
__group_219__new_moment_1 (optional) - T4: New averaged gradients
__group_219__new_moment_2 (optional) - T4: New averaged squared gradients
__group_219__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_220__new_weights (optional) - T2: New weights
__group_220__new_gradients (optional) - T3: New gradients
__group_220__new_moment_1 (optional) - T4: New averaged gradients
__group_220__new_moment_2 (optional) - T4: New averaged squared gradients
__group_220__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_221__new_weights (optional) - T2: New weights
__group_221__new_gradients (optional) - T3: New gradients
__group_221__new_moment_1 (optional) - T4: New averaged gradients
__group_221__new_moment_2 (optional) - T4: New averaged squared gradients
__group_221__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_222__new_weights (optional) - T2: New weights
__group_222__new_gradients (optional) - T3: New gradients
__group_222__new_moment_1 (optional) - T4: New averaged gradients
__group_222__new_moment_2 (optional) - T4: New averaged squared gradients
__group_222__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_223__new_weights (optional) - T2: New weights
__group_223__new_gradients (optional) - T3: New gradients
__group_223__new_moment_1 (optional) - T4: New averaged gradients
__group_223__new_moment_2 (optional) - T4: New averaged squared gradients
__group_223__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_224__new_weights (optional) - T2: New weights
__group_224__new_gradients (optional) - T3: New gradients
__group_224__new_moment_1 (optional) - T4: New averaged gradients
__group_224__new_moment_2 (optional) - T4: New averaged squared gradients
__group_224__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_225__new_weights (optional) - T2: New weights
__group_225__new_gradients (optional) - T3: New gradients
__group_225__new_moment_1 (optional) - T4: New averaged gradients
__group_225__new_moment_2 (optional) - T4: New averaged squared gradients
__group_225__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_226__new_weights (optional) - T2: New weights
__group_226__new_gradients (optional) - T3: New gradients
__group_226__new_moment_1 (optional) - T4: New averaged gradients
__group_226__new_moment_2 (optional) - T4: New averaged squared gradients
__group_226__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_227__new_weights (optional) - T2: New weights
__group_227__new_gradients (optional) - T3: New gradients
__group_227__new_moment_1 (optional) - T4: New averaged gradients
__group_227__new_moment_2 (optional) - T4: New averaged squared gradients
__group_227__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_228__new_weights (optional) - T2: New weights
__group_228__new_gradients (optional) - T3: New gradients
__group_228__new_moment_1 (optional) - T4: New averaged gradients
__group_228__new_moment_2 (optional) - T4: New averaged squared gradients
__group_228__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_229__new_weights (optional) - T2: New weights
__group_229__new_gradients (optional) - T3: New gradients
__group_229__new_moment_1 (optional) - T4: New averaged gradients
__group_229__new_moment_2 (optional) - T4: New averaged squared gradients
__group_229__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_230__new_weights (optional) - T2: New weights
__group_230__new_gradients (optional) - T3: New gradients
__group_230__new_moment_1 (optional) - T4: New averaged gradients
__group_230__new_moment_2 (optional) - T4: New averaged squared gradients
__group_230__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_231__new_weights (optional) - T2: New weights
__group_231__new_gradients (optional) - T3: New gradients
__group_231__new_moment_1 (optional) - T4: New averaged gradients
__group_231__new_moment_2 (optional) - T4: New averaged squared gradients
__group_231__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_232__new_weights (optional) - T2: New weights
__group_232__new_gradients (optional) - T3: New gradients
__group_232__new_moment_1 (optional) - T4: New averaged gradients
__group_232__new_moment_2 (optional) - T4: New averaged squared gradients
__group_232__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_233__new_weights (optional) - T2: New weights
__group_233__new_gradients (optional) - T3: New gradients
__group_233__new_moment_1 (optional) - T4: New averaged gradients
__group_233__new_moment_2 (optional) - T4: New averaged squared gradients
__group_233__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_234__new_weights (optional) - T2: New weights
__group_234__new_gradients (optional) - T3: New gradients
__group_234__new_moment_1 (optional) - T4: New averaged gradients
__group_234__new_moment_2 (optional) - T4: New averaged squared gradients
__group_234__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_235__new_weights (optional) - T2: New weights
__group_235__new_gradients (optional) - T3: New gradients
__group_235__new_moment_1 (optional) - T4: New averaged gradients
__group_235__new_moment_2 (optional) - T4: New averaged squared gradients
__group_235__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_236__new_weights (optional) - T2: New weights
__group_236__new_gradients (optional) - T3: New gradients
__group_236__new_moment_1 (optional) - T4: New averaged gradients
__group_236__new_moment_2 (optional) - T4: New averaged squared gradients
__group_236__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_237__new_weights (optional) - T2: New weights
__group_237__new_gradients (optional) - T3: New gradients
__group_237__new_moment_1 (optional) - T4: New averaged gradients
__group_237__new_moment_2 (optional) - T4: New averaged squared gradients
__group_237__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_238__new_weights (optional) - T2: New weights
__group_238__new_gradients (optional) - T3: New gradients
__group_238__new_moment_1 (optional) - T4: New averaged gradients
__group_238__new_moment_2 (optional) - T4: New averaged squared gradients
__group_238__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_239__new_weights (optional) - T2: New weights
__group_239__new_gradients (optional) - T3: New gradients
__group_239__new_moment_1 (optional) - T4: New averaged gradients
__group_239__new_moment_2 (optional) - T4: New averaged squared gradients
__group_239__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_240__new_weights (optional) - T2: New weights
__group_240__new_gradients (optional) - T3: New gradients
__group_240__new_moment_1 (optional) - T4: New averaged gradients
__group_240__new_moment_2 (optional) - T4: New averaged squared gradients
__group_240__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_241__new_weights (optional) - T2: New weights
__group_241__new_gradients (optional) - T3: New gradients
__group_241__new_moment_1 (optional) - T4: New averaged gradients
__group_241__new_moment_2 (optional) - T4: New averaged squared gradients
__group_241__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_242__new_weights (optional) - T2: New weights
__group_242__new_gradients (optional) - T3: New gradients
__group_242__new_moment_1 (optional) - T4: New averaged gradients
__group_242__new_moment_2 (optional) - T4: New averaged squared gradients
__group_242__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_243__new_weights (optional) - T2: New weights
__group_243__new_gradients (optional) - T3: New gradients
__group_243__new_moment_1 (optional) - T4: New averaged gradients
__group_243__new_moment_2 (optional) - T4: New averaged squared gradients
__group_243__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_244__new_weights (optional) - T2: New weights
__group_244__new_gradients (optional) - T3: New gradients
__group_244__new_moment_1 (optional) - T4: New averaged gradients
__group_244__new_moment_2 (optional) - T4: New averaged squared gradients
__group_244__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_245__new_weights (optional) - T2: New weights
__group_245__new_gradients (optional) - T3: New gradients
__group_245__new_moment_1 (optional) - T4: New averaged gradients
__group_245__new_moment_2 (optional) - T4: New averaged squared gradients
__group_245__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_246__new_weights (optional) - T2: New weights
__group_246__new_gradients (optional) - T3: New gradients
__group_246__new_moment_1 (optional) - T4: New averaged gradients
__group_246__new_moment_2 (optional) - T4: New averaged squared gradients
__group_246__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_247__new_weights (optional) - T2: New weights
__group_247__new_gradients (optional) - T3: New gradients
__group_247__new_moment_1 (optional) - T4: New averaged gradients
__group_247__new_moment_2 (optional) - T4: New averaged squared gradients
__group_247__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_248__new_weights (optional) - T2: New weights
__group_248__new_gradients (optional) - T3: New gradients
__group_248__new_moment_1 (optional) - T4: New averaged gradients
__group_248__new_moment_2 (optional) - T4: New averaged squared gradients
__group_248__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_249__new_weights (optional) - T2: New weights
__group_249__new_gradients (optional) - T3: New gradients
__group_249__new_moment_1 (optional) - T4: New averaged gradients
__group_249__new_moment_2 (optional) - T4: New averaged squared gradients
__group_249__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_250__new_weights (optional) - T2: New weights
__group_250__new_gradients (optional) - T3: New gradients
__group_250__new_moment_1 (optional) - T4: New averaged gradients
__group_250__new_moment_2 (optional) - T4: New averaged squared gradients
__group_250__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_251__new_weights (optional) - T2: New weights
__group_251__new_gradients (optional) - T3: New gradients
__group_251__new_moment_1 (optional) - T4: New averaged gradients
__group_251__new_moment_2 (optional) - T4: New averaged squared gradients
__group_251__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_252__new_weights (optional) - T2: New weights
__group_252__new_gradients (optional) - T3: New gradients
__group_252__new_moment_1 (optional) - T4: New averaged gradients
__group_252__new_moment_2 (optional) - T4: New averaged squared gradients
__group_252__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_253__new_weights (optional) - T2: New weights
__group_253__new_gradients (optional) - T3: New gradients
__group_253__new_moment_1 (optional) - T4: New averaged gradients
__group_253__new_moment_2 (optional) - T4: New averaged squared gradients
__group_253__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_254__new_weights (optional) - T2: New weights
__group_254__new_gradients (optional) - T3: New gradients
__group_254__new_moment_1 (optional) - T4: New averaged gradients
__group_254__new_moment_2 (optional) - T4: New averaged squared gradients
__group_254__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_255__new_weights (optional) - T2: New weights
__group_255__new_gradients (optional) - T3: New gradients
__group_255__new_moment_1 (optional) - T4: New averaged gradients
__group_255__new_moment_2 (optional) - T4: New averaged squared gradients
__group_255__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_256__new_weights (optional) - T2: New weights
__group_256__new_gradients (optional) - T3: New gradients
__group_256__new_moment_1 (optional) - T4: New averaged gradients
__group_256__new_moment_2 (optional) - T4: New averaged squared gradients
__group_256__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_257__new_weights (optional) - T2: New weights
__group_257__new_gradients (optional) - T3: New gradients
__group_257__new_moment_1 (optional) - T4: New averaged gradients
__group_257__new_moment_2 (optional) - T4: New averaged squared gradients
__group_257__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_258__new_weights (optional) - T2: New weights
__group_258__new_gradients (optional) - T3: New gradients
__group_258__new_moment_1 (optional) - T4: New averaged gradients
__group_258__new_moment_2 (optional) - T4: New averaged squared gradients
__group_258__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_259__new_weights (optional) - T2: New weights
__group_259__new_gradients (optional) - T3: New gradients
__group_259__new_moment_1 (optional) - T4: New averaged gradients
__group_259__new_moment_2 (optional) - T4: New averaged squared gradients
__group_259__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_260__new_weights (optional) - T2: New weights
__group_260__new_gradients (optional) - T3: New gradients
__group_260__new_moment_1 (optional) - T4: New averaged gradients
__group_260__new_moment_2 (optional) - T4: New averaged squared gradients
__group_260__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_261__new_weights (optional) - T2: New weights
__group_261__new_gradients (optional) - T3: New gradients
__group_261__new_moment_1 (optional) - T4: New averaged gradients
__group_261__new_moment_2 (optional) - T4: New averaged squared gradients
__group_261__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_262__new_weights (optional) - T2: New weights
__group_262__new_gradients (optional) - T3: New gradients
__group_262__new_moment_1 (optional) - T4: New averaged gradients
__group_262__new_moment_2 (optional) - T4: New averaged squared gradients
__group_262__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_263__new_weights (optional) - T2: New weights
__group_263__new_gradients (optional) - T3: New gradients
__group_263__new_moment_1 (optional) - T4: New averaged gradients
__group_263__new_moment_2 (optional) - T4: New averaged squared gradients
__group_263__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_264__new_weights (optional) - T2: New weights
__group_264__new_gradients (optional) - T3: New gradients
__group_264__new_moment_1 (optional) - T4: New averaged gradients
__group_264__new_moment_2 (optional) - T4: New averaged squared gradients
__group_264__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_265__new_weights (optional) - T2: New weights
__group_265__new_gradients (optional) - T3: New gradients
__group_265__new_moment_1 (optional) - T4: New averaged gradients
__group_265__new_moment_2 (optional) - T4: New averaged squared gradients
__group_265__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_266__new_weights (optional) - T2: New weights
__group_266__new_gradients (optional) - T3: New gradients
__group_266__new_moment_1 (optional) - T4: New averaged gradients
__group_266__new_moment_2 (optional) - T4: New averaged squared gradients
__group_266__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_267__new_weights (optional) - T2: New weights
__group_267__new_gradients (optional) - T3: New gradients
__group_267__new_moment_1 (optional) - T4: New averaged gradients
__group_267__new_moment_2 (optional) - T4: New averaged squared gradients
__group_267__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_268__new_weights (optional) - T2: New weights
__group_268__new_gradients (optional) - T3: New gradients
__group_268__new_moment_1 (optional) - T4: New averaged gradients
__group_268__new_moment_2 (optional) - T4: New averaged squared gradients
__group_268__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_269__new_weights (optional) - T2: New weights
__group_269__new_gradients (optional) - T3: New gradients
__group_269__new_moment_1 (optional) - T4: New averaged gradients
__group_269__new_moment_2 (optional) - T4: New averaged squared gradients
__group_269__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_270__new_weights (optional) - T2: New weights
__group_270__new_gradients (optional) - T3: New gradients
__group_270__new_moment_1 (optional) - T4: New averaged gradients
__group_270__new_moment_2 (optional) - T4: New averaged squared gradients
__group_270__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_271__new_weights (optional) - T2: New weights
__group_271__new_gradients (optional) - T3: New gradients
__group_271__new_moment_1 (optional) - T4: New averaged gradients
__group_271__new_moment_2 (optional) - T4: New averaged squared gradients
__group_271__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_272__new_weights (optional) - T2: New weights
__group_272__new_gradients (optional) - T3: New gradients
__group_272__new_moment_1 (optional) - T4: New averaged gradients
__group_272__new_moment_2 (optional) - T4: New averaged squared gradients
__group_272__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_273__new_weights (optional) - T2: New weights
__group_273__new_gradients (optional) - T3: New gradients
__group_273__new_moment_1 (optional) - T4: New averaged gradients
__group_273__new_moment_2 (optional) - T4: New averaged squared gradients
__group_273__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_274__new_weights (optional) - T2: New weights
__group_274__new_gradients (optional) - T3: New gradients
__group_274__new_moment_1 (optional) - T4: New averaged gradients
__group_274__new_moment_2 (optional) - T4: New averaged squared gradients
__group_274__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_275__new_weights (optional) - T2: New weights
__group_275__new_gradients (optional) - T3: New gradients
__group_275__new_moment_1 (optional) - T4: New averaged gradients
__group_275__new_moment_2 (optional) - T4: New averaged squared gradients
__group_275__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_276__new_weights (optional) - T2: New weights
__group_276__new_gradients (optional) - T3: New gradients
__group_276__new_moment_1 (optional) - T4: New averaged gradients
__group_276__new_moment_2 (optional) - T4: New averaged squared gradients
__group_276__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_277__new_weights (optional) - T2: New weights
__group_277__new_gradients (optional) - T3: New gradients
__group_277__new_moment_1 (optional) - T4: New averaged gradients
__group_277__new_moment_2 (optional) - T4: New averaged squared gradients
__group_277__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_278__new_weights (optional) - T2: New weights
__group_278__new_gradients (optional) - T3: New gradients
__group_278__new_moment_1 (optional) - T4: New averaged gradients
__group_278__new_moment_2 (optional) - T4: New averaged squared gradients
__group_278__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_279__new_weights (optional) - T2: New weights
__group_279__new_gradients (optional) - T3: New gradients
__group_279__new_moment_1 (optional) - T4: New averaged gradients
__group_279__new_moment_2 (optional) - T4: New averaged squared gradients
__group_279__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_280__new_weights (optional) - T2: New weights
__group_280__new_gradients (optional) - T3: New gradients
__group_280__new_moment_1 (optional) - T4: New averaged gradients
__group_280__new_moment_2 (optional) - T4: New averaged squared gradients
__group_280__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_281__new_weights (optional) - T2: New weights
__group_281__new_gradients (optional) - T3: New gradients
__group_281__new_moment_1 (optional) - T4: New averaged gradients
__group_281__new_moment_2 (optional) - T4: New averaged squared gradients
__group_281__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_282__new_weights (optional) - T2: New weights
__group_282__new_gradients (optional) - T3: New gradients
__group_282__new_moment_1 (optional) - T4: New averaged gradients
__group_282__new_moment_2 (optional) - T4: New averaged squared gradients
__group_282__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_283__new_weights (optional) - T2: New weights
__group_283__new_gradients (optional) - T3: New gradients
__group_283__new_moment_1 (optional) - T4: New averaged gradients
__group_283__new_moment_2 (optional) - T4: New averaged squared gradients
__group_283__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_284__new_weights (optional) - T2: New weights
__group_284__new_gradients (optional) - T3: New gradients
__group_284__new_moment_1 (optional) - T4: New averaged gradients
__group_284__new_moment_2 (optional) - T4: New averaged squared gradients
__group_284__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_285__new_weights (optional) - T2: New weights
__group_285__new_gradients (optional) - T3: New gradients
__group_285__new_moment_1 (optional) - T4: New averaged gradients
__group_285__new_moment_2 (optional) - T4: New averaged squared gradients
__group_285__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_286__new_weights (optional) - T2: New weights
__group_286__new_gradients (optional) - T3: New gradients
__group_286__new_moment_1 (optional) - T4: New averaged gradients
__group_286__new_moment_2 (optional) - T4: New averaged squared gradients
__group_286__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_287__new_weights (optional) - T2: New weights
__group_287__new_gradients (optional) - T3: New gradients
__group_287__new_moment_1 (optional) - T4: New averaged gradients
__group_287__new_moment_2 (optional) - T4: New averaged squared gradients
__group_287__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_288__new_weights (optional) - T2: New weights
__group_288__new_gradients (optional) - T3: New gradients
__group_288__new_moment_1 (optional) - T4: New averaged gradients
__group_288__new_moment_2 (optional) - T4: New averaged squared gradients
__group_288__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_289__new_weights (optional) - T2: New weights
__group_289__new_gradients (optional) - T3: New gradients
__group_289__new_moment_1 (optional) - T4: New averaged gradients
__group_289__new_moment_2 (optional) - T4: New averaged squared gradients
__group_289__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_290__new_weights (optional) - T2: New weights
__group_290__new_gradients (optional) - T3: New gradients
__group_290__new_moment_1 (optional) - T4: New averaged gradients
__group_290__new_moment_2 (optional) - T4: New averaged squared gradients
__group_290__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_291__new_weights (optional) - T2: New weights
__group_291__new_gradients (optional) - T3: New gradients
__group_291__new_moment_1 (optional) - T4: New averaged gradients
__group_291__new_moment_2 (optional) - T4: New averaged squared gradients
__group_291__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_292__new_weights (optional) - T2: New weights
__group_292__new_gradients (optional) - T3: New gradients
__group_292__new_moment_1 (optional) - T4: New averaged gradients
__group_292__new_moment_2 (optional) - T4: New averaged squared gradients
__group_292__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_293__new_weights (optional) - T2: New weights
__group_293__new_gradients (optional) - T3: New gradients
__group_293__new_moment_1 (optional) - T4: New averaged gradients
__group_293__new_moment_2 (optional) - T4: New averaged squared gradients
__group_293__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_294__new_weights (optional) - T2: New weights
__group_294__new_gradients (optional) - T3: New gradients
__group_294__new_moment_1 (optional) - T4: New averaged gradients
__group_294__new_moment_2 (optional) - T4: New averaged squared gradients
__group_294__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_295__new_weights (optional) - T2: New weights
__group_295__new_gradients (optional) - T3: New gradients
__group_295__new_moment_1 (optional) - T4: New averaged gradients
__group_295__new_moment_2 (optional) - T4: New averaged squared gradients
__group_295__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_296__new_weights (optional) - T2: New weights
__group_296__new_gradients (optional) - T3: New gradients
__group_296__new_moment_1 (optional) - T4: New averaged gradients
__group_296__new_moment_2 (optional) - T4: New averaged squared gradients
__group_296__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_297__new_weights (optional) - T2: New weights
__group_297__new_gradients (optional) - T3: New gradients
__group_297__new_moment_1 (optional) - T4: New averaged gradients
__group_297__new_moment_2 (optional) - T4: New averaged squared gradients
__group_297__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_298__new_weights (optional) - T2: New weights
__group_298__new_gradients (optional) - T3: New gradients
__group_298__new_moment_1 (optional) - T4: New averaged gradients
__group_298__new_moment_2 (optional) - T4: New averaged squared gradients
__group_298__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_299__new_weights (optional) - T2: New weights
__group_299__new_gradients (optional) - T3: New gradients
__group_299__new_moment_1 (optional) - T4: New averaged gradients
__group_299__new_moment_2 (optional) - T4: New averaged squared gradients
__group_299__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_300__new_weights (optional) - T2: New weights
__group_300__new_gradients (optional) - T3: New gradients
__group_300__new_moment_1 (optional) - T4: New averaged gradients
__group_300__new_moment_2 (optional) - T4: New averaged squared gradients
__group_300__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_301__new_weights (optional) - T2: New weights
__group_301__new_gradients (optional) - T3: New gradients
__group_301__new_moment_1 (optional) - T4: New averaged gradients
__group_301__new_moment_2 (optional) - T4: New averaged squared gradients
__group_301__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_302__new_weights (optional) - T2: New weights
__group_302__new_gradients (optional) - T3: New gradients
__group_302__new_moment_1 (optional) - T4: New averaged gradients
__group_302__new_moment_2 (optional) - T4: New averaged squared gradients
__group_302__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_303__new_weights (optional) - T2: New weights
__group_303__new_gradients (optional) - T3: New gradients
__group_303__new_moment_1 (optional) - T4: New averaged gradients
__group_303__new_moment_2 (optional) - T4: New averaged squared gradients
__group_303__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_304__new_weights (optional) - T2: New weights
__group_304__new_gradients (optional) - T3: New gradients
__group_304__new_moment_1 (optional) - T4: New averaged gradients
__group_304__new_moment_2 (optional) - T4: New averaged squared gradients
__group_304__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_305__new_weights (optional) - T2: New weights
__group_305__new_gradients (optional) - T3: New gradients
__group_305__new_moment_1 (optional) - T4: New averaged gradients
__group_305__new_moment_2 (optional) - T4: New averaged squared gradients
__group_305__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_306__new_weights (optional) - T2: New weights
__group_306__new_gradients (optional) - T3: New gradients
__group_306__new_moment_1 (optional) - T4: New averaged gradients
__group_306__new_moment_2 (optional) - T4: New averaged squared gradients
__group_306__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_307__new_weights (optional) - T2: New weights
__group_307__new_gradients (optional) - T3: New gradients
__group_307__new_moment_1 (optional) - T4: New averaged gradients
__group_307__new_moment_2 (optional) - T4: New averaged squared gradients
__group_307__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_308__new_weights (optional) - T2: New weights
__group_308__new_gradients (optional) - T3: New gradients
__group_308__new_moment_1 (optional) - T4: New averaged gradients
__group_308__new_moment_2 (optional) - T4: New averaged squared gradients
__group_308__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_309__new_weights (optional) - T2: New weights
__group_309__new_gradients (optional) - T3: New gradients
__group_309__new_moment_1 (optional) - T4: New averaged gradients
__group_309__new_moment_2 (optional) - T4: New averaged squared gradients
__group_309__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_310__new_weights (optional) - T2: New weights
__group_310__new_gradients (optional) - T3: New gradients
__group_310__new_moment_1 (optional) - T4: New averaged gradients
__group_310__new_moment_2 (optional) - T4: New averaged squared gradients
__group_310__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_311__new_weights (optional) - T2: New weights
__group_311__new_gradients (optional) - T3: New gradients
__group_311__new_moment_1 (optional) - T4: New averaged gradients
__group_311__new_moment_2 (optional) - T4: New averaged squared gradients
__group_311__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_312__new_weights (optional) - T2: New weights
__group_312__new_gradients (optional) - T3: New gradients
__group_312__new_moment_1 (optional) - T4: New averaged gradients
__group_312__new_moment_2 (optional) - T4: New averaged squared gradients
__group_312__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_313__new_weights (optional) - T2: New weights
__group_313__new_gradients (optional) - T3: New gradients
__group_313__new_moment_1 (optional) - T4: New averaged gradients
__group_313__new_moment_2 (optional) - T4: New averaged squared gradients
__group_313__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_314__new_weights (optional) - T2: New weights
__group_314__new_gradients (optional) - T3: New gradients
__group_314__new_moment_1 (optional) - T4: New averaged gradients
__group_314__new_moment_2 (optional) - T4: New averaged squared gradients
__group_314__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_315__new_weights (optional) - T2: New weights
__group_315__new_gradients (optional) - T3: New gradients
__group_315__new_moment_1 (optional) - T4: New averaged gradients
__group_315__new_moment_2 (optional) - T4: New averaged squared gradients
__group_315__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_316__new_weights (optional) - T2: New weights
__group_316__new_gradients (optional) - T3: New gradients
__group_316__new_moment_1 (optional) - T4: New averaged gradients
__group_316__new_moment_2 (optional) - T4: New averaged squared gradients
__group_316__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_317__new_weights (optional) - T2: New weights
__group_317__new_gradients (optional) - T3: New gradients
__group_317__new_moment_1 (optional) - T4: New averaged gradients
__group_317__new_moment_2 (optional) - T4: New averaged squared gradients
__group_317__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_318__new_weights (optional) - T2: New weights
__group_318__new_gradients (optional) - T3: New gradients
__group_318__new_moment_1 (optional) - T4: New averaged gradients
__group_318__new_moment_2 (optional) - T4: New averaged squared gradients
__group_318__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_319__new_weights (optional) - T2: New weights
__group_319__new_gradients (optional) - T3: New gradients
__group_319__new_moment_1 (optional) - T4: New averaged gradients
__group_319__new_moment_2 (optional) - T4: New averaged squared gradients
__group_319__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_320__new_weights (optional) - T2: New weights
__group_320__new_gradients (optional) - T3: New gradients
__group_320__new_moment_1 (optional) - T4: New averaged gradients
__group_320__new_moment_2 (optional) - T4: New averaged squared gradients
__group_320__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_321__new_weights (optional) - T2: New weights
__group_321__new_gradients (optional) - T3: New gradients
__group_321__new_moment_1 (optional) - T4: New averaged gradients
__group_321__new_moment_2 (optional) - T4: New averaged squared gradients
__group_321__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_322__new_weights (optional) - T2: New weights
__group_322__new_gradients (optional) - T3: New gradients
__group_322__new_moment_1 (optional) - T4: New averaged gradients
__group_322__new_moment_2 (optional) - T4: New averaged squared gradients
__group_322__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_323__new_weights (optional) - T2: New weights
__group_323__new_gradients (optional) - T3: New gradients
__group_323__new_moment_1 (optional) - T4: New averaged gradients
__group_323__new_moment_2 (optional) - T4: New averaged squared gradients
__group_323__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_324__new_weights (optional) - T2: New weights
__group_324__new_gradients (optional) - T3: New gradients
__group_324__new_moment_1 (optional) - T4: New averaged gradients
__group_324__new_moment_2 (optional) - T4: New averaged squared gradients
__group_324__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_325__new_weights (optional) - T2: New weights
__group_325__new_gradients (optional) - T3: New gradients
__group_325__new_moment_1 (optional) - T4: New averaged gradients
__group_325__new_moment_2 (optional) - T4: New averaged squared gradients
__group_325__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_326__new_weights (optional) - T2: New weights
__group_326__new_gradients (optional) - T3: New gradients
__group_326__new_moment_1 (optional) - T4: New averaged gradients
__group_326__new_moment_2 (optional) - T4: New averaged squared gradients
__group_326__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_327__new_weights (optional) - T2: New weights
__group_327__new_gradients (optional) - T3: New gradients
__group_327__new_moment_1 (optional) - T4: New averaged gradients
__group_327__new_moment_2 (optional) - T4: New averaged squared gradients
__group_327__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_328__new_weights (optional) - T2: New weights
__group_328__new_gradients (optional) - T3: New gradients
__group_328__new_moment_1 (optional) - T4: New averaged gradients
__group_328__new_moment_2 (optional) - T4: New averaged squared gradients
__group_328__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_329__new_weights (optional) - T2: New weights
__group_329__new_gradients (optional) - T3: New gradients
__group_329__new_moment_1 (optional) - T4: New averaged gradients
__group_329__new_moment_2 (optional) - T4: New averaged squared gradients
__group_329__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_330__new_weights (optional) - T2: New weights
__group_330__new_gradients (optional) - T3: New gradients
__group_330__new_moment_1 (optional) - T4: New averaged gradients
__group_330__new_moment_2 (optional) - T4: New averaged squared gradients
__group_330__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_331__new_weights (optional) - T2: New weights
__group_331__new_gradients (optional) - T3: New gradients
__group_331__new_moment_1 (optional) - T4: New averaged gradients
__group_331__new_moment_2 (optional) - T4: New averaged squared gradients
__group_331__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_332__new_weights (optional) - T2: New weights
__group_332__new_gradients (optional) - T3: New gradients
__group_332__new_moment_1 (optional) - T4: New averaged gradients
__group_332__new_moment_2 (optional) - T4: New averaged squared gradients
__group_332__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_333__new_weights (optional) - T2: New weights
__group_333__new_gradients (optional) - T3: New gradients
__group_333__new_moment_1 (optional) - T4: New averaged gradients
__group_333__new_moment_2 (optional) - T4: New averaged squared gradients
__group_333__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_334__new_weights (optional) - T2: New weights
__group_334__new_gradients (optional) - T3: New gradients
__group_334__new_moment_1 (optional) - T4: New averaged gradients
__group_334__new_moment_2 (optional) - T4: New averaged squared gradients
__group_334__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_335__new_weights (optional) - T2: New weights
__group_335__new_gradients (optional) - T3: New gradients
__group_335__new_moment_1 (optional) - T4: New averaged gradients
__group_335__new_moment_2 (optional) - T4: New averaged squared gradients
__group_335__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_336__new_weights (optional) - T2: New weights
__group_336__new_gradients (optional) - T3: New gradients
__group_336__new_moment_1 (optional) - T4: New averaged gradients
__group_336__new_moment_2 (optional) - T4: New averaged squared gradients
__group_336__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_337__new_weights (optional) - T2: New weights
__group_337__new_gradients (optional) - T3: New gradients
__group_337__new_moment_1 (optional) - T4: New averaged gradients
__group_337__new_moment_2 (optional) - T4: New averaged squared gradients
__group_337__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_338__new_weights (optional) - T2: New weights
__group_338__new_gradients (optional) - T3: New gradients
__group_338__new_moment_1 (optional) - T4: New averaged gradients
__group_338__new_moment_2 (optional) - T4: New averaged squared gradients
__group_338__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_339__new_weights (optional) - T2: New weights
__group_339__new_gradients (optional) - T3: New gradients
__group_339__new_moment_1 (optional) - T4: New averaged gradients
__group_339__new_moment_2 (optional) - T4: New averaged squared gradients
__group_339__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_340__new_weights (optional) - T2: New weights
__group_340__new_gradients (optional) - T3: New gradients
__group_340__new_moment_1 (optional) - T4: New averaged gradients
__group_340__new_moment_2 (optional) - T4: New averaged squared gradients
__group_340__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_341__new_weights (optional) - T2: New weights
__group_341__new_gradients (optional) - T3: New gradients
__group_341__new_moment_1 (optional) - T4: New averaged gradients
__group_341__new_moment_2 (optional) - T4: New averaged squared gradients
__group_341__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_342__new_weights (optional) - T2: New weights
__group_342__new_gradients (optional) - T3: New gradients
__group_342__new_moment_1 (optional) - T4: New averaged gradients
__group_342__new_moment_2 (optional) - T4: New averaged squared gradients
__group_342__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_343__new_weights (optional) - T2: New weights
__group_343__new_gradients (optional) - T3: New gradients
__group_343__new_moment_1 (optional) - T4: New averaged gradients
__group_343__new_moment_2 (optional) - T4: New averaged squared gradients
__group_343__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_344__new_weights (optional) - T2: New weights
__group_344__new_gradients (optional) - T3: New gradients
__group_344__new_moment_1 (optional) - T4: New averaged gradients
__group_344__new_moment_2 (optional) - T4: New averaged squared gradients
__group_344__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_345__new_weights (optional) - T2: New weights
__group_345__new_gradients (optional) - T3: New gradients
__group_345__new_moment_1 (optional) - T4: New averaged gradients
__group_345__new_moment_2 (optional) - T4: New averaged squared gradients
__group_345__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_346__new_weights (optional) - T2: New weights
__group_346__new_gradients (optional) - T3: New gradients
__group_346__new_moment_1 (optional) - T4: New averaged gradients
__group_346__new_moment_2 (optional) - T4: New averaged squared gradients
__group_346__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_347__new_weights (optional) - T2: New weights
__group_347__new_gradients (optional) - T3: New gradients
__group_347__new_moment_1 (optional) - T4: New averaged gradients
__group_347__new_moment_2 (optional) - T4: New averaged squared gradients
__group_347__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_348__new_weights (optional) - T2: New weights
__group_348__new_gradients (optional) - T3: New gradients
__group_348__new_moment_1 (optional) - T4: New averaged gradients
__group_348__new_moment_2 (optional) - T4: New averaged squared gradients
__group_348__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_349__new_weights (optional) - T2: New weights
__group_349__new_gradients (optional) - T3: New gradients
__group_349__new_moment_1 (optional) - T4: New averaged gradients
__group_349__new_moment_2 (optional) - T4: New averaged squared gradients
__group_349__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_350__new_weights (optional) - T2: New weights
__group_350__new_gradients (optional) - T3: New gradients
__group_350__new_moment_1 (optional) - T4: New averaged gradients
__group_350__new_moment_2 (optional) - T4: New averaged squared gradients
__group_350__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_351__new_weights (optional) - T2: New weights
__group_351__new_gradients (optional) - T3: New gradients
__group_351__new_moment_1 (optional) - T4: New averaged gradients
__group_351__new_moment_2 (optional) - T4: New averaged squared gradients
__group_351__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_352__new_weights (optional) - T2: New weights
__group_352__new_gradients (optional) - T3: New gradients
__group_352__new_moment_1 (optional) - T4: New averaged gradients
__group_352__new_moment_2 (optional) - T4: New averaged squared gradients
__group_352__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_353__new_weights (optional) - T2: New weights
__group_353__new_gradients (optional) - T3: New gradients
__group_353__new_moment_1 (optional) - T4: New averaged gradients
__group_353__new_moment_2 (optional) - T4: New averaged squared gradients
__group_353__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_354__new_weights (optional) - T2: New weights
__group_354__new_gradients (optional) - T3: New gradients
__group_354__new_moment_1 (optional) - T4: New averaged gradients
__group_354__new_moment_2 (optional) - T4: New averaged squared gradients
__group_354__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_355__new_weights (optional) - T2: New weights
__group_355__new_gradients (optional) - T3: New gradients
__group_355__new_moment_1 (optional) - T4: New averaged gradients
__group_355__new_moment_2 (optional) - T4: New averaged squared gradients
__group_355__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_356__new_weights (optional) - T2: New weights
__group_356__new_gradients (optional) - T3: New gradients
__group_356__new_moment_1 (optional) - T4: New averaged gradients
__group_356__new_moment_2 (optional) - T4: New averaged squared gradients
__group_356__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_357__new_weights (optional) - T2: New weights
__group_357__new_gradients (optional) - T3: New gradients
__group_357__new_moment_1 (optional) - T4: New averaged gradients
__group_357__new_moment_2 (optional) - T4: New averaged squared gradients
__group_357__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_358__new_weights (optional) - T2: New weights
__group_358__new_gradients (optional) - T3: New gradients
__group_358__new_moment_1 (optional) - T4: New averaged gradients
__group_358__new_moment_2 (optional) - T4: New averaged squared gradients
__group_358__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_359__new_weights (optional) - T2: New weights
__group_359__new_gradients (optional) - T3: New gradients
__group_359__new_moment_1 (optional) - T4: New averaged gradients
__group_359__new_moment_2 (optional) - T4: New averaged squared gradients
__group_359__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_360__new_weights (optional) - T2: New weights
__group_360__new_gradients (optional) - T3: New gradients
__group_360__new_moment_1 (optional) - T4: New averaged gradients
__group_360__new_moment_2 (optional) - T4: New averaged squared gradients
__group_360__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_361__new_weights (optional) - T2: New weights
__group_361__new_gradients (optional) - T3: New gradients
__group_361__new_moment_1 (optional) - T4: New averaged gradients
__group_361__new_moment_2 (optional) - T4: New averaged squared gradients
__group_361__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_362__new_weights (optional) - T2: New weights
__group_362__new_gradients (optional) - T3: New gradients
__group_362__new_moment_1 (optional) - T4: New averaged gradients
__group_362__new_moment_2 (optional) - T4: New averaged squared gradients
__group_362__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_363__new_weights (optional) - T2: New weights
__group_363__new_gradients (optional) - T3: New gradients
__group_363__new_moment_1 (optional) - T4: New averaged gradients
__group_363__new_moment_2 (optional) - T4: New averaged squared gradients
__group_363__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_364__new_weights (optional) - T2: New weights
__group_364__new_gradients (optional) - T3: New gradients
__group_364__new_moment_1 (optional) - T4: New averaged gradients
__group_364__new_moment_2 (optional) - T4: New averaged squared gradients
__group_364__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_365__new_weights (optional) - T2: New weights
__group_365__new_gradients (optional) - T3: New gradients
__group_365__new_moment_1 (optional) - T4: New averaged gradients
__group_365__new_moment_2 (optional) - T4: New averaged squared gradients
__group_365__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_366__new_weights (optional) - T2: New weights
__group_366__new_gradients (optional) - T3: New gradients
__group_366__new_moment_1 (optional) - T4: New averaged gradients
__group_366__new_moment_2 (optional) - T4: New averaged squared gradients
__group_366__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_367__new_weights (optional) - T2: New weights
__group_367__new_gradients (optional) - T3: New gradients
__group_367__new_moment_1 (optional) - T4: New averaged gradients
__group_367__new_moment_2 (optional) - T4: New averaged squared gradients
__group_367__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_368__new_weights (optional) - T2: New weights
__group_368__new_gradients (optional) - T3: New gradients
__group_368__new_moment_1 (optional) - T4: New averaged gradients
__group_368__new_moment_2 (optional) - T4: New averaged squared gradients
__group_368__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_369__new_weights (optional) - T2: New weights
__group_369__new_gradients (optional) - T3: New gradients
__group_369__new_moment_1 (optional) - T4: New averaged gradients
__group_369__new_moment_2 (optional) - T4: New averaged squared gradients
__group_369__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_370__new_weights (optional) - T2: New weights
__group_370__new_gradients (optional) - T3: New gradients
__group_370__new_moment_1 (optional) - T4: New averaged gradients
__group_370__new_moment_2 (optional) - T4: New averaged squared gradients
__group_370__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_371__new_weights (optional) - T2: New weights
__group_371__new_gradients (optional) - T3: New gradients
__group_371__new_moment_1 (optional) - T4: New averaged gradients
__group_371__new_moment_2 (optional) - T4: New averaged squared gradients
__group_371__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_372__new_weights (optional) - T2: New weights
__group_372__new_gradients (optional) - T3: New gradients
__group_372__new_moment_1 (optional) - T4: New averaged gradients
__group_372__new_moment_2 (optional) - T4: New averaged squared gradients
__group_372__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_373__new_weights (optional) - T2: New weights
__group_373__new_gradients (optional) - T3: New gradients
__group_373__new_moment_1 (optional) - T4: New averaged gradients
__group_373__new_moment_2 (optional) - T4: New averaged squared gradients
__group_373__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_374__new_weights (optional) - T2: New weights
__group_374__new_gradients (optional) - T3: New gradients
__group_374__new_moment_1 (optional) - T4: New averaged gradients
__group_374__new_moment_2 (optional) - T4: New averaged squared gradients
__group_374__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_375__new_weights (optional) - T2: New weights
__group_375__new_gradients (optional) - T3: New gradients
__group_375__new_moment_1 (optional) - T4: New averaged gradients
__group_375__new_moment_2 (optional) - T4: New averaged squared gradients
__group_375__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_376__new_weights (optional) - T2: New weights
__group_376__new_gradients (optional) - T3: New gradients
__group_376__new_moment_1 (optional) - T4: New averaged gradients
__group_376__new_moment_2 (optional) - T4: New averaged squared gradients
__group_376__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_377__new_weights (optional) - T2: New weights
__group_377__new_gradients (optional) - T3: New gradients
__group_377__new_moment_1 (optional) - T4: New averaged gradients
__group_377__new_moment_2 (optional) - T4: New averaged squared gradients
__group_377__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_378__new_weights (optional) - T2: New weights
__group_378__new_gradients (optional) - T3: New gradients
__group_378__new_moment_1 (optional) - T4: New averaged gradients
__group_378__new_moment_2 (optional) - T4: New averaged squared gradients
__group_378__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_379__new_weights (optional) - T2: New weights
__group_379__new_gradients (optional) - T3: New gradients
__group_379__new_moment_1 (optional) - T4: New averaged gradients
__group_379__new_moment_2 (optional) - T4: New averaged squared gradients
__group_379__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_380__new_weights (optional) - T2: New weights
__group_380__new_gradients (optional) - T3: New gradients
__group_380__new_moment_1 (optional) - T4: New averaged gradients
__group_380__new_moment_2 (optional) - T4: New averaged squared gradients
__group_380__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_381__new_weights (optional) - T2: New weights
__group_381__new_gradients (optional) - T3: New gradients
__group_381__new_moment_1 (optional) - T4: New averaged gradients
__group_381__new_moment_2 (optional) - T4: New averaged squared gradients
__group_381__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_382__new_weights (optional) - T2: New weights
__group_382__new_gradients (optional) - T3: New gradients
__group_382__new_moment_1 (optional) - T4: New averaged gradients
__group_382__new_moment_2 (optional) - T4: New averaged squared gradients
__group_382__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_383__new_weights (optional) - T2: New weights
__group_383__new_gradients (optional) - T3: New gradients
__group_383__new_moment_1 (optional) - T4: New averaged gradients
__group_383__new_moment_2 (optional) - T4: New averaged squared gradients
__group_383__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_384__new_weights (optional) - T2: New weights
__group_384__new_gradients (optional) - T3: New gradients
__group_384__new_moment_1 (optional) - T4: New averaged gradients
__group_384__new_moment_2 (optional) - T4: New averaged squared gradients
__group_384__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_385__new_weights (optional) - T2: New weights
__group_385__new_gradients (optional) - T3: New gradients
__group_385__new_moment_1 (optional) - T4: New averaged gradients
__group_385__new_moment_2 (optional) - T4: New averaged squared gradients
__group_385__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_386__new_weights (optional) - T2: New weights
__group_386__new_gradients (optional) - T3: New gradients
__group_386__new_moment_1 (optional) - T4: New averaged gradients
__group_386__new_moment_2 (optional) - T4: New averaged squared gradients
__group_386__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_387__new_weights (optional) - T2: New weights
__group_387__new_gradients (optional) - T3: New gradients
__group_387__new_moment_1 (optional) - T4: New averaged gradients
__group_387__new_moment_2 (optional) - T4: New averaged squared gradients
__group_387__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_388__new_weights (optional) - T2: New weights
__group_388__new_gradients (optional) - T3: New gradients
__group_388__new_moment_1 (optional) - T4: New averaged gradients
__group_388__new_moment_2 (optional) - T4: New averaged squared gradients
__group_388__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_389__new_weights (optional) - T2: New weights
__group_389__new_gradients (optional) - T3: New gradients
__group_389__new_moment_1 (optional) - T4: New averaged gradients
__group_389__new_moment_2 (optional) - T4: New averaged squared gradients
__group_389__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_390__new_weights (optional) - T2: New weights
__group_390__new_gradients (optional) - T3: New gradients
__group_390__new_moment_1 (optional) - T4: New averaged gradients
__group_390__new_moment_2 (optional) - T4: New averaged squared gradients
__group_390__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_391__new_weights (optional) - T2: New weights
__group_391__new_gradients (optional) - T3: New gradients
__group_391__new_moment_1 (optional) - T4: New averaged gradients
__group_391__new_moment_2 (optional) - T4: New averaged squared gradients
__group_391__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_392__new_weights (optional) - T2: New weights
__group_392__new_gradients (optional) - T3: New gradients
__group_392__new_moment_1 (optional) - T4: New averaged gradients
__group_392__new_moment_2 (optional) - T4: New averaged squared gradients
__group_392__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_393__new_weights (optional) - T2: New weights
__group_393__new_gradients (optional) - T3: New gradients
__group_393__new_moment_1 (optional) - T4: New averaged gradients
__group_393__new_moment_2 (optional) - T4: New averaged squared gradients
__group_393__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_394__new_weights (optional) - T2: New weights
__group_394__new_gradients (optional) - T3: New gradients
__group_394__new_moment_1 (optional) - T4: New averaged gradients
__group_394__new_moment_2 (optional) - T4: New averaged squared gradients
__group_394__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_395__new_weights (optional) - T2: New weights
__group_395__new_gradients (optional) - T3: New gradients
__group_395__new_moment_1 (optional) - T4: New averaged gradients
__group_395__new_moment_2 (optional) - T4: New averaged squared gradients
__group_395__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_396__new_weights (optional) - T2: New weights
__group_396__new_gradients (optional) - T3: New gradients
__group_396__new_moment_1 (optional) - T4: New averaged gradients
__group_396__new_moment_2 (optional) - T4: New averaged squared gradients
__group_396__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_397__new_weights (optional) - T2: New weights
__group_397__new_gradients (optional) - T3: New gradients
__group_397__new_moment_1 (optional) - T4: New averaged gradients
__group_397__new_moment_2 (optional) - T4: New averaged squared gradients
__group_397__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_398__new_weights (optional) - T2: New weights
__group_398__new_gradients (optional) - T3: New gradients
__group_398__new_moment_1 (optional) - T4: New averaged gradients
__group_398__new_moment_2 (optional) - T4: New averaged squared gradients
__group_398__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_399__new_weights (optional) - T2: New weights
__group_399__new_gradients (optional) - T3: New gradients
__group_399__new_moment_1 (optional) - T4: New averaged gradients
__group_399__new_moment_2 (optional) - T4: New averaged squared gradients
__group_399__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_400__new_weights (optional) - T2: New weights
__group_400__new_gradients (optional) - T3: New gradients
__group_400__new_moment_1 (optional) - T4: New averaged gradients
__group_400__new_moment_2 (optional) - T4: New averaged squared gradients
__group_400__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_401__new_weights (optional) - T2: New weights
__group_401__new_gradients (optional) - T3: New gradients
__group_401__new_moment_1 (optional) - T4: New averaged gradients
__group_401__new_moment_2 (optional) - T4: New averaged squared gradients
__group_401__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_402__new_weights (optional) - T2: New weights
__group_402__new_gradients (optional) - T3: New gradients
__group_402__new_moment_1 (optional) - T4: New averaged gradients
__group_402__new_moment_2 (optional) - T4: New averaged squared gradients
__group_402__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_403__new_weights (optional) - T2: New weights
__group_403__new_gradients (optional) - T3: New gradients
__group_403__new_moment_1 (optional) - T4: New averaged gradients
__group_403__new_moment_2 (optional) - T4: New averaged squared gradients
__group_403__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_404__new_weights (optional) - T2: New weights
__group_404__new_gradients (optional) - T3: New gradients
__group_404__new_moment_1 (optional) - T4: New averaged gradients
__group_404__new_moment_2 (optional) - T4: New averaged squared gradients
__group_404__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_405__new_weights (optional) - T2: New weights
__group_405__new_gradients (optional) - T3: New gradients
__group_405__new_moment_1 (optional) - T4: New averaged gradients
__group_405__new_moment_2 (optional) - T4: New averaged squared gradients
__group_405__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_406__new_weights (optional) - T2: New weights
__group_406__new_gradients (optional) - T3: New gradients
__group_406__new_moment_1 (optional) - T4: New averaged gradients
__group_406__new_moment_2 (optional) - T4: New averaged squared gradients
__group_406__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_407__new_weights (optional) - T2: New weights
__group_407__new_gradients (optional) - T3: New gradients
__group_407__new_moment_1 (optional) - T4: New averaged gradients
__group_407__new_moment_2 (optional) - T4: New averaged squared gradients
__group_407__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_408__new_weights (optional) - T2: New weights
__group_408__new_gradients (optional) - T3: New gradients
__group_408__new_moment_1 (optional) - T4: New averaged gradients
__group_408__new_moment_2 (optional) - T4: New averaged squared gradients
__group_408__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_409__new_weights (optional) - T2: New weights
__group_409__new_gradients (optional) - T3: New gradients
__group_409__new_moment_1 (optional) - T4: New averaged gradients
__group_409__new_moment_2 (optional) - T4: New averaged squared gradients
__group_409__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_410__new_weights (optional) - T2: New weights
__group_410__new_gradients (optional) - T3: New gradients
__group_410__new_moment_1 (optional) - T4: New averaged gradients
__group_410__new_moment_2 (optional) - T4: New averaged squared gradients
__group_410__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_411__new_weights (optional) - T2: New weights
__group_411__new_gradients (optional) - T3: New gradients
__group_411__new_moment_1 (optional) - T4: New averaged gradients
__group_411__new_moment_2 (optional) - T4: New averaged squared gradients
__group_411__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_412__new_weights (optional) - T2: New weights
__group_412__new_gradients (optional) - T3: New gradients
__group_412__new_moment_1 (optional) - T4: New averaged gradients
__group_412__new_moment_2 (optional) - T4: New averaged squared gradients
__group_412__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_413__new_weights (optional) - T2: New weights
__group_413__new_gradients (optional) - T3: New gradients
__group_413__new_moment_1 (optional) - T4: New averaged gradients
__group_413__new_moment_2 (optional) - T4: New averaged squared gradients
__group_413__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_414__new_weights (optional) - T2: New weights
__group_414__new_gradients (optional) - T3: New gradients
__group_414__new_moment_1 (optional) - T4: New averaged gradients
__group_414__new_moment_2 (optional) - T4: New averaged squared gradients
__group_414__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_415__new_weights (optional) - T2: New weights
__group_415__new_gradients (optional) - T3: New gradients
__group_415__new_moment_1 (optional) - T4: New averaged gradients
__group_415__new_moment_2 (optional) - T4: New averaged squared gradients
__group_415__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_416__new_weights (optional) - T2: New weights
__group_416__new_gradients (optional) - T3: New gradients
__group_416__new_moment_1 (optional) - T4: New averaged gradients
__group_416__new_moment_2 (optional) - T4: New averaged squared gradients
__group_416__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_417__new_weights (optional) - T2: New weights
__group_417__new_gradients (optional) - T3: New gradients
__group_417__new_moment_1 (optional) - T4: New averaged gradients
__group_417__new_moment_2 (optional) - T4: New averaged squared gradients
__group_417__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_418__new_weights (optional) - T2: New weights
__group_418__new_gradients (optional) - T3: New gradients
__group_418__new_moment_1 (optional) - T4: New averaged gradients
__group_418__new_moment_2 (optional) - T4: New averaged squared gradients
__group_418__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_419__new_weights (optional) - T2: New weights
__group_419__new_gradients (optional) - T3: New gradients
__group_419__new_moment_1 (optional) - T4: New averaged gradients
__group_419__new_moment_2 (optional) - T4: New averaged squared gradients
__group_419__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_420__new_weights (optional) - T2: New weights
__group_420__new_gradients (optional) - T3: New gradients
__group_420__new_moment_1 (optional) - T4: New averaged gradients
__group_420__new_moment_2 (optional) - T4: New averaged squared gradients
__group_420__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_421__new_weights (optional) - T2: New weights
__group_421__new_gradients (optional) - T3: New gradients
__group_421__new_moment_1 (optional) - T4: New averaged gradients
__group_421__new_moment_2 (optional) - T4: New averaged squared gradients
__group_421__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_422__new_weights (optional) - T2: New weights
__group_422__new_gradients (optional) - T3: New gradients
__group_422__new_moment_1 (optional) - T4: New averaged gradients
__group_422__new_moment_2 (optional) - T4: New averaged squared gradients
__group_422__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_423__new_weights (optional) - T2: New weights
__group_423__new_gradients (optional) - T3: New gradients
__group_423__new_moment_1 (optional) - T4: New averaged gradients
__group_423__new_moment_2 (optional) - T4: New averaged squared gradients
__group_423__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_424__new_weights (optional) - T2: New weights
__group_424__new_gradients (optional) - T3: New gradients
__group_424__new_moment_1 (optional) - T4: New averaged gradients
__group_424__new_moment_2 (optional) - T4: New averaged squared gradients
__group_424__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_425__new_weights (optional) - T2: New weights
__group_425__new_gradients (optional) - T3: New gradients
__group_425__new_moment_1 (optional) - T4: New averaged gradients
__group_425__new_moment_2 (optional) - T4: New averaged squared gradients
__group_425__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_426__new_weights (optional) - T2: New weights
__group_426__new_gradients (optional) - T3: New gradients
__group_426__new_moment_1 (optional) - T4: New averaged gradients
__group_426__new_moment_2 (optional) - T4: New averaged squared gradients
__group_426__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_427__new_weights (optional) - T2: New weights
__group_427__new_gradients (optional) - T3: New gradients
__group_427__new_moment_1 (optional) - T4: New averaged gradients
__group_427__new_moment_2 (optional) - T4: New averaged squared gradients
__group_427__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_428__new_weights (optional) - T2: New weights
__group_428__new_gradients (optional) - T3: New gradients
__group_428__new_moment_1 (optional) - T4: New averaged gradients
__group_428__new_moment_2 (optional) - T4: New averaged squared gradients
__group_428__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_429__new_weights (optional) - T2: New weights
__group_429__new_gradients (optional) - T3: New gradients
__group_429__new_moment_1 (optional) - T4: New averaged gradients
__group_429__new_moment_2 (optional) - T4: New averaged squared gradients
__group_429__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_430__new_weights (optional) - T2: New weights
__group_430__new_gradients (optional) - T3: New gradients
__group_430__new_moment_1 (optional) - T4: New averaged gradients
__group_430__new_moment_2 (optional) - T4: New averaged squared gradients
__group_430__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_431__new_weights (optional) - T2: New weights
__group_431__new_gradients (optional) - T3: New gradients
__group_431__new_moment_1 (optional) - T4: New averaged gradients
__group_431__new_moment_2 (optional) - T4: New averaged squared gradients
__group_431__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_432__new_weights (optional) - T2: New weights
__group_432__new_gradients (optional) - T3: New gradients
__group_432__new_moment_1 (optional) - T4: New averaged gradients
__group_432__new_moment_2 (optional) - T4: New averaged squared gradients
__group_432__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_433__new_weights (optional) - T2: New weights
__group_433__new_gradients (optional) - T3: New gradients
__group_433__new_moment_1 (optional) - T4: New averaged gradients
__group_433__new_moment_2 (optional) - T4: New averaged squared gradients
__group_433__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_434__new_weights (optional) - T2: New weights
__group_434__new_gradients (optional) - T3: New gradients
__group_434__new_moment_1 (optional) - T4: New averaged gradients
__group_434__new_moment_2 (optional) - T4: New averaged squared gradients
__group_434__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_435__new_weights (optional) - T2: New weights
__group_435__new_gradients (optional) - T3: New gradients
__group_435__new_moment_1 (optional) - T4: New averaged gradients
__group_435__new_moment_2 (optional) - T4: New averaged squared gradients
__group_435__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_436__new_weights (optional) - T2: New weights
__group_436__new_gradients (optional) - T3: New gradients
__group_436__new_moment_1 (optional) - T4: New averaged gradients
__group_436__new_moment_2 (optional) - T4: New averaged squared gradients
__group_436__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_437__new_weights (optional) - T2: New weights
__group_437__new_gradients (optional) - T3: New gradients
__group_437__new_moment_1 (optional) - T4: New averaged gradients
__group_437__new_moment_2 (optional) - T4: New averaged squared gradients
__group_437__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_438__new_weights (optional) - T2: New weights
__group_438__new_gradients (optional) - T3: New gradients
__group_438__new_moment_1 (optional) - T4: New averaged gradients
__group_438__new_moment_2 (optional) - T4: New averaged squared gradients
__group_438__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_439__new_weights (optional) - T2: New weights
__group_439__new_gradients (optional) - T3: New gradients
__group_439__new_moment_1 (optional) - T4: New averaged gradients
__group_439__new_moment_2 (optional) - T4: New averaged squared gradients
__group_439__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_440__new_weights (optional) - T2: New weights
__group_440__new_gradients (optional) - T3: New gradients
__group_440__new_moment_1 (optional) - T4: New averaged gradients
__group_440__new_moment_2 (optional) - T4: New averaged squared gradients
__group_440__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_441__new_weights (optional) - T2: New weights
__group_441__new_gradients (optional) - T3: New gradients
__group_441__new_moment_1 (optional) - T4: New averaged gradients
__group_441__new_moment_2 (optional) - T4: New averaged squared gradients
__group_441__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_442__new_weights (optional) - T2: New weights
__group_442__new_gradients (optional) - T3: New gradients
__group_442__new_moment_1 (optional) - T4: New averaged gradients
__group_442__new_moment_2 (optional) - T4: New averaged squared gradients
__group_442__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_443__new_weights (optional) - T2: New weights
__group_443__new_gradients (optional) - T3: New gradients
__group_443__new_moment_1 (optional) - T4: New averaged gradients
__group_443__new_moment_2 (optional) - T4: New averaged squared gradients
__group_443__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_444__new_weights (optional) - T2: New weights
__group_444__new_gradients (optional) - T3: New gradients
__group_444__new_moment_1 (optional) - T4: New averaged gradients
__group_444__new_moment_2 (optional) - T4: New averaged squared gradients
__group_444__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_445__new_weights (optional) - T2: New weights
__group_445__new_gradients (optional) - T3: New gradients
__group_445__new_moment_1 (optional) - T4: New averaged gradients
__group_445__new_moment_2 (optional) - T4: New averaged squared gradients
__group_445__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_446__new_weights (optional) - T2: New weights
__group_446__new_gradients (optional) - T3: New gradients
__group_446__new_moment_1 (optional) - T4: New averaged gradients
__group_446__new_moment_2 (optional) - T4: New averaged squared gradients
__group_446__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_447__new_weights (optional) - T2: New weights
__group_447__new_gradients (optional) - T3: New gradients
__group_447__new_moment_1 (optional) - T4: New averaged gradients
__group_447__new_moment_2 (optional) - T4: New averaged squared gradients
__group_447__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_448__new_weights (optional) - T2: New weights
__group_448__new_gradients (optional) - T3: New gradients
__group_448__new_moment_1 (optional) - T4: New averaged gradients
__group_448__new_moment_2 (optional) - T4: New averaged squared gradients
__group_448__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_449__new_weights (optional) - T2: New weights
__group_449__new_gradients (optional) - T3: New gradients
__group_449__new_moment_1 (optional) - T4: New averaged gradients
__group_449__new_moment_2 (optional) - T4: New averaged squared gradients
__group_449__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_450__new_weights (optional) - T2: New weights
__group_450__new_gradients (optional) - T3: New gradients
__group_450__new_moment_1 (optional) - T4: New averaged gradients
__group_450__new_moment_2 (optional) - T4: New averaged squared gradients
__group_450__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_451__new_weights (optional) - T2: New weights
__group_451__new_gradients (optional) - T3: New gradients
__group_451__new_moment_1 (optional) - T4: New averaged gradients
__group_451__new_moment_2 (optional) - T4: New averaged squared gradients
__group_451__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_452__new_weights (optional) - T2: New weights
__group_452__new_gradients (optional) - T3: New gradients
__group_452__new_moment_1 (optional) - T4: New averaged gradients
__group_452__new_moment_2 (optional) - T4: New averaged squared gradients
__group_452__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_453__new_weights (optional) - T2: New weights
__group_453__new_gradients (optional) - T3: New gradients
__group_453__new_moment_1 (optional) - T4: New averaged gradients
__group_453__new_moment_2 (optional) - T4: New averaged squared gradients
__group_453__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_454__new_weights (optional) - T2: New weights
__group_454__new_gradients (optional) - T3: New gradients
__group_454__new_moment_1 (optional) - T4: New averaged gradients
__group_454__new_moment_2 (optional) - T4: New averaged squared gradients
__group_454__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_455__new_weights (optional) - T2: New weights
__group_455__new_gradients (optional) - T3: New gradients
__group_455__new_moment_1 (optional) - T4: New averaged gradients
__group_455__new_moment_2 (optional) - T4: New averaged squared gradients
__group_455__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_456__new_weights (optional) - T2: New weights
__group_456__new_gradients (optional) - T3: New gradients
__group_456__new_moment_1 (optional) - T4: New averaged gradients
__group_456__new_moment_2 (optional) - T4: New averaged squared gradients
__group_456__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_457__new_weights (optional) - T2: New weights
__group_457__new_gradients (optional) - T3: New gradients
__group_457__new_moment_1 (optional) - T4: New averaged gradients
__group_457__new_moment_2 (optional) - T4: New averaged squared gradients
__group_457__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_458__new_weights (optional) - T2: New weights
__group_458__new_gradients (optional) - T3: New gradients
__group_458__new_moment_1 (optional) - T4: New averaged gradients
__group_458__new_moment_2 (optional) - T4: New averaged squared gradients
__group_458__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_459__new_weights (optional) - T2: New weights
__group_459__new_gradients (optional) - T3: New gradients
__group_459__new_moment_1 (optional) - T4: New averaged gradients
__group_459__new_moment_2 (optional) - T4: New averaged squared gradients
__group_459__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_460__new_weights (optional) - T2: New weights
__group_460__new_gradients (optional) - T3: New gradients
__group_460__new_moment_1 (optional) - T4: New averaged gradients
__group_460__new_moment_2 (optional) - T4: New averaged squared gradients
__group_460__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_461__new_weights (optional) - T2: New weights
__group_461__new_gradients (optional) - T3: New gradients
__group_461__new_moment_1 (optional) - T4: New averaged gradients
__group_461__new_moment_2 (optional) - T4: New averaged squared gradients
__group_461__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_462__new_weights (optional) - T2: New weights
__group_462__new_gradients (optional) - T3: New gradients
__group_462__new_moment_1 (optional) - T4: New averaged gradients
__group_462__new_moment_2 (optional) - T4: New averaged squared gradients
__group_462__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_463__new_weights (optional) - T2: New weights
__group_463__new_gradients (optional) - T3: New gradients
__group_463__new_moment_1 (optional) - T4: New averaged gradients
__group_463__new_moment_2 (optional) - T4: New averaged squared gradients
__group_463__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_464__new_weights (optional) - T2: New weights
__group_464__new_gradients (optional) - T3: New gradients
__group_464__new_moment_1 (optional) - T4: New averaged gradients
__group_464__new_moment_2 (optional) - T4: New averaged squared gradients
__group_464__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_465__new_weights (optional) - T2: New weights
__group_465__new_gradients (optional) - T3: New gradients
__group_465__new_moment_1 (optional) - T4: New averaged gradients
__group_465__new_moment_2 (optional) - T4: New averaged squared gradients
__group_465__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_466__new_weights (optional) - T2: New weights
__group_466__new_gradients (optional) - T3: New gradients
__group_466__new_moment_1 (optional) - T4: New averaged gradients
__group_466__new_moment_2 (optional) - T4: New averaged squared gradients
__group_466__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_467__new_weights (optional) - T2: New weights
__group_467__new_gradients (optional) - T3: New gradients
__group_467__new_moment_1 (optional) - T4: New averaged gradients
__group_467__new_moment_2 (optional) - T4: New averaged squared gradients
__group_467__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_468__new_weights (optional) - T2: New weights
__group_468__new_gradients (optional) - T3: New gradients
__group_468__new_moment_1 (optional) - T4: New averaged gradients
__group_468__new_moment_2 (optional) - T4: New averaged squared gradients
__group_468__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_469__new_weights (optional) - T2: New weights
__group_469__new_gradients (optional) - T3: New gradients
__group_469__new_moment_1 (optional) - T4: New averaged gradients
__group_469__new_moment_2 (optional) - T4: New averaged squared gradients
__group_469__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_470__new_weights (optional) - T2: New weights
__group_470__new_gradients (optional) - T3: New gradients
__group_470__new_moment_1 (optional) - T4: New averaged gradients
__group_470__new_moment_2 (optional) - T4: New averaged squared gradients
__group_470__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_471__new_weights (optional) - T2: New weights
__group_471__new_gradients (optional) - T3: New gradients
__group_471__new_moment_1 (optional) - T4: New averaged gradients
__group_471__new_moment_2 (optional) - T4: New averaged squared gradients
__group_471__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_472__new_weights (optional) - T2: New weights
__group_472__new_gradients (optional) - T3: New gradients
__group_472__new_moment_1 (optional) - T4: New averaged gradients
__group_472__new_moment_2 (optional) - T4: New averaged squared gradients
__group_472__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_473__new_weights (optional) - T2: New weights
__group_473__new_gradients (optional) - T3: New gradients
__group_473__new_moment_1 (optional) - T4: New averaged gradients
__group_473__new_moment_2 (optional) - T4: New averaged squared gradients
__group_473__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_474__new_weights (optional) - T2: New weights
__group_474__new_gradients (optional) - T3: New gradients
__group_474__new_moment_1 (optional) - T4: New averaged gradients
__group_474__new_moment_2 (optional) - T4: New averaged squared gradients
__group_474__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_475__new_weights (optional) - T2: New weights
__group_475__new_gradients (optional) - T3: New gradients
__group_475__new_moment_1 (optional) - T4: New averaged gradients
__group_475__new_moment_2 (optional) - T4: New averaged squared gradients
__group_475__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_476__new_weights (optional) - T2: New weights
__group_476__new_gradients (optional) - T3: New gradients
__group_476__new_moment_1 (optional) - T4: New averaged gradients
__group_476__new_moment_2 (optional) - T4: New averaged squared gradients
__group_476__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_477__new_weights (optional) - T2: New weights
__group_477__new_gradients (optional) - T3: New gradients
__group_477__new_moment_1 (optional) - T4: New averaged gradients
__group_477__new_moment_2 (optional) - T4: New averaged squared gradients
__group_477__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_478__new_weights (optional) - T2: New weights
__group_478__new_gradients (optional) - T3: New gradients
__group_478__new_moment_1 (optional) - T4: New averaged gradients
__group_478__new_moment_2 (optional) - T4: New averaged squared gradients
__group_478__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_479__new_weights (optional) - T2: New weights
__group_479__new_gradients (optional) - T3: New gradients
__group_479__new_moment_1 (optional) - T4: New averaged gradients
__group_479__new_moment_2 (optional) - T4: New averaged squared gradients
__group_479__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_480__new_weights (optional) - T2: New weights
__group_480__new_gradients (optional) - T3: New gradients
__group_480__new_moment_1 (optional) - T4: New averaged gradients
__group_480__new_moment_2 (optional) - T4: New averaged squared gradients
__group_480__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_481__new_weights (optional) - T2: New weights
__group_481__new_gradients (optional) - T3: New gradients
__group_481__new_moment_1 (optional) - T4: New averaged gradients
__group_481__new_moment_2 (optional) - T4: New averaged squared gradients
__group_481__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_482__new_weights (optional) - T2: New weights
__group_482__new_gradients (optional) - T3: New gradients
__group_482__new_moment_1 (optional) - T4: New averaged gradients
__group_482__new_moment_2 (optional) - T4: New averaged squared gradients
__group_482__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_483__new_weights (optional) - T2: New weights
__group_483__new_gradients (optional) - T3: New gradients
__group_483__new_moment_1 (optional) - T4: New averaged gradients
__group_483__new_moment_2 (optional) - T4: New averaged squared gradients
__group_483__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_484__new_weights (optional) - T2: New weights
__group_484__new_gradients (optional) - T3: New gradients
__group_484__new_moment_1 (optional) - T4: New averaged gradients
__group_484__new_moment_2 (optional) - T4: New averaged squared gradients
__group_484__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_485__new_weights (optional) - T2: New weights
__group_485__new_gradients (optional) - T3: New gradients
__group_485__new_moment_1 (optional) - T4: New averaged gradients
__group_485__new_moment_2 (optional) - T4: New averaged squared gradients
__group_485__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_486__new_weights (optional) - T2: New weights
__group_486__new_gradients (optional) - T3: New gradients
__group_486__new_moment_1 (optional) - T4: New averaged gradients
__group_486__new_moment_2 (optional) - T4: New averaged squared gradients
__group_486__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_487__new_weights (optional) - T2: New weights
__group_487__new_gradients (optional) - T3: New gradients
__group_487__new_moment_1 (optional) - T4: New averaged gradients
__group_487__new_moment_2 (optional) - T4: New averaged squared gradients
__group_487__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_488__new_weights (optional) - T2: New weights
__group_488__new_gradients (optional) - T3: New gradients
__group_488__new_moment_1 (optional) - T4: New averaged gradients
__group_488__new_moment_2 (optional) - T4: New averaged squared gradients
__group_488__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_489__new_weights (optional) - T2: New weights
__group_489__new_gradients (optional) - T3: New gradients
__group_489__new_moment_1 (optional) - T4: New averaged gradients
__group_489__new_moment_2 (optional) - T4: New averaged squared gradients
__group_489__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_490__new_weights (optional) - T2: New weights
__group_490__new_gradients (optional) - T3: New gradients
__group_490__new_moment_1 (optional) - T4: New averaged gradients
__group_490__new_moment_2 (optional) - T4: New averaged squared gradients
__group_490__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_491__new_weights (optional) - T2: New weights
__group_491__new_gradients (optional) - T3: New gradients
__group_491__new_moment_1 (optional) - T4: New averaged gradients
__group_491__new_moment_2 (optional) - T4: New averaged squared gradients
__group_491__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_492__new_weights (optional) - T2: New weights
__group_492__new_gradients (optional) - T3: New gradients
__group_492__new_moment_1 (optional) - T4: New averaged gradients
__group_492__new_moment_2 (optional) - T4: New averaged squared gradients
__group_492__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_493__new_weights (optional) - T2: New weights
__group_493__new_gradients (optional) - T3: New gradients
__group_493__new_moment_1 (optional) - T4: New averaged gradients
__group_493__new_moment_2 (optional) - T4: New averaged squared gradients
__group_493__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_494__new_weights (optional) - T2: New weights
__group_494__new_gradients (optional) - T3: New gradients
__group_494__new_moment_1 (optional) - T4: New averaged gradients
__group_494__new_moment_2 (optional) - T4: New averaged squared gradients
__group_494__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_495__new_weights (optional) - T2: New weights
__group_495__new_gradients (optional) - T3: New gradients
__group_495__new_moment_1 (optional) - T4: New averaged gradients
__group_495__new_moment_2 (optional) - T4: New averaged squared gradients
__group_495__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_496__new_weights (optional) - T2: New weights
__group_496__new_gradients (optional) - T3: New gradients
__group_496__new_moment_1 (optional) - T4: New averaged gradients
__group_496__new_moment_2 (optional) - T4: New averaged squared gradients
__group_496__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_497__new_weights (optional) - T2: New weights
__group_497__new_gradients (optional) - T3: New gradients
__group_497__new_moment_1 (optional) - T4: New averaged gradients
__group_497__new_moment_2 (optional) - T4: New averaged squared gradients
__group_497__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_498__new_weights (optional) - T2: New weights
__group_498__new_gradients (optional) - T3: New gradients
__group_498__new_moment_1 (optional) - T4: New averaged gradients
__group_498__new_moment_2 (optional) - T4: New averaged squared gradients
__group_498__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_499__new_weights (optional) - T2: New weights
__group_499__new_gradients (optional) - T3: New gradients
__group_499__new_moment_1 (optional) - T4: New averaged gradients
__group_499__new_moment_2 (optional) - T4: New averaged squared gradients
__group_499__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_500__new_weights (optional) - T2: New weights
__group_500__new_gradients (optional) - T3: New gradients
__group_500__new_moment_1 (optional) - T4: New averaged gradients
__group_500__new_moment_2 (optional) - T4: New averaged squared gradients
__group_500__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_501__new_weights (optional) - T2: New weights
__group_501__new_gradients (optional) - T3: New gradients
__group_501__new_moment_1 (optional) - T4: New averaged gradients
__group_501__new_moment_2 (optional) - T4: New averaged squared gradients
__group_501__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_502__new_weights (optional) - T2: New weights
__group_502__new_gradients (optional) - T3: New gradients
__group_502__new_moment_1 (optional) - T4: New averaged gradients
__group_502__new_moment_2 (optional) - T4: New averaged squared gradients
__group_502__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_503__new_weights (optional) - T2: New weights
__group_503__new_gradients (optional) - T3: New gradients
__group_503__new_moment_1 (optional) - T4: New averaged gradients
__group_503__new_moment_2 (optional) - T4: New averaged squared gradients
__group_503__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_504__new_weights (optional) - T2: New weights
__group_504__new_gradients (optional) - T3: New gradients
__group_504__new_moment_1 (optional) - T4: New averaged gradients
__group_504__new_moment_2 (optional) - T4: New averaged squared gradients
__group_504__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_505__new_weights (optional) - T2: New weights
__group_505__new_gradients (optional) - T3: New gradients
__group_505__new_moment_1 (optional) - T4: New averaged gradients
__group_505__new_moment_2 (optional) - T4: New averaged squared gradients
__group_505__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_506__new_weights (optional) - T2: New weights
__group_506__new_gradients (optional) - T3: New gradients
__group_506__new_moment_1 (optional) - T4: New averaged gradients
__group_506__new_moment_2 (optional) - T4: New averaged squared gradients
__group_506__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_507__new_weights (optional) - T2: New weights
__group_507__new_gradients (optional) - T3: New gradients
__group_507__new_moment_1 (optional) - T4: New averaged gradients
__group_507__new_moment_2 (optional) - T4: New averaged squared gradients
__group_507__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_508__new_weights (optional) - T2: New weights
__group_508__new_gradients (optional) - T3: New gradients
__group_508__new_moment_1 (optional) - T4: New averaged gradients
__group_508__new_moment_2 (optional) - T4: New averaged squared gradients
__group_508__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_509__new_weights (optional) - T2: New weights
__group_509__new_gradients (optional) - T3: New gradients
__group_509__new_moment_1 (optional) - T4: New averaged gradients
__group_509__new_moment_2 (optional) - T4: New averaged squared gradients
__group_509__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_510__new_weights (optional) - T2: New weights
__group_510__new_gradients (optional) - T3: New gradients
__group_510__new_moment_1 (optional) - T4: New averaged gradients
__group_510__new_moment_2 (optional) - T4: New averaged squared gradients
__group_510__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_511__new_weights (optional) - T2: New weights
__group_511__new_gradients (optional) - T3: New gradients
__group_511__new_moment_1 (optional) - T4: New averaged gradients
__group_511__new_moment_2 (optional) - T4: New averaged squared gradients
__group_511__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_512__new_weights (optional) - T2: New weights
__group_512__new_gradients (optional) - T3: New gradients
__group_512__new_moment_1 (optional) - T4: New averaged gradients
__group_512__new_moment_2 (optional) - T4: New averaged squared gradients
__group_512__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_513__new_weights (optional) - T2: New weights
__group_513__new_gradients (optional) - T3: New gradients
__group_513__new_moment_1 (optional) - T4: New averaged gradients
__group_513__new_moment_2 (optional) - T4: New averaged squared gradients
__group_513__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_514__new_weights (optional) - T2: New weights
__group_514__new_gradients (optional) - T3: New gradients
__group_514__new_moment_1 (optional) - T4: New averaged gradients
__group_514__new_moment_2 (optional) - T4: New averaged squared gradients
__group_514__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_515__new_weights (optional) - T2: New weights
__group_515__new_gradients (optional) - T3: New gradients
__group_515__new_moment_1 (optional) - T4: New averaged gradients
__group_515__new_moment_2 (optional) - T4: New averaged squared gradients
__group_515__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_516__new_weights (optional) - T2: New weights
__group_516__new_gradients (optional) - T3: New gradients
__group_516__new_moment_1 (optional) - T4: New averaged gradients
__group_516__new_moment_2 (optional) - T4: New averaged squared gradients
__group_516__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_517__new_weights (optional) - T2: New weights
__group_517__new_gradients (optional) - T3: New gradients
__group_517__new_moment_1 (optional) - T4: New averaged gradients
__group_517__new_moment_2 (optional) - T4: New averaged squared gradients
__group_517__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_518__new_weights (optional) - T2: New weights
__group_518__new_gradients (optional) - T3: New gradients
__group_518__new_moment_1 (optional) - T4: New averaged gradients
__group_518__new_moment_2 (optional) - T4: New averaged squared gradients
__group_518__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_519__new_weights (optional) - T2: New weights
__group_519__new_gradients (optional) - T3: New gradients
__group_519__new_moment_1 (optional) - T4: New averaged gradients
__group_519__new_moment_2 (optional) - T4: New averaged squared gradients
__group_519__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_520__new_weights (optional) - T2: New weights
__group_520__new_gradients (optional) - T3: New gradients
__group_520__new_moment_1 (optional) - T4: New averaged gradients
__group_520__new_moment_2 (optional) - T4: New averaged squared gradients
__group_520__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_521__new_weights (optional) - T2: New weights
__group_521__new_gradients (optional) - T3: New gradients
__group_521__new_moment_1 (optional) - T4: New averaged gradients
__group_521__new_moment_2 (optional) - T4: New averaged squared gradients
__group_521__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_522__new_weights (optional) - T2: New weights
__group_522__new_gradients (optional) - T3: New gradients
__group_522__new_moment_1 (optional) - T4: New averaged gradients
__group_522__new_moment_2 (optional) - T4: New averaged squared gradients
__group_522__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_523__new_weights (optional) - T2: New weights
__group_523__new_gradients (optional) - T3: New gradients
__group_523__new_moment_1 (optional) - T4: New averaged gradients
__group_523__new_moment_2 (optional) - T4: New averaged squared gradients
__group_523__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_524__new_weights (optional) - T2: New weights
__group_524__new_gradients (optional) - T3: New gradients
__group_524__new_moment_1 (optional) - T4: New averaged gradients
__group_524__new_moment_2 (optional) - T4: New averaged squared gradients
__group_524__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_525__new_weights (optional) - T2: New weights
__group_525__new_gradients (optional) - T3: New gradients
__group_525__new_moment_1 (optional) - T4: New averaged gradients
__group_525__new_moment_2 (optional) - T4: New averaged squared gradients
__group_525__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_526__new_weights (optional) - T2: New weights
__group_526__new_gradients (optional) - T3: New gradients
__group_526__new_moment_1 (optional) - T4: New averaged gradients
__group_526__new_moment_2 (optional) - T4: New averaged squared gradients
__group_526__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_527__new_weights (optional) - T2: New weights
__group_527__new_gradients (optional) - T3: New gradients
__group_527__new_moment_1 (optional) - T4: New averaged gradients
__group_527__new_moment_2 (optional) - T4: New averaged squared gradients
__group_527__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_528__new_weights (optional) - T2: New weights
__group_528__new_gradients (optional) - T3: New gradients
__group_528__new_moment_1 (optional) - T4: New averaged gradients
__group_528__new_moment_2 (optional) - T4: New averaged squared gradients
__group_528__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_529__new_weights (optional) - T2: New weights
__group_529__new_gradients (optional) - T3: New gradients
__group_529__new_moment_1 (optional) - T4: New averaged gradients
__group_529__new_moment_2 (optional) - T4: New averaged squared gradients
__group_529__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_530__new_weights (optional) - T2: New weights
__group_530__new_gradients (optional) - T3: New gradients
__group_530__new_moment_1 (optional) - T4: New averaged gradients
__group_530__new_moment_2 (optional) - T4: New averaged squared gradients
__group_530__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_531__new_weights (optional) - T2: New weights
__group_531__new_gradients (optional) - T3: New gradients
__group_531__new_moment_1 (optional) - T4: New averaged gradients
__group_531__new_moment_2 (optional) - T4: New averaged squared gradients
__group_531__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_532__new_weights (optional) - T2: New weights
__group_532__new_gradients (optional) - T3: New gradients
__group_532__new_moment_1 (optional) - T4: New averaged gradients
__group_532__new_moment_2 (optional) - T4: New averaged squared gradients
__group_532__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_533__new_weights (optional) - T2: New weights
__group_533__new_gradients (optional) - T3: New gradients
__group_533__new_moment_1 (optional) - T4: New averaged gradients
__group_533__new_moment_2 (optional) - T4: New averaged squared gradients
__group_533__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_534__new_weights (optional) - T2: New weights
__group_534__new_gradients (optional) - T3: New gradients
__group_534__new_moment_1 (optional) - T4: New averaged gradients
__group_534__new_moment_2 (optional) - T4: New averaged squared gradients
__group_534__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_535__new_weights (optional) - T2: New weights
__group_535__new_gradients (optional) - T3: New gradients
__group_535__new_moment_1 (optional) - T4: New averaged gradients
__group_535__new_moment_2 (optional) - T4: New averaged squared gradients
__group_535__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_536__new_weights (optional) - T2: New weights
__group_536__new_gradients (optional) - T3: New gradients
__group_536__new_moment_1 (optional) - T4: New averaged gradients
__group_536__new_moment_2 (optional) - T4: New averaged squared gradients
__group_536__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_537__new_weights (optional) - T2: New weights
__group_537__new_gradients (optional) - T3: New gradients
__group_537__new_moment_1 (optional) - T4: New averaged gradients
__group_537__new_moment_2 (optional) - T4: New averaged squared gradients
__group_537__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_538__new_weights (optional) - T2: New weights
__group_538__new_gradients (optional) - T3: New gradients
__group_538__new_moment_1 (optional) - T4: New averaged gradients
__group_538__new_moment_2 (optional) - T4: New averaged squared gradients
__group_538__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_539__new_weights (optional) - T2: New weights
__group_539__new_gradients (optional) - T3: New gradients
__group_539__new_moment_1 (optional) - T4: New averaged gradients
__group_539__new_moment_2 (optional) - T4: New averaged squared gradients
__group_539__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_540__new_weights (optional) - T2: New weights
__group_540__new_gradients (optional) - T3: New gradients
__group_540__new_moment_1 (optional) - T4: New averaged gradients
__group_540__new_moment_2 (optional) - T4: New averaged squared gradients
__group_540__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_541__new_weights (optional) - T2: New weights
__group_541__new_gradients (optional) - T3: New gradients
__group_541__new_moment_1 (optional) - T4: New averaged gradients
__group_541__new_moment_2 (optional) - T4: New averaged squared gradients
__group_541__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_542__new_weights (optional) - T2: New weights
__group_542__new_gradients (optional) - T3: New gradients
__group_542__new_moment_1 (optional) - T4: New averaged gradients
__group_542__new_moment_2 (optional) - T4: New averaged squared gradients
__group_542__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_543__new_weights (optional) - T2: New weights
__group_543__new_gradients (optional) - T3: New gradients
__group_543__new_moment_1 (optional) - T4: New averaged gradients
__group_543__new_moment_2 (optional) - T4: New averaged squared gradients
__group_543__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_544__new_weights (optional) - T2: New weights
__group_544__new_gradients (optional) - T3: New gradients
__group_544__new_moment_1 (optional) - T4: New averaged gradients
__group_544__new_moment_2 (optional) - T4: New averaged squared gradients
__group_544__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_545__new_weights (optional) - T2: New weights
__group_545__new_gradients (optional) - T3: New gradients
__group_545__new_moment_1 (optional) - T4: New averaged gradients
__group_545__new_moment_2 (optional) - T4: New averaged squared gradients
__group_545__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_546__new_weights (optional) - T2: New weights
__group_546__new_gradients (optional) - T3: New gradients
__group_546__new_moment_1 (optional) - T4: New averaged gradients
__group_546__new_moment_2 (optional) - T4: New averaged squared gradients
__group_546__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_547__new_weights (optional) - T2: New weights
__group_547__new_gradients (optional) - T3: New gradients
__group_547__new_moment_1 (optional) - T4: New averaged gradients
__group_547__new_moment_2 (optional) - T4: New averaged squared gradients
__group_547__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_548__new_weights (optional) - T2: New weights
__group_548__new_gradients (optional) - T3: New gradients
__group_548__new_moment_1 (optional) - T4: New averaged gradients
__group_548__new_moment_2 (optional) - T4: New averaged squared gradients
__group_548__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_549__new_weights (optional) - T2: New weights
__group_549__new_gradients (optional) - T3: New gradients
__group_549__new_moment_1 (optional) - T4: New averaged gradients
__group_549__new_moment_2 (optional) - T4: New averaged squared gradients
__group_549__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_550__new_weights (optional) - T2: New weights
__group_550__new_gradients (optional) - T3: New gradients
__group_550__new_moment_1 (optional) - T4: New averaged gradients
__group_550__new_moment_2 (optional) - T4: New averaged squared gradients
__group_550__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_551__new_weights (optional) - T2: New weights
__group_551__new_gradients (optional) - T3: New gradients
__group_551__new_moment_1 (optional) - T4: New averaged gradients
__group_551__new_moment_2 (optional) - T4: New averaged squared gradients
__group_551__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_552__new_weights (optional) - T2: New weights
__group_552__new_gradients (optional) - T3: New gradients
__group_552__new_moment_1 (optional) - T4: New averaged gradients
__group_552__new_moment_2 (optional) - T4: New averaged squared gradients
__group_552__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_553__new_weights (optional) - T2: New weights
__group_553__new_gradients (optional) - T3: New gradients
__group_553__new_moment_1 (optional) - T4: New averaged gradients
__group_553__new_moment_2 (optional) - T4: New averaged squared gradients
__group_553__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_554__new_weights (optional) - T2: New weights
__group_554__new_gradients (optional) - T3: New gradients
__group_554__new_moment_1 (optional) - T4: New averaged gradients
__group_554__new_moment_2 (optional) - T4: New averaged squared gradients
__group_554__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_555__new_weights (optional) - T2: New weights
__group_555__new_gradients (optional) - T3: New gradients
__group_555__new_moment_1 (optional) - T4: New averaged gradients
__group_555__new_moment_2 (optional) - T4: New averaged squared gradients
__group_555__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_556__new_weights (optional) - T2: New weights
__group_556__new_gradients (optional) - T3: New gradients
__group_556__new_moment_1 (optional) - T4: New averaged gradients
__group_556__new_moment_2 (optional) - T4: New averaged squared gradients
__group_556__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_557__new_weights (optional) - T2: New weights
__group_557__new_gradients (optional) - T3: New gradients
__group_557__new_moment_1 (optional) - T4: New averaged gradients
__group_557__new_moment_2 (optional) - T4: New averaged squared gradients
__group_557__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_558__new_weights (optional) - T2: New weights
__group_558__new_gradients (optional) - T3: New gradients
__group_558__new_moment_1 (optional) - T4: New averaged gradients
__group_558__new_moment_2 (optional) - T4: New averaged squared gradients
__group_558__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_559__new_weights (optional) - T2: New weights
__group_559__new_gradients (optional) - T3: New gradients
__group_559__new_moment_1 (optional) - T4: New averaged gradients
__group_559__new_moment_2 (optional) - T4: New averaged squared gradients
__group_559__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_560__new_weights (optional) - T2: New weights
__group_560__new_gradients (optional) - T3: New gradients
__group_560__new_moment_1 (optional) - T4: New averaged gradients
__group_560__new_moment_2 (optional) - T4: New averaged squared gradients
__group_560__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_561__new_weights (optional) - T2: New weights
__group_561__new_gradients (optional) - T3: New gradients
__group_561__new_moment_1 (optional) - T4: New averaged gradients
__group_561__new_moment_2 (optional) - T4: New averaged squared gradients
__group_561__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_562__new_weights (optional) - T2: New weights
__group_562__new_gradients (optional) - T3: New gradients
__group_562__new_moment_1 (optional) - T4: New averaged gradients
__group_562__new_moment_2 (optional) - T4: New averaged squared gradients
__group_562__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_563__new_weights (optional) - T2: New weights
__group_563__new_gradients (optional) - T3: New gradients
__group_563__new_moment_1 (optional) - T4: New averaged gradients
__group_563__new_moment_2 (optional) - T4: New averaged squared gradients
__group_563__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_564__new_weights (optional) - T2: New weights
__group_564__new_gradients (optional) - T3: New gradients
__group_564__new_moment_1 (optional) - T4: New averaged gradients
__group_564__new_moment_2 (optional) - T4: New averaged squared gradients
__group_564__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_565__new_weights (optional) - T2: New weights
__group_565__new_gradients (optional) - T3: New gradients
__group_565__new_moment_1 (optional) - T4: New averaged gradients
__group_565__new_moment_2 (optional) - T4: New averaged squared gradients
__group_565__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_566__new_weights (optional) - T2: New weights
__group_566__new_gradients (optional) - T3: New gradients
__group_566__new_moment_1 (optional) - T4: New averaged gradients
__group_566__new_moment_2 (optional) - T4: New averaged squared gradients
__group_566__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_567__new_weights (optional) - T2: New weights
__group_567__new_gradients (optional) - T3: New gradients
__group_567__new_moment_1 (optional) - T4: New averaged gradients
__group_567__new_moment_2 (optional) - T4: New averaged squared gradients
__group_567__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_568__new_weights (optional) - T2: New weights
__group_568__new_gradients (optional) - T3: New gradients
__group_568__new_moment_1 (optional) - T4: New averaged gradients
__group_568__new_moment_2 (optional) - T4: New averaged squared gradients
__group_568__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_569__new_weights (optional) - T2: New weights
__group_569__new_gradients (optional) - T3: New gradients
__group_569__new_moment_1 (optional) - T4: New averaged gradients
__group_569__new_moment_2 (optional) - T4: New averaged squared gradients
__group_569__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_570__new_weights (optional) - T2: New weights
__group_570__new_gradients (optional) - T3: New gradients
__group_570__new_moment_1 (optional) - T4: New averaged gradients
__group_570__new_moment_2 (optional) - T4: New averaged squared gradients
__group_570__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_571__new_weights (optional) - T2: New weights
__group_571__new_gradients (optional) - T3: New gradients
__group_571__new_moment_1 (optional) - T4: New averaged gradients
__group_571__new_moment_2 (optional) - T4: New averaged squared gradients
__group_571__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_572__new_weights (optional) - T2: New weights
__group_572__new_gradients (optional) - T3: New gradients
__group_572__new_moment_1 (optional) - T4: New averaged gradients
__group_572__new_moment_2 (optional) - T4: New averaged squared gradients
__group_572__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_573__new_weights (optional) - T2: New weights
__group_573__new_gradients (optional) - T3: New gradients
__group_573__new_moment_1 (optional) - T4: New averaged gradients
__group_573__new_moment_2 (optional) - T4: New averaged squared gradients
__group_573__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_574__new_weights (optional) - T2: New weights
__group_574__new_gradients (optional) - T3: New gradients
__group_574__new_moment_1 (optional) - T4: New averaged gradients
__group_574__new_moment_2 (optional) - T4: New averaged squared gradients
__group_574__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_575__new_weights (optional) - T2: New weights
__group_575__new_gradients (optional) - T3: New gradients
__group_575__new_moment_1 (optional) - T4: New averaged gradients
__group_575__new_moment_2 (optional) - T4: New averaged squared gradients
__group_575__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_576__new_weights (optional) - T2: New weights
__group_576__new_gradients (optional) - T3: New gradients
__group_576__new_moment_1 (optional) - T4: New averaged gradients
__group_576__new_moment_2 (optional) - T4: New averaged squared gradients
__group_576__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_577__new_weights (optional) - T2: New weights
__group_577__new_gradients (optional) - T3: New gradients
__group_577__new_moment_1 (optional) - T4: New averaged gradients
__group_577__new_moment_2 (optional) - T4: New averaged squared gradients
__group_577__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_578__new_weights (optional) - T2: New weights
__group_578__new_gradients (optional) - T3: New gradients
__group_578__new_moment_1 (optional) - T4: New averaged gradients
__group_578__new_moment_2 (optional) - T4: New averaged squared gradients
__group_578__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_579__new_weights (optional) - T2: New weights
__group_579__new_gradients (optional) - T3: New gradients
__group_579__new_moment_1 (optional) - T4: New averaged gradients
__group_579__new_moment_2 (optional) - T4: New averaged squared gradients
__group_579__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_580__new_weights (optional) - T2: New weights
__group_580__new_gradients (optional) - T3: New gradients
__group_580__new_moment_1 (optional) - T4: New averaged gradients
__group_580__new_moment_2 (optional) - T4: New averaged squared gradients
__group_580__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_581__new_weights (optional) - T2: New weights
__group_581__new_gradients (optional) - T3: New gradients
__group_581__new_moment_1 (optional) - T4: New averaged gradients
__group_581__new_moment_2 (optional) - T4: New averaged squared gradients
__group_581__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_582__new_weights (optional) - T2: New weights
__group_582__new_gradients (optional) - T3: New gradients
__group_582__new_moment_1 (optional) - T4: New averaged gradients
__group_582__new_moment_2 (optional) - T4: New averaged squared gradients
__group_582__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_583__new_weights (optional) - T2: New weights
__group_583__new_gradients (optional) - T3: New gradients
__group_583__new_moment_1 (optional) - T4: New averaged gradients
__group_583__new_moment_2 (optional) - T4: New averaged squared gradients
__group_583__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_584__new_weights (optional) - T2: New weights
__group_584__new_gradients (optional) - T3: New gradients
__group_584__new_moment_1 (optional) - T4: New averaged gradients
__group_584__new_moment_2 (optional) - T4: New averaged squared gradients
__group_584__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_585__new_weights (optional) - T2: New weights
__group_585__new_gradients (optional) - T3: New gradients
__group_585__new_moment_1 (optional) - T4: New averaged gradients
__group_585__new_moment_2 (optional) - T4: New averaged squared gradients
__group_585__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_586__new_weights (optional) - T2: New weights
__group_586__new_gradients (optional) - T3: New gradients
__group_586__new_moment_1 (optional) - T4: New averaged gradients
__group_586__new_moment_2 (optional) - T4: New averaged squared gradients
__group_586__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_587__new_weights (optional) - T2: New weights
__group_587__new_gradients (optional) - T3: New gradients
__group_587__new_moment_1 (optional) - T4: New averaged gradients
__group_587__new_moment_2 (optional) - T4: New averaged squared gradients
__group_587__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_588__new_weights (optional) - T2: New weights
__group_588__new_gradients (optional) - T3: New gradients
__group_588__new_moment_1 (optional) - T4: New averaged gradients
__group_588__new_moment_2 (optional) - T4: New averaged squared gradients
__group_588__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_589__new_weights (optional) - T2: New weights
__group_589__new_gradients (optional) - T3: New gradients
__group_589__new_moment_1 (optional) - T4: New averaged gradients
__group_589__new_moment_2 (optional) - T4: New averaged squared gradients
__group_589__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_590__new_weights (optional) - T2: New weights
__group_590__new_gradients (optional) - T3: New gradients
__group_590__new_moment_1 (optional) - T4: New averaged gradients
__group_590__new_moment_2 (optional) - T4: New averaged squared gradients
__group_590__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_591__new_weights (optional) - T2: New weights
__group_591__new_gradients (optional) - T3: New gradients
__group_591__new_moment_1 (optional) - T4: New averaged gradients
__group_591__new_moment_2 (optional) - T4: New averaged squared gradients
__group_591__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_592__new_weights (optional) - T2: New weights
__group_592__new_gradients (optional) - T3: New gradients
__group_592__new_moment_1 (optional) - T4: New averaged gradients
__group_592__new_moment_2 (optional) - T4: New averaged squared gradients
__group_592__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_593__new_weights (optional) - T2: New weights
__group_593__new_gradients (optional) - T3: New gradients
__group_593__new_moment_1 (optional) - T4: New averaged gradients
__group_593__new_moment_2 (optional) - T4: New averaged squared gradients
__group_593__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_594__new_weights (optional) - T2: New weights
__group_594__new_gradients (optional) - T3: New gradients
__group_594__new_moment_1 (optional) - T4: New averaged gradients
__group_594__new_moment_2 (optional) - T4: New averaged squared gradients
__group_594__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_595__new_weights (optional) - T2: New weights
__group_595__new_gradients (optional) - T3: New gradients
__group_595__new_moment_1 (optional) - T4: New averaged gradients
__group_595__new_moment_2 (optional) - T4: New averaged squared gradients
__group_595__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_596__new_weights (optional) - T2: New weights
__group_596__new_gradients (optional) - T3: New gradients
__group_596__new_moment_1 (optional) - T4: New averaged gradients
__group_596__new_moment_2 (optional) - T4: New averaged squared gradients
__group_596__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_597__new_weights (optional) - T2: New weights
__group_597__new_gradients (optional) - T3: New gradients
__group_597__new_moment_1 (optional) - T4: New averaged gradients
__group_597__new_moment_2 (optional) - T4: New averaged squared gradients
__group_597__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_598__new_weights (optional) - T2: New weights
__group_598__new_gradients (optional) - T3: New gradients
__group_598__new_moment_1 (optional) - T4: New averaged gradients
__group_598__new_moment_2 (optional) - T4: New averaged squared gradients
__group_598__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_599__new_weights (optional) - T2: New weights
__group_599__new_gradients (optional) - T3: New gradients
__group_599__new_moment_1 (optional) - T4: New averaged gradients
__group_599__new_moment_2 (optional) - T4: New averaged squared gradients
__group_599__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_600__new_weights (optional) - T2: New weights
__group_600__new_gradients (optional) - T3: New gradients
__group_600__new_moment_1 (optional) - T4: New averaged gradients
__group_600__new_moment_2 (optional) - T4: New averaged squared gradients
__group_600__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_601__new_weights (optional) - T2: New weights
__group_601__new_gradients (optional) - T3: New gradients
__group_601__new_moment_1 (optional) - T4: New averaged gradients
__group_601__new_moment_2 (optional) - T4: New averaged squared gradients
__group_601__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_602__new_weights (optional) - T2: New weights
__group_602__new_gradients (optional) - T3: New gradients
__group_602__new_moment_1 (optional) - T4: New averaged gradients
__group_602__new_moment_2 (optional) - T4: New averaged squared gradients
__group_602__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_603__new_weights (optional) - T2: New weights
__group_603__new_gradients (optional) - T3: New gradients
__group_603__new_moment_1 (optional) - T4: New averaged gradients
__group_603__new_moment_2 (optional) - T4: New averaged squared gradients
__group_603__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_604__new_weights (optional) - T2: New weights
__group_604__new_gradients (optional) - T3: New gradients
__group_604__new_moment_1 (optional) - T4: New averaged gradients
__group_604__new_moment_2 (optional) - T4: New averaged squared gradients
__group_604__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_605__new_weights (optional) - T2: New weights
__group_605__new_gradients (optional) - T3: New gradients
__group_605__new_moment_1 (optional) - T4: New averaged gradients
__group_605__new_moment_2 (optional) - T4: New averaged squared gradients
__group_605__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_606__new_weights (optional) - T2: New weights
__group_606__new_gradients (optional) - T3: New gradients
__group_606__new_moment_1 (optional) - T4: New averaged gradients
__group_606__new_moment_2 (optional) - T4: New averaged squared gradients
__group_606__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_607__new_weights (optional) - T2: New weights
__group_607__new_gradients (optional) - T3: New gradients
__group_607__new_moment_1 (optional) - T4: New averaged gradients
__group_607__new_moment_2 (optional) - T4: New averaged squared gradients
__group_607__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_608__new_weights (optional) - T2: New weights
__group_608__new_gradients (optional) - T3: New gradients
__group_608__new_moment_1 (optional) - T4: New averaged gradients
__group_608__new_moment_2 (optional) - T4: New averaged squared gradients
__group_608__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_609__new_weights (optional) - T2: New weights
__group_609__new_gradients (optional) - T3: New gradients
__group_609__new_moment_1 (optional) - T4: New averaged gradients
__group_609__new_moment_2 (optional) - T4: New averaged squared gradients
__group_609__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_610__new_weights (optional) - T2: New weights
__group_610__new_gradients (optional) - T3: New gradients
__group_610__new_moment_1 (optional) - T4: New averaged gradients
__group_610__new_moment_2 (optional) - T4: New averaged squared gradients
__group_610__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_611__new_weights (optional) - T2: New weights
__group_611__new_gradients (optional) - T3: New gradients
__group_611__new_moment_1 (optional) - T4: New averaged gradients
__group_611__new_moment_2 (optional) - T4: New averaged squared gradients
__group_611__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_612__new_weights (optional) - T2: New weights
__group_612__new_gradients (optional) - T3: New gradients
__group_612__new_moment_1 (optional) - T4: New averaged gradients
__group_612__new_moment_2 (optional) - T4: New averaged squared gradients
__group_612__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_613__new_weights (optional) - T2: New weights
__group_613__new_gradients (optional) - T3: New gradients
__group_613__new_moment_1 (optional) - T4: New averaged gradients
__group_613__new_moment_2 (optional) - T4: New averaged squared gradients
__group_613__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_614__new_weights (optional) - T2: New weights
__group_614__new_gradients (optional) - T3: New gradients
__group_614__new_moment_1 (optional) - T4: New averaged gradients
__group_614__new_moment_2 (optional) - T4: New averaged squared gradients
__group_614__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_615__new_weights (optional) - T2: New weights
__group_615__new_gradients (optional) - T3: New gradients
__group_615__new_moment_1 (optional) - T4: New averaged gradients
__group_615__new_moment_2 (optional) - T4: New averaged squared gradients
__group_615__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_616__new_weights (optional) - T2: New weights
__group_616__new_gradients (optional) - T3: New gradients
__group_616__new_moment_1 (optional) - T4: New averaged gradients
__group_616__new_moment_2 (optional) - T4: New averaged squared gradients
__group_616__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_617__new_weights (optional) - T2: New weights
__group_617__new_gradients (optional) - T3: New gradients
__group_617__new_moment_1 (optional) - T4: New averaged gradients
__group_617__new_moment_2 (optional) - T4: New averaged squared gradients
__group_617__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_618__new_weights (optional) - T2: New weights
__group_618__new_gradients (optional) - T3: New gradients
__group_618__new_moment_1 (optional) - T4: New averaged gradients
__group_618__new_moment_2 (optional) - T4: New averaged squared gradients
__group_618__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_619__new_weights (optional) - T2: New weights
__group_619__new_gradients (optional) - T3: New gradients
__group_619__new_moment_1 (optional) - T4: New averaged gradients
__group_619__new_moment_2 (optional) - T4: New averaged squared gradients
__group_619__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_620__new_weights (optional) - T2: New weights
__group_620__new_gradients (optional) - T3: New gradients
__group_620__new_moment_1 (optional) - T4: New averaged gradients
__group_620__new_moment_2 (optional) - T4: New averaged squared gradients
__group_620__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_621__new_weights (optional) - T2: New weights
__group_621__new_gradients (optional) - T3: New gradients
__group_621__new_moment_1 (optional) - T4: New averaged gradients
__group_621__new_moment_2 (optional) - T4: New averaged squared gradients
__group_621__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_622__new_weights (optional) - T2: New weights
__group_622__new_gradients (optional) - T3: New gradients
__group_622__new_moment_1 (optional) - T4: New averaged gradients
__group_622__new_moment_2 (optional) - T4: New averaged squared gradients
__group_622__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_623__new_weights (optional) - T2: New weights
__group_623__new_gradients (optional) - T3: New gradients
__group_623__new_moment_1 (optional) - T4: New averaged gradients
__group_623__new_moment_2 (optional) - T4: New averaged squared gradients
__group_623__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_624__new_weights (optional) - T2: New weights
__group_624__new_gradients (optional) - T3: New gradients
__group_624__new_moment_1 (optional) - T4: New averaged gradients
__group_624__new_moment_2 (optional) - T4: New averaged squared gradients
__group_624__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_625__new_weights (optional) - T2: New weights
__group_625__new_gradients (optional) - T3: New gradients
__group_625__new_moment_1 (optional) - T4: New averaged gradients
__group_625__new_moment_2 (optional) - T4: New averaged squared gradients
__group_625__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_626__new_weights (optional) - T2: New weights
__group_626__new_gradients (optional) - T3: New gradients
__group_626__new_moment_1 (optional) - T4: New averaged gradients
__group_626__new_moment_2 (optional) - T4: New averaged squared gradients
__group_626__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_627__new_weights (optional) - T2: New weights
__group_627__new_gradients (optional) - T3: New gradients
__group_627__new_moment_1 (optional) - T4: New averaged gradients
__group_627__new_moment_2 (optional) - T4: New averaged squared gradients
__group_627__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_628__new_weights (optional) - T2: New weights
__group_628__new_gradients (optional) - T3: New gradients
__group_628__new_moment_1 (optional) - T4: New averaged gradients
__group_628__new_moment_2 (optional) - T4: New averaged squared gradients
__group_628__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_629__new_weights (optional) - T2: New weights
__group_629__new_gradients (optional) - T3: New gradients
__group_629__new_moment_1 (optional) - T4: New averaged gradients
__group_629__new_moment_2 (optional) - T4: New averaged squared gradients
__group_629__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_630__new_weights (optional) - T2: New weights
__group_630__new_gradients (optional) - T3: New gradients
__group_630__new_moment_1 (optional) - T4: New averaged gradients
__group_630__new_moment_2 (optional) - T4: New averaged squared gradients
__group_630__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_631__new_weights (optional) - T2: New weights
__group_631__new_gradients (optional) - T3: New gradients
__group_631__new_moment_1 (optional) - T4: New averaged gradients
__group_631__new_moment_2 (optional) - T4: New averaged squared gradients
__group_631__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_632__new_weights (optional) - T2: New weights
__group_632__new_gradients (optional) - T3: New gradients
__group_632__new_moment_1 (optional) - T4: New averaged gradients
__group_632__new_moment_2 (optional) - T4: New averaged squared gradients
__group_632__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_633__new_weights (optional) - T2: New weights
__group_633__new_gradients (optional) - T3: New gradients
__group_633__new_moment_1 (optional) - T4: New averaged gradients
__group_633__new_moment_2 (optional) - T4: New averaged squared gradients
__group_633__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_634__new_weights (optional) - T2: New weights
__group_634__new_gradients (optional) - T3: New gradients
__group_634__new_moment_1 (optional) - T4: New averaged gradients
__group_634__new_moment_2 (optional) - T4: New averaged squared gradients
__group_634__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_635__new_weights (optional) - T2: New weights
__group_635__new_gradients (optional) - T3: New gradients
__group_635__new_moment_1 (optional) - T4: New averaged gradients
__group_635__new_moment_2 (optional) - T4: New averaged squared gradients
__group_635__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_636__new_weights (optional) - T2: New weights
__group_636__new_gradients (optional) - T3: New gradients
__group_636__new_moment_1 (optional) - T4: New averaged gradients
__group_636__new_moment_2 (optional) - T4: New averaged squared gradients
__group_636__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_637__new_weights (optional) - T2: New weights
__group_637__new_gradients (optional) - T3: New gradients
__group_637__new_moment_1 (optional) - T4: New averaged gradients
__group_637__new_moment_2 (optional) - T4: New averaged squared gradients
__group_637__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_638__new_weights (optional) - T2: New weights
__group_638__new_gradients (optional) - T3: New gradients
__group_638__new_moment_1 (optional) - T4: New averaged gradients
__group_638__new_moment_2 (optional) - T4: New averaged squared gradients
__group_638__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_639__new_weights (optional) - T2: New weights
__group_639__new_gradients (optional) - T3: New gradients
__group_639__new_moment_1 (optional) - T4: New averaged gradients
__group_639__new_moment_2 (optional) - T4: New averaged squared gradients
__group_639__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_640__new_weights (optional) - T2: New weights
__group_640__new_gradients (optional) - T3: New gradients
__group_640__new_moment_1 (optional) - T4: New averaged gradients
__group_640__new_moment_2 (optional) - T4: New averaged squared gradients
__group_640__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_641__new_weights (optional) - T2: New weights
__group_641__new_gradients (optional) - T3: New gradients
__group_641__new_moment_1 (optional) - T4: New averaged gradients
__group_641__new_moment_2 (optional) - T4: New averaged squared gradients
__group_641__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_642__new_weights (optional) - T2: New weights
__group_642__new_gradients (optional) - T3: New gradients
__group_642__new_moment_1 (optional) - T4: New averaged gradients
__group_642__new_moment_2 (optional) - T4: New averaged squared gradients
__group_642__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_643__new_weights (optional) - T2: New weights
__group_643__new_gradients (optional) - T3: New gradients
__group_643__new_moment_1 (optional) - T4: New averaged gradients
__group_643__new_moment_2 (optional) - T4: New averaged squared gradients
__group_643__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_644__new_weights (optional) - T2: New weights
__group_644__new_gradients (optional) - T3: New gradients
__group_644__new_moment_1 (optional) - T4: New averaged gradients
__group_644__new_moment_2 (optional) - T4: New averaged squared gradients
__group_644__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_645__new_weights (optional) - T2: New weights
__group_645__new_gradients (optional) - T3: New gradients
__group_645__new_moment_1 (optional) - T4: New averaged gradients
__group_645__new_moment_2 (optional) - T4: New averaged squared gradients
__group_645__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_646__new_weights (optional) - T2: New weights
__group_646__new_gradients (optional) - T3: New gradients
__group_646__new_moment_1 (optional) - T4: New averaged gradients
__group_646__new_moment_2 (optional) - T4: New averaged squared gradients
__group_646__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_647__new_weights (optional) - T2: New weights
__group_647__new_gradients (optional) - T3: New gradients
__group_647__new_moment_1 (optional) - T4: New averaged gradients
__group_647__new_moment_2 (optional) - T4: New averaged squared gradients
__group_647__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_648__new_weights (optional) - T2: New weights
__group_648__new_gradients (optional) - T3: New gradients
__group_648__new_moment_1 (optional) - T4: New averaged gradients
__group_648__new_moment_2 (optional) - T4: New averaged squared gradients
__group_648__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_649__new_weights (optional) - T2: New weights
__group_649__new_gradients (optional) - T3: New gradients
__group_649__new_moment_1 (optional) - T4: New averaged gradients
__group_649__new_moment_2 (optional) - T4: New averaged squared gradients
__group_649__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_650__new_weights (optional) - T2: New weights
__group_650__new_gradients (optional) - T3: New gradients
__group_650__new_moment_1 (optional) - T4: New averaged gradients
__group_650__new_moment_2 (optional) - T4: New averaged squared gradients
__group_650__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_651__new_weights (optional) - T2: New weights
__group_651__new_gradients (optional) - T3: New gradients
__group_651__new_moment_1 (optional) - T4: New averaged gradients
__group_651__new_moment_2 (optional) - T4: New averaged squared gradients
__group_651__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_652__new_weights (optional) - T2: New weights
__group_652__new_gradients (optional) - T3: New gradients
__group_652__new_moment_1 (optional) - T4: New averaged gradients
__group_652__new_moment_2 (optional) - T4: New averaged squared gradients
__group_652__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_653__new_weights (optional) - T2: New weights
__group_653__new_gradients (optional) - T3: New gradients
__group_653__new_moment_1 (optional) - T4: New averaged gradients
__group_653__new_moment_2 (optional) - T4: New averaged squared gradients
__group_653__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_654__new_weights (optional) - T2: New weights
__group_654__new_gradients (optional) - T3: New gradients
__group_654__new_moment_1 (optional) - T4: New averaged gradients
__group_654__new_moment_2 (optional) - T4: New averaged squared gradients
__group_654__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_655__new_weights (optional) - T2: New weights
__group_655__new_gradients (optional) - T3: New gradients
__group_655__new_moment_1 (optional) - T4: New averaged gradients
__group_655__new_moment_2 (optional) - T4: New averaged squared gradients
__group_655__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_656__new_weights (optional) - T2: New weights
__group_656__new_gradients (optional) - T3: New gradients
__group_656__new_moment_1 (optional) - T4: New averaged gradients
__group_656__new_moment_2 (optional) - T4: New averaged squared gradients
__group_656__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_657__new_weights (optional) - T2: New weights
__group_657__new_gradients (optional) - T3: New gradients
__group_657__new_moment_1 (optional) - T4: New averaged gradients
__group_657__new_moment_2 (optional) - T4: New averaged squared gradients
__group_657__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_658__new_weights (optional) - T2: New weights
__group_658__new_gradients (optional) - T3: New gradients
__group_658__new_moment_1 (optional) - T4: New averaged gradients
__group_658__new_moment_2 (optional) - T4: New averaged squared gradients
__group_658__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_659__new_weights (optional) - T2: New weights
__group_659__new_gradients (optional) - T3: New gradients
__group_659__new_moment_1 (optional) - T4: New averaged gradients
__group_659__new_moment_2 (optional) - T4: New averaged squared gradients
__group_659__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_660__new_weights (optional) - T2: New weights
__group_660__new_gradients (optional) - T3: New gradients
__group_660__new_moment_1 (optional) - T4: New averaged gradients
__group_660__new_moment_2 (optional) - T4: New averaged squared gradients
__group_660__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_661__new_weights (optional) - T2: New weights
__group_661__new_gradients (optional) - T3: New gradients
__group_661__new_moment_1 (optional) - T4: New averaged gradients
__group_661__new_moment_2 (optional) - T4: New averaged squared gradients
__group_661__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_662__new_weights (optional) - T2: New weights
__group_662__new_gradients (optional) - T3: New gradients
__group_662__new_moment_1 (optional) - T4: New averaged gradients
__group_662__new_moment_2 (optional) - T4: New averaged squared gradients
__group_662__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_663__new_weights (optional) - T2: New weights
__group_663__new_gradients (optional) - T3: New gradients
__group_663__new_moment_1 (optional) - T4: New averaged gradients
__group_663__new_moment_2 (optional) - T4: New averaged squared gradients
__group_663__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_664__new_weights (optional) - T2: New weights
__group_664__new_gradients (optional) - T3: New gradients
__group_664__new_moment_1 (optional) - T4: New averaged gradients
__group_664__new_moment_2 (optional) - T4: New averaged squared gradients
__group_664__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_665__new_weights (optional) - T2: New weights
__group_665__new_gradients (optional) - T3: New gradients
__group_665__new_moment_1 (optional) - T4: New averaged gradients
__group_665__new_moment_2 (optional) - T4: New averaged squared gradients
__group_665__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_666__new_weights (optional) - T2: New weights
__group_666__new_gradients (optional) - T3: New gradients
__group_666__new_moment_1 (optional) - T4: New averaged gradients
__group_666__new_moment_2 (optional) - T4: New averaged squared gradients
__group_666__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_667__new_weights (optional) - T2: New weights
__group_667__new_gradients (optional) - T3: New gradients
__group_667__new_moment_1 (optional) - T4: New averaged gradients
__group_667__new_moment_2 (optional) - T4: New averaged squared gradients
__group_667__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_668__new_weights (optional) - T2: New weights
__group_668__new_gradients (optional) - T3: New gradients
__group_668__new_moment_1 (optional) - T4: New averaged gradients
__group_668__new_moment_2 (optional) - T4: New averaged squared gradients
__group_668__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_669__new_weights (optional) - T2: New weights
__group_669__new_gradients (optional) - T3: New gradients
__group_669__new_moment_1 (optional) - T4: New averaged gradients
__group_669__new_moment_2 (optional) - T4: New averaged squared gradients
__group_669__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_670__new_weights (optional) - T2: New weights
__group_670__new_gradients (optional) - T3: New gradients
__group_670__new_moment_1 (optional) - T4: New averaged gradients
__group_670__new_moment_2 (optional) - T4: New averaged squared gradients
__group_670__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_671__new_weights (optional) - T2: New weights
__group_671__new_gradients (optional) - T3: New gradients
__group_671__new_moment_1 (optional) - T4: New averaged gradients
__group_671__new_moment_2 (optional) - T4: New averaged squared gradients
__group_671__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_672__new_weights (optional) - T2: New weights
__group_672__new_gradients (optional) - T3: New gradients
__group_672__new_moment_1 (optional) - T4: New averaged gradients
__group_672__new_moment_2 (optional) - T4: New averaged squared gradients
__group_672__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_673__new_weights (optional) - T2: New weights
__group_673__new_gradients (optional) - T3: New gradients
__group_673__new_moment_1 (optional) - T4: New averaged gradients
__group_673__new_moment_2 (optional) - T4: New averaged squared gradients
__group_673__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_674__new_weights (optional) - T2: New weights
__group_674__new_gradients (optional) - T3: New gradients
__group_674__new_moment_1 (optional) - T4: New averaged gradients
__group_674__new_moment_2 (optional) - T4: New averaged squared gradients
__group_674__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_675__new_weights (optional) - T2: New weights
__group_675__new_gradients (optional) - T3: New gradients
__group_675__new_moment_1 (optional) - T4: New averaged gradients
__group_675__new_moment_2 (optional) - T4: New averaged squared gradients
__group_675__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_676__new_weights (optional) - T2: New weights
__group_676__new_gradients (optional) - T3: New gradients
__group_676__new_moment_1 (optional) - T4: New averaged gradients
__group_676__new_moment_2 (optional) - T4: New averaged squared gradients
__group_676__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_677__new_weights (optional) - T2: New weights
__group_677__new_gradients (optional) - T3: New gradients
__group_677__new_moment_1 (optional) - T4: New averaged gradients
__group_677__new_moment_2 (optional) - T4: New averaged squared gradients
__group_677__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_678__new_weights (optional) - T2: New weights
__group_678__new_gradients (optional) - T3: New gradients
__group_678__new_moment_1 (optional) - T4: New averaged gradients
__group_678__new_moment_2 (optional) - T4: New averaged squared gradients
__group_678__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_679__new_weights (optional) - T2: New weights
__group_679__new_gradients (optional) - T3: New gradients
__group_679__new_moment_1 (optional) - T4: New averaged gradients
__group_679__new_moment_2 (optional) - T4: New averaged squared gradients
__group_679__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_680__new_weights (optional) - T2: New weights
__group_680__new_gradients (optional) - T3: New gradients
__group_680__new_moment_1 (optional) - T4: New averaged gradients
__group_680__new_moment_2 (optional) - T4: New averaged squared gradients
__group_680__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_681__new_weights (optional) - T2: New weights
__group_681__new_gradients (optional) - T3: New gradients
__group_681__new_moment_1 (optional) - T4: New averaged gradients
__group_681__new_moment_2 (optional) - T4: New averaged squared gradients
__group_681__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_682__new_weights (optional) - T2: New weights
__group_682__new_gradients (optional) - T3: New gradients
__group_682__new_moment_1 (optional) - T4: New averaged gradients
__group_682__new_moment_2 (optional) - T4: New averaged squared gradients
__group_682__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_683__new_weights (optional) - T2: New weights
__group_683__new_gradients (optional) - T3: New gradients
__group_683__new_moment_1 (optional) - T4: New averaged gradients
__group_683__new_moment_2 (optional) - T4: New averaged squared gradients
__group_683__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_684__new_weights (optional) - T2: New weights
__group_684__new_gradients (optional) - T3: New gradients
__group_684__new_moment_1 (optional) - T4: New averaged gradients
__group_684__new_moment_2 (optional) - T4: New averaged squared gradients
__group_684__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_685__new_weights (optional) - T2: New weights
__group_685__new_gradients (optional) - T3: New gradients
__group_685__new_moment_1 (optional) - T4: New averaged gradients
__group_685__new_moment_2 (optional) - T4: New averaged squared gradients
__group_685__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_686__new_weights (optional) - T2: New weights
__group_686__new_gradients (optional) - T3: New gradients
__group_686__new_moment_1 (optional) - T4: New averaged gradients
__group_686__new_moment_2 (optional) - T4: New averaged squared gradients
__group_686__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_687__new_weights (optional) - T2: New weights
__group_687__new_gradients (optional) - T3: New gradients
__group_687__new_moment_1 (optional) - T4: New averaged gradients
__group_687__new_moment_2 (optional) - T4: New averaged squared gradients
__group_687__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_688__new_weights (optional) - T2: New weights
__group_688__new_gradients (optional) - T3: New gradients
__group_688__new_moment_1 (optional) - T4: New averaged gradients
__group_688__new_moment_2 (optional) - T4: New averaged squared gradients
__group_688__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_689__new_weights (optional) - T2: New weights
__group_689__new_gradients (optional) - T3: New gradients
__group_689__new_moment_1 (optional) - T4: New averaged gradients
__group_689__new_moment_2 (optional) - T4: New averaged squared gradients
__group_689__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_690__new_weights (optional) - T2: New weights
__group_690__new_gradients (optional) - T3: New gradients
__group_690__new_moment_1 (optional) - T4: New averaged gradients
__group_690__new_moment_2 (optional) - T4: New averaged squared gradients
__group_690__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_691__new_weights (optional) - T2: New weights
__group_691__new_gradients (optional) - T3: New gradients
__group_691__new_moment_1 (optional) - T4: New averaged gradients
__group_691__new_moment_2 (optional) - T4: New averaged squared gradients
__group_691__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_692__new_weights (optional) - T2: New weights
__group_692__new_gradients (optional) - T3: New gradients
__group_692__new_moment_1 (optional) - T4: New averaged gradients
__group_692__new_moment_2 (optional) - T4: New averaged squared gradients
__group_692__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_693__new_weights (optional) - T2: New weights
__group_693__new_gradients (optional) - T3: New gradients
__group_693__new_moment_1 (optional) - T4: New averaged gradients
__group_693__new_moment_2 (optional) - T4: New averaged squared gradients
__group_693__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_694__new_weights (optional) - T2: New weights
__group_694__new_gradients (optional) - T3: New gradients
__group_694__new_moment_1 (optional) - T4: New averaged gradients
__group_694__new_moment_2 (optional) - T4: New averaged squared gradients
__group_694__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_695__new_weights (optional) - T2: New weights
__group_695__new_gradients (optional) - T3: New gradients
__group_695__new_moment_1 (optional) - T4: New averaged gradients
__group_695__new_moment_2 (optional) - T4: New averaged squared gradients
__group_695__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_696__new_weights (optional) - T2: New weights
__group_696__new_gradients (optional) - T3: New gradients
__group_696__new_moment_1 (optional) - T4: New averaged gradients
__group_696__new_moment_2 (optional) - T4: New averaged squared gradients
__group_696__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_697__new_weights (optional) - T2: New weights
__group_697__new_gradients (optional) - T3: New gradients
__group_697__new_moment_1 (optional) - T4: New averaged gradients
__group_697__new_moment_2 (optional) - T4: New averaged squared gradients
__group_697__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_698__new_weights (optional) - T2: New weights
__group_698__new_gradients (optional) - T3: New gradients
__group_698__new_moment_1 (optional) - T4: New averaged gradients
__group_698__new_moment_2 (optional) - T4: New averaged squared gradients
__group_698__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_699__new_weights (optional) - T2: New weights
__group_699__new_gradients (optional) - T3: New gradients
__group_699__new_moment_1 (optional) - T4: New averaged gradients
__group_699__new_moment_2 (optional) - T4: New averaged squared gradients
__group_699__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_700__new_weights (optional) - T2: New weights
__group_700__new_gradients (optional) - T3: New gradients
__group_700__new_moment_1 (optional) - T4: New averaged gradients
__group_700__new_moment_2 (optional) - T4: New averaged squared gradients
__group_700__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_701__new_weights (optional) - T2: New weights
__group_701__new_gradients (optional) - T3: New gradients
__group_701__new_moment_1 (optional) - T4: New averaged gradients
__group_701__new_moment_2 (optional) - T4: New averaged squared gradients
__group_701__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_702__new_weights (optional) - T2: New weights
__group_702__new_gradients (optional) - T3: New gradients
__group_702__new_moment_1 (optional) - T4: New averaged gradients
__group_702__new_moment_2 (optional) - T4: New averaged squared gradients
__group_702__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_703__new_weights (optional) - T2: New weights
__group_703__new_gradients (optional) - T3: New gradients
__group_703__new_moment_1 (optional) - T4: New averaged gradients
__group_703__new_moment_2 (optional) - T4: New averaged squared gradients
__group_703__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_704__new_weights (optional) - T2: New weights
__group_704__new_gradients (optional) - T3: New gradients
__group_704__new_moment_1 (optional) - T4: New averaged gradients
__group_704__new_moment_2 (optional) - T4: New averaged squared gradients
__group_704__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_705__new_weights (optional) - T2: New weights
__group_705__new_gradients (optional) - T3: New gradients
__group_705__new_moment_1 (optional) - T4: New averaged gradients
__group_705__new_moment_2 (optional) - T4: New averaged squared gradients
__group_705__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_706__new_weights (optional) - T2: New weights
__group_706__new_gradients (optional) - T3: New gradients
__group_706__new_moment_1 (optional) - T4: New averaged gradients
__group_706__new_moment_2 (optional) - T4: New averaged squared gradients
__group_706__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_707__new_weights (optional) - T2: New weights
__group_707__new_gradients (optional) - T3: New gradients
__group_707__new_moment_1 (optional) - T4: New averaged gradients
__group_707__new_moment_2 (optional) - T4: New averaged squared gradients
__group_707__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_708__new_weights (optional) - T2: New weights
__group_708__new_gradients (optional) - T3: New gradients
__group_708__new_moment_1 (optional) - T4: New averaged gradients
__group_708__new_moment_2 (optional) - T4: New averaged squared gradients
__group_708__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_709__new_weights (optional) - T2: New weights
__group_709__new_gradients (optional) - T3: New gradients
__group_709__new_moment_1 (optional) - T4: New averaged gradients
__group_709__new_moment_2 (optional) - T4: New averaged squared gradients
__group_709__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_710__new_weights (optional) - T2: New weights
__group_710__new_gradients (optional) - T3: New gradients
__group_710__new_moment_1 (optional) - T4: New averaged gradients
__group_710__new_moment_2 (optional) - T4: New averaged squared gradients
__group_710__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_711__new_weights (optional) - T2: New weights
__group_711__new_gradients (optional) - T3: New gradients
__group_711__new_moment_1 (optional) - T4: New averaged gradients
__group_711__new_moment_2 (optional) - T4: New averaged squared gradients
__group_711__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_712__new_weights (optional) - T2: New weights
__group_712__new_gradients (optional) - T3: New gradients
__group_712__new_moment_1 (optional) - T4: New averaged gradients
__group_712__new_moment_2 (optional) - T4: New averaged squared gradients
__group_712__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_713__new_weights (optional) - T2: New weights
__group_713__new_gradients (optional) - T3: New gradients
__group_713__new_moment_1 (optional) - T4: New averaged gradients
__group_713__new_moment_2 (optional) - T4: New averaged squared gradients
__group_713__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_714__new_weights (optional) - T2: New weights
__group_714__new_gradients (optional) - T3: New gradients
__group_714__new_moment_1 (optional) - T4: New averaged gradients
__group_714__new_moment_2 (optional) - T4: New averaged squared gradients
__group_714__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_715__new_weights (optional) - T2: New weights
__group_715__new_gradients (optional) - T3: New gradients
__group_715__new_moment_1 (optional) - T4: New averaged gradients
__group_715__new_moment_2 (optional) - T4: New averaged squared gradients
__group_715__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_716__new_weights (optional) - T2: New weights
__group_716__new_gradients (optional) - T3: New gradients
__group_716__new_moment_1 (optional) - T4: New averaged gradients
__group_716__new_moment_2 (optional) - T4: New averaged squared gradients
__group_716__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_717__new_weights (optional) - T2: New weights
__group_717__new_gradients (optional) - T3: New gradients
__group_717__new_moment_1 (optional) - T4: New averaged gradients
__group_717__new_moment_2 (optional) - T4: New averaged squared gradients
__group_717__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_718__new_weights (optional) - T2: New weights
__group_718__new_gradients (optional) - T3: New gradients
__group_718__new_moment_1 (optional) - T4: New averaged gradients
__group_718__new_moment_2 (optional) - T4: New averaged squared gradients
__group_718__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_719__new_weights (optional) - T2: New weights
__group_719__new_gradients (optional) - T3: New gradients
__group_719__new_moment_1 (optional) - T4: New averaged gradients
__group_719__new_moment_2 (optional) - T4: New averaged squared gradients
__group_719__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_720__new_weights (optional) - T2: New weights
__group_720__new_gradients (optional) - T3: New gradients
__group_720__new_moment_1 (optional) - T4: New averaged gradients
__group_720__new_moment_2 (optional) - T4: New averaged squared gradients
__group_720__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_721__new_weights (optional) - T2: New weights
__group_721__new_gradients (optional) - T3: New gradients
__group_721__new_moment_1 (optional) - T4: New averaged gradients
__group_721__new_moment_2 (optional) - T4: New averaged squared gradients
__group_721__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_722__new_weights (optional) - T2: New weights
__group_722__new_gradients (optional) - T3: New gradients
__group_722__new_moment_1 (optional) - T4: New averaged gradients
__group_722__new_moment_2 (optional) - T4: New averaged squared gradients
__group_722__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_723__new_weights (optional) - T2: New weights
__group_723__new_gradients (optional) - T3: New gradients
__group_723__new_moment_1 (optional) - T4: New averaged gradients
__group_723__new_moment_2 (optional) - T4: New averaged squared gradients
__group_723__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_724__new_weights (optional) - T2: New weights
__group_724__new_gradients (optional) - T3: New gradients
__group_724__new_moment_1 (optional) - T4: New averaged gradients
__group_724__new_moment_2 (optional) - T4: New averaged squared gradients
__group_724__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_725__new_weights (optional) - T2: New weights
__group_725__new_gradients (optional) - T3: New gradients
__group_725__new_moment_1 (optional) - T4: New averaged gradients
__group_725__new_moment_2 (optional) - T4: New averaged squared gradients
__group_725__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_726__new_weights (optional) - T2: New weights
__group_726__new_gradients (optional) - T3: New gradients
__group_726__new_moment_1 (optional) - T4: New averaged gradients
__group_726__new_moment_2 (optional) - T4: New averaged squared gradients
__group_726__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_727__new_weights (optional) - T2: New weights
__group_727__new_gradients (optional) - T3: New gradients
__group_727__new_moment_1 (optional) - T4: New averaged gradients
__group_727__new_moment_2 (optional) - T4: New averaged squared gradients
__group_727__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_728__new_weights (optional) - T2: New weights
__group_728__new_gradients (optional) - T3: New gradients
__group_728__new_moment_1 (optional) - T4: New averaged gradients
__group_728__new_moment_2 (optional) - T4: New averaged squared gradients
__group_728__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_729__new_weights (optional) - T2: New weights
__group_729__new_gradients (optional) - T3: New gradients
__group_729__new_moment_1 (optional) - T4: New averaged gradients
__group_729__new_moment_2 (optional) - T4: New averaged squared gradients
__group_729__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_730__new_weights (optional) - T2: New weights
__group_730__new_gradients (optional) - T3: New gradients
__group_730__new_moment_1 (optional) - T4: New averaged gradients
__group_730__new_moment_2 (optional) - T4: New averaged squared gradients
__group_730__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_731__new_weights (optional) - T2: New weights
__group_731__new_gradients (optional) - T3: New gradients
__group_731__new_moment_1 (optional) - T4: New averaged gradients
__group_731__new_moment_2 (optional) - T4: New averaged squared gradients
__group_731__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_732__new_weights (optional) - T2: New weights
__group_732__new_gradients (optional) - T3: New gradients
__group_732__new_moment_1 (optional) - T4: New averaged gradients
__group_732__new_moment_2 (optional) - T4: New averaged squared gradients
__group_732__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_733__new_weights (optional) - T2: New weights
__group_733__new_gradients (optional) - T3: New gradients
__group_733__new_moment_1 (optional) - T4: New averaged gradients
__group_733__new_moment_2 (optional) - T4: New averaged squared gradients
__group_733__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_734__new_weights (optional) - T2: New weights
__group_734__new_gradients (optional) - T3: New gradients
__group_734__new_moment_1 (optional) - T4: New averaged gradients
__group_734__new_moment_2 (optional) - T4: New averaged squared gradients
__group_734__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_735__new_weights (optional) - T2: New weights
__group_735__new_gradients (optional) - T3: New gradients
__group_735__new_moment_1 (optional) - T4: New averaged gradients
__group_735__new_moment_2 (optional) - T4: New averaged squared gradients
__group_735__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_736__new_weights (optional) - T2: New weights
__group_736__new_gradients (optional) - T3: New gradients
__group_736__new_moment_1 (optional) - T4: New averaged gradients
__group_736__new_moment_2 (optional) - T4: New averaged squared gradients
__group_736__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_737__new_weights (optional) - T2: New weights
__group_737__new_gradients (optional) - T3: New gradients
__group_737__new_moment_1 (optional) - T4: New averaged gradients
__group_737__new_moment_2 (optional) - T4: New averaged squared gradients
__group_737__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_738__new_weights (optional) - T2: New weights
__group_738__new_gradients (optional) - T3: New gradients
__group_738__new_moment_1 (optional) - T4: New averaged gradients
__group_738__new_moment_2 (optional) - T4: New averaged squared gradients
__group_738__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_739__new_weights (optional) - T2: New weights
__group_739__new_gradients (optional) - T3: New gradients
__group_739__new_moment_1 (optional) - T4: New averaged gradients
__group_739__new_moment_2 (optional) - T4: New averaged squared gradients
__group_739__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_740__new_weights (optional) - T2: New weights
__group_740__new_gradients (optional) - T3: New gradients
__group_740__new_moment_1 (optional) - T4: New averaged gradients
__group_740__new_moment_2 (optional) - T4: New averaged squared gradients
__group_740__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_741__new_weights (optional) - T2: New weights
__group_741__new_gradients (optional) - T3: New gradients
__group_741__new_moment_1 (optional) - T4: New averaged gradients
__group_741__new_moment_2 (optional) - T4: New averaged squared gradients
__group_741__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_742__new_weights (optional) - T2: New weights
__group_742__new_gradients (optional) - T3: New gradients
__group_742__new_moment_1 (optional) - T4: New averaged gradients
__group_742__new_moment_2 (optional) - T4: New averaged squared gradients
__group_742__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_743__new_weights (optional) - T2: New weights
__group_743__new_gradients (optional) - T3: New gradients
__group_743__new_moment_1 (optional) - T4: New averaged gradients
__group_743__new_moment_2 (optional) - T4: New averaged squared gradients
__group_743__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_744__new_weights (optional) - T2: New weights
__group_744__new_gradients (optional) - T3: New gradients
__group_744__new_moment_1 (optional) - T4: New averaged gradients
__group_744__new_moment_2 (optional) - T4: New averaged squared gradients
__group_744__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_745__new_weights (optional) - T2: New weights
__group_745__new_gradients (optional) - T3: New gradients
__group_745__new_moment_1 (optional) - T4: New averaged gradients
__group_745__new_moment_2 (optional) - T4: New averaged squared gradients
__group_745__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_746__new_weights (optional) - T2: New weights
__group_746__new_gradients (optional) - T3: New gradients
__group_746__new_moment_1 (optional) - T4: New averaged gradients
__group_746__new_moment_2 (optional) - T4: New averaged squared gradients
__group_746__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_747__new_weights (optional) - T2: New weights
__group_747__new_gradients (optional) - T3: New gradients
__group_747__new_moment_1 (optional) - T4: New averaged gradients
__group_747__new_moment_2 (optional) - T4: New averaged squared gradients
__group_747__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_748__new_weights (optional) - T2: New weights
__group_748__new_gradients (optional) - T3: New gradients
__group_748__new_moment_1 (optional) - T4: New averaged gradients
__group_748__new_moment_2 (optional) - T4: New averaged squared gradients
__group_748__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_749__new_weights (optional) - T2: New weights
__group_749__new_gradients (optional) - T3: New gradients
__group_749__new_moment_1 (optional) - T4: New averaged gradients
__group_749__new_moment_2 (optional) - T4: New averaged squared gradients
__group_749__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_750__new_weights (optional) - T2: New weights
__group_750__new_gradients (optional) - T3: New gradients
__group_750__new_moment_1 (optional) - T4: New averaged gradients
__group_750__new_moment_2 (optional) - T4: New averaged squared gradients
__group_750__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_751__new_weights (optional) - T2: New weights
__group_751__new_gradients (optional) - T3: New gradients
__group_751__new_moment_1 (optional) - T4: New averaged gradients
__group_751__new_moment_2 (optional) - T4: New averaged squared gradients
__group_751__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_752__new_weights (optional) - T2: New weights
__group_752__new_gradients (optional) - T3: New gradients
__group_752__new_moment_1 (optional) - T4: New averaged gradients
__group_752__new_moment_2 (optional) - T4: New averaged squared gradients
__group_752__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_753__new_weights (optional) - T2: New weights
__group_753__new_gradients (optional) - T3: New gradients
__group_753__new_moment_1 (optional) - T4: New averaged gradients
__group_753__new_moment_2 (optional) - T4: New averaged squared gradients
__group_753__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_754__new_weights (optional) - T2: New weights
__group_754__new_gradients (optional) - T3: New gradients
__group_754__new_moment_1 (optional) - T4: New averaged gradients
__group_754__new_moment_2 (optional) - T4: New averaged squared gradients
__group_754__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_755__new_weights (optional) - T2: New weights
__group_755__new_gradients (optional) - T3: New gradients
__group_755__new_moment_1 (optional) - T4: New averaged gradients
__group_755__new_moment_2 (optional) - T4: New averaged squared gradients
__group_755__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_756__new_weights (optional) - T2: New weights
__group_756__new_gradients (optional) - T3: New gradients
__group_756__new_moment_1 (optional) - T4: New averaged gradients
__group_756__new_moment_2 (optional) - T4: New averaged squared gradients
__group_756__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_757__new_weights (optional) - T2: New weights
__group_757__new_gradients (optional) - T3: New gradients
__group_757__new_moment_1 (optional) - T4: New averaged gradients
__group_757__new_moment_2 (optional) - T4: New averaged squared gradients
__group_757__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_758__new_weights (optional) - T2: New weights
__group_758__new_gradients (optional) - T3: New gradients
__group_758__new_moment_1 (optional) - T4: New averaged gradients
__group_758__new_moment_2 (optional) - T4: New averaged squared gradients
__group_758__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_759__new_weights (optional) - T2: New weights
__group_759__new_gradients (optional) - T3: New gradients
__group_759__new_moment_1 (optional) - T4: New averaged gradients
__group_759__new_moment_2 (optional) - T4: New averaged squared gradients
__group_759__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_760__new_weights (optional) - T2: New weights
__group_760__new_gradients (optional) - T3: New gradients
__group_760__new_moment_1 (optional) - T4: New averaged gradients
__group_760__new_moment_2 (optional) - T4: New averaged squared gradients
__group_760__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_761__new_weights (optional) - T2: New weights
__group_761__new_gradients (optional) - T3: New gradients
__group_761__new_moment_1 (optional) - T4: New averaged gradients
__group_761__new_moment_2 (optional) - T4: New averaged squared gradients
__group_761__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_762__new_weights (optional) - T2: New weights
__group_762__new_gradients (optional) - T3: New gradients
__group_762__new_moment_1 (optional) - T4: New averaged gradients
__group_762__new_moment_2 (optional) - T4: New averaged squared gradients
__group_762__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_763__new_weights (optional) - T2: New weights
__group_763__new_gradients (optional) - T3: New gradients
__group_763__new_moment_1 (optional) - T4: New averaged gradients
__group_763__new_moment_2 (optional) - T4: New averaged squared gradients
__group_763__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_764__new_weights (optional) - T2: New weights
__group_764__new_gradients (optional) - T3: New gradients
__group_764__new_moment_1 (optional) - T4: New averaged gradients
__group_764__new_moment_2 (optional) - T4: New averaged squared gradients
__group_764__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_765__new_weights (optional) - T2: New weights
__group_765__new_gradients (optional) - T3: New gradients
__group_765__new_moment_1 (optional) - T4: New averaged gradients
__group_765__new_moment_2 (optional) - T4: New averaged squared gradients
__group_765__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_766__new_weights (optional) - T2: New weights
__group_766__new_gradients (optional) - T3: New gradients
__group_766__new_moment_1 (optional) - T4: New averaged gradients
__group_766__new_moment_2 (optional) - T4: New averaged squared gradients
__group_766__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_767__new_weights (optional) - T2: New weights
__group_767__new_gradients (optional) - T3: New gradients
__group_767__new_moment_1 (optional) - T4: New averaged gradients
__group_767__new_moment_2 (optional) - T4: New averaged squared gradients
__group_767__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_768__new_weights (optional) - T2: New weights
__group_768__new_gradients (optional) - T3: New gradients
__group_768__new_moment_1 (optional) - T4: New averaged gradients
__group_768__new_moment_2 (optional) - T4: New averaged squared gradients
__group_768__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_769__new_weights (optional) - T2: New weights
__group_769__new_gradients (optional) - T3: New gradients
__group_769__new_moment_1 (optional) - T4: New averaged gradients
__group_769__new_moment_2 (optional) - T4: New averaged squared gradients
__group_769__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_770__new_weights (optional) - T2: New weights
__group_770__new_gradients (optional) - T3: New gradients
__group_770__new_moment_1 (optional) - T4: New averaged gradients
__group_770__new_moment_2 (optional) - T4: New averaged squared gradients
__group_770__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_771__new_weights (optional) - T2: New weights
__group_771__new_gradients (optional) - T3: New gradients
__group_771__new_moment_1 (optional) - T4: New averaged gradients
__group_771__new_moment_2 (optional) - T4: New averaged squared gradients
__group_771__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_772__new_weights (optional) - T2: New weights
__group_772__new_gradients (optional) - T3: New gradients
__group_772__new_moment_1 (optional) - T4: New averaged gradients
__group_772__new_moment_2 (optional) - T4: New averaged squared gradients
__group_772__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_773__new_weights (optional) - T2: New weights
__group_773__new_gradients (optional) - T3: New gradients
__group_773__new_moment_1 (optional) - T4: New averaged gradients
__group_773__new_moment_2 (optional) - T4: New averaged squared gradients
__group_773__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_774__new_weights (optional) - T2: New weights
__group_774__new_gradients (optional) - T3: New gradients
__group_774__new_moment_1 (optional) - T4: New averaged gradients
__group_774__new_moment_2 (optional) - T4: New averaged squared gradients
__group_774__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_775__new_weights (optional) - T2: New weights
__group_775__new_gradients (optional) - T3: New gradients
__group_775__new_moment_1 (optional) - T4: New averaged gradients
__group_775__new_moment_2 (optional) - T4: New averaged squared gradients
__group_775__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_776__new_weights (optional) - T2: New weights
__group_776__new_gradients (optional) - T3: New gradients
__group_776__new_moment_1 (optional) - T4: New averaged gradients
__group_776__new_moment_2 (optional) - T4: New averaged squared gradients
__group_776__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_777__new_weights (optional) - T2: New weights
__group_777__new_gradients (optional) - T3: New gradients
__group_777__new_moment_1 (optional) - T4: New averaged gradients
__group_777__new_moment_2 (optional) - T4: New averaged squared gradients
__group_777__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_778__new_weights (optional) - T2: New weights
__group_778__new_gradients (optional) - T3: New gradients
__group_778__new_moment_1 (optional) - T4: New averaged gradients
__group_778__new_moment_2 (optional) - T4: New averaged squared gradients
__group_778__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_779__new_weights (optional) - T2: New weights
__group_779__new_gradients (optional) - T3: New gradients
__group_779__new_moment_1 (optional) - T4: New averaged gradients
__group_779__new_moment_2 (optional) - T4: New averaged squared gradients
__group_779__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_780__new_weights (optional) - T2: New weights
__group_780__new_gradients (optional) - T3: New gradients
__group_780__new_moment_1 (optional) - T4: New averaged gradients
__group_780__new_moment_2 (optional) - T4: New averaged squared gradients
__group_780__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_781__new_weights (optional) - T2: New weights
__group_781__new_gradients (optional) - T3: New gradients
__group_781__new_moment_1 (optional) - T4: New averaged gradients
__group_781__new_moment_2 (optional) - T4: New averaged squared gradients
__group_781__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_782__new_weights (optional) - T2: New weights
__group_782__new_gradients (optional) - T3: New gradients
__group_782__new_moment_1 (optional) - T4: New averaged gradients
__group_782__new_moment_2 (optional) - T4: New averaged squared gradients
__group_782__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_783__new_weights (optional) - T2: New weights
__group_783__new_gradients (optional) - T3: New gradients
__group_783__new_moment_1 (optional) - T4: New averaged gradients
__group_783__new_moment_2 (optional) - T4: New averaged squared gradients
__group_783__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_784__new_weights (optional) - T2: New weights
__group_784__new_gradients (optional) - T3: New gradients
__group_784__new_moment_1 (optional) - T4: New averaged gradients
__group_784__new_moment_2 (optional) - T4: New averaged squared gradients
__group_784__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_785__new_weights (optional) - T2: New weights
__group_785__new_gradients (optional) - T3: New gradients
__group_785__new_moment_1 (optional) - T4: New averaged gradients
__group_785__new_moment_2 (optional) - T4: New averaged squared gradients
__group_785__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_786__new_weights (optional) - T2: New weights
__group_786__new_gradients (optional) - T3: New gradients
__group_786__new_moment_1 (optional) - T4: New averaged gradients
__group_786__new_moment_2 (optional) - T4: New averaged squared gradients
__group_786__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_787__new_weights (optional) - T2: New weights
__group_787__new_gradients (optional) - T3: New gradients
__group_787__new_moment_1 (optional) - T4: New averaged gradients
__group_787__new_moment_2 (optional) - T4: New averaged squared gradients
__group_787__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_788__new_weights (optional) - T2: New weights
__group_788__new_gradients (optional) - T3: New gradients
__group_788__new_moment_1 (optional) - T4: New averaged gradients
__group_788__new_moment_2 (optional) - T4: New averaged squared gradients
__group_788__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_789__new_weights (optional) - T2: New weights
__group_789__new_gradients (optional) - T3: New gradients
__group_789__new_moment_1 (optional) - T4: New averaged gradients
__group_789__new_moment_2 (optional) - T4: New averaged squared gradients
__group_789__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_790__new_weights (optional) - T2: New weights
__group_790__new_gradients (optional) - T3: New gradients
__group_790__new_moment_1 (optional) - T4: New averaged gradients
__group_790__new_moment_2 (optional) - T4: New averaged squared gradients
__group_790__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_791__new_weights (optional) - T2: New weights
__group_791__new_gradients (optional) - T3: New gradients
__group_791__new_moment_1 (optional) - T4: New averaged gradients
__group_791__new_moment_2 (optional) - T4: New averaged squared gradients
__group_791__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_792__new_weights (optional) - T2: New weights
__group_792__new_gradients (optional) - T3: New gradients
__group_792__new_moment_1 (optional) - T4: New averaged gradients
__group_792__new_moment_2 (optional) - T4: New averaged squared gradients
__group_792__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_793__new_weights (optional) - T2: New weights
__group_793__new_gradients (optional) - T3: New gradients
__group_793__new_moment_1 (optional) - T4: New averaged gradients
__group_793__new_moment_2 (optional) - T4: New averaged squared gradients
__group_793__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_794__new_weights (optional) - T2: New weights
__group_794__new_gradients (optional) - T3: New gradients
__group_794__new_moment_1 (optional) - T4: New averaged gradients
__group_794__new_moment_2 (optional) - T4: New averaged squared gradients
__group_794__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_795__new_weights (optional) - T2: New weights
__group_795__new_gradients (optional) - T3: New gradients
__group_795__new_moment_1 (optional) - T4: New averaged gradients
__group_795__new_moment_2 (optional) - T4: New averaged squared gradients
__group_795__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_796__new_weights (optional) - T2: New weights
__group_796__new_gradients (optional) - T3: New gradients
__group_796__new_moment_1 (optional) - T4: New averaged gradients
__group_796__new_moment_2 (optional) - T4: New averaged squared gradients
__group_796__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_797__new_weights (optional) - T2: New weights
__group_797__new_gradients (optional) - T3: New gradients
__group_797__new_moment_1 (optional) - T4: New averaged gradients
__group_797__new_moment_2 (optional) - T4: New averaged squared gradients
__group_797__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_798__new_weights (optional) - T2: New weights
__group_798__new_gradients (optional) - T3: New gradients
__group_798__new_moment_1 (optional) - T4: New averaged gradients
__group_798__new_moment_2 (optional) - T4: New averaged squared gradients
__group_798__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_799__new_weights (optional) - T2: New weights
__group_799__new_gradients (optional) - T3: New gradients
__group_799__new_moment_1 (optional) - T4: New averaged gradients
__group_799__new_moment_2 (optional) - T4: New averaged squared gradients
__group_799__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_800__new_weights (optional) - T2: New weights
__group_800__new_gradients (optional) - T3: New gradients
__group_800__new_moment_1 (optional) - T4: New averaged gradients
__group_800__new_moment_2 (optional) - T4: New averaged squared gradients
__group_800__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_801__new_weights (optional) - T2: New weights
__group_801__new_gradients (optional) - T3: New gradients
__group_801__new_moment_1 (optional) - T4: New averaged gradients
__group_801__new_moment_2 (optional) - T4: New averaged squared gradients
__group_801__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_802__new_weights (optional) - T2: New weights
__group_802__new_gradients (optional) - T3: New gradients
__group_802__new_moment_1 (optional) - T4: New averaged gradients
__group_802__new_moment_2 (optional) - T4: New averaged squared gradients
__group_802__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_803__new_weights (optional) - T2: New weights
__group_803__new_gradients (optional) - T3: New gradients
__group_803__new_moment_1 (optional) - T4: New averaged gradients
__group_803__new_moment_2 (optional) - T4: New averaged squared gradients
__group_803__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_804__new_weights (optional) - T2: New weights
__group_804__new_gradients (optional) - T3: New gradients
__group_804__new_moment_1 (optional) - T4: New averaged gradients
__group_804__new_moment_2 (optional) - T4: New averaged squared gradients
__group_804__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_805__new_weights (optional) - T2: New weights
__group_805__new_gradients (optional) - T3: New gradients
__group_805__new_moment_1 (optional) - T4: New averaged gradients
__group_805__new_moment_2 (optional) - T4: New averaged squared gradients
__group_805__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_806__new_weights (optional) - T2: New weights
__group_806__new_gradients (optional) - T3: New gradients
__group_806__new_moment_1 (optional) - T4: New averaged gradients
__group_806__new_moment_2 (optional) - T4: New averaged squared gradients
__group_806__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_807__new_weights (optional) - T2: New weights
__group_807__new_gradients (optional) - T3: New gradients
__group_807__new_moment_1 (optional) - T4: New averaged gradients
__group_807__new_moment_2 (optional) - T4: New averaged squared gradients
__group_807__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_808__new_weights (optional) - T2: New weights
__group_808__new_gradients (optional) - T3: New gradients
__group_808__new_moment_1 (optional) - T4: New averaged gradients
__group_808__new_moment_2 (optional) - T4: New averaged squared gradients
__group_808__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_809__new_weights (optional) - T2: New weights
__group_809__new_gradients (optional) - T3: New gradients
__group_809__new_moment_1 (optional) - T4: New averaged gradients
__group_809__new_moment_2 (optional) - T4: New averaged squared gradients
__group_809__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_810__new_weights (optional) - T2: New weights
__group_810__new_gradients (optional) - T3: New gradients
__group_810__new_moment_1 (optional) - T4: New averaged gradients
__group_810__new_moment_2 (optional) - T4: New averaged squared gradients
__group_810__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_811__new_weights (optional) - T2: New weights
__group_811__new_gradients (optional) - T3: New gradients
__group_811__new_moment_1 (optional) - T4: New averaged gradients
__group_811__new_moment_2 (optional) - T4: New averaged squared gradients
__group_811__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_812__new_weights (optional) - T2: New weights
__group_812__new_gradients (optional) - T3: New gradients
__group_812__new_moment_1 (optional) - T4: New averaged gradients
__group_812__new_moment_2 (optional) - T4: New averaged squared gradients
__group_812__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_813__new_weights (optional) - T2: New weights
__group_813__new_gradients (optional) - T3: New gradients
__group_813__new_moment_1 (optional) - T4: New averaged gradients
__group_813__new_moment_2 (optional) - T4: New averaged squared gradients
__group_813__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_814__new_weights (optional) - T2: New weights
__group_814__new_gradients (optional) - T3: New gradients
__group_814__new_moment_1 (optional) - T4: New averaged gradients
__group_814__new_moment_2 (optional) - T4: New averaged squared gradients
__group_814__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_815__new_weights (optional) - T2: New weights
__group_815__new_gradients (optional) - T3: New gradients
__group_815__new_moment_1 (optional) - T4: New averaged gradients
__group_815__new_moment_2 (optional) - T4: New averaged squared gradients
__group_815__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_816__new_weights (optional) - T2: New weights
__group_816__new_gradients (optional) - T3: New gradients
__group_816__new_moment_1 (optional) - T4: New averaged gradients
__group_816__new_moment_2 (optional) - T4: New averaged squared gradients
__group_816__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_817__new_weights (optional) - T2: New weights
__group_817__new_gradients (optional) - T3: New gradients
__group_817__new_moment_1 (optional) - T4: New averaged gradients
__group_817__new_moment_2 (optional) - T4: New averaged squared gradients
__group_817__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_818__new_weights (optional) - T2: New weights
__group_818__new_gradients (optional) - T3: New gradients
__group_818__new_moment_1 (optional) - T4: New averaged gradients
__group_818__new_moment_2 (optional) - T4: New averaged squared gradients
__group_818__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_819__new_weights (optional) - T2: New weights
__group_819__new_gradients (optional) - T3: New gradients
__group_819__new_moment_1 (optional) - T4: New averaged gradients
__group_819__new_moment_2 (optional) - T4: New averaged squared gradients
__group_819__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_820__new_weights (optional) - T2: New weights
__group_820__new_gradients (optional) - T3: New gradients
__group_820__new_moment_1 (optional) - T4: New averaged gradients
__group_820__new_moment_2 (optional) - T4: New averaged squared gradients
__group_820__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_821__new_weights (optional) - T2: New weights
__group_821__new_gradients (optional) - T3: New gradients
__group_821__new_moment_1 (optional) - T4: New averaged gradients
__group_821__new_moment_2 (optional) - T4: New averaged squared gradients
__group_821__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_822__new_weights (optional) - T2: New weights
__group_822__new_gradients (optional) - T3: New gradients
__group_822__new_moment_1 (optional) - T4: New averaged gradients
__group_822__new_moment_2 (optional) - T4: New averaged squared gradients
__group_822__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_823__new_weights (optional) - T2: New weights
__group_823__new_gradients (optional) - T3: New gradients
__group_823__new_moment_1 (optional) - T4: New averaged gradients
__group_823__new_moment_2 (optional) - T4: New averaged squared gradients
__group_823__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_824__new_weights (optional) - T2: New weights
__group_824__new_gradients (optional) - T3: New gradients
__group_824__new_moment_1 (optional) - T4: New averaged gradients
__group_824__new_moment_2 (optional) - T4: New averaged squared gradients
__group_824__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_825__new_weights (optional) - T2: New weights
__group_825__new_gradients (optional) - T3: New gradients
__group_825__new_moment_1 (optional) - T4: New averaged gradients
__group_825__new_moment_2 (optional) - T4: New averaged squared gradients
__group_825__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_826__new_weights (optional) - T2: New weights
__group_826__new_gradients (optional) - T3: New gradients
__group_826__new_moment_1 (optional) - T4: New averaged gradients
__group_826__new_moment_2 (optional) - T4: New averaged squared gradients
__group_826__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_827__new_weights (optional) - T2: New weights
__group_827__new_gradients (optional) - T3: New gradients
__group_827__new_moment_1 (optional) - T4: New averaged gradients
__group_827__new_moment_2 (optional) - T4: New averaged squared gradients
__group_827__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_828__new_weights (optional) - T2: New weights
__group_828__new_gradients (optional) - T3: New gradients
__group_828__new_moment_1 (optional) - T4: New averaged gradients
__group_828__new_moment_2 (optional) - T4: New averaged squared gradients
__group_828__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_829__new_weights (optional) - T2: New weights
__group_829__new_gradients (optional) - T3: New gradients
__group_829__new_moment_1 (optional) - T4: New averaged gradients
__group_829__new_moment_2 (optional) - T4: New averaged squared gradients
__group_829__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_830__new_weights (optional) - T2: New weights
__group_830__new_gradients (optional) - T3: New gradients
__group_830__new_moment_1 (optional) - T4: New averaged gradients
__group_830__new_moment_2 (optional) - T4: New averaged squared gradients
__group_830__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_831__new_weights (optional) - T2: New weights
__group_831__new_gradients (optional) - T3: New gradients
__group_831__new_moment_1 (optional) - T4: New averaged gradients
__group_831__new_moment_2 (optional) - T4: New averaged squared gradients
__group_831__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_832__new_weights (optional) - T2: New weights
__group_832__new_gradients (optional) - T3: New gradients
__group_832__new_moment_1 (optional) - T4: New averaged gradients
__group_832__new_moment_2 (optional) - T4: New averaged squared gradients
__group_832__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_833__new_weights (optional) - T2: New weights
__group_833__new_gradients (optional) - T3: New gradients
__group_833__new_moment_1 (optional) - T4: New averaged gradients
__group_833__new_moment_2 (optional) - T4: New averaged squared gradients
__group_833__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_834__new_weights (optional) - T2: New weights
__group_834__new_gradients (optional) - T3: New gradients
__group_834__new_moment_1 (optional) - T4: New averaged gradients
__group_834__new_moment_2 (optional) - T4: New averaged squared gradients
__group_834__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_835__new_weights (optional) - T2: New weights
__group_835__new_gradients (optional) - T3: New gradients
__group_835__new_moment_1 (optional) - T4: New averaged gradients
__group_835__new_moment_2 (optional) - T4: New averaged squared gradients
__group_835__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_836__new_weights (optional) - T2: New weights
__group_836__new_gradients (optional) - T3: New gradients
__group_836__new_moment_1 (optional) - T4: New averaged gradients
__group_836__new_moment_2 (optional) - T4: New averaged squared gradients
__group_836__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_837__new_weights (optional) - T2: New weights
__group_837__new_gradients (optional) - T3: New gradients
__group_837__new_moment_1 (optional) - T4: New averaged gradients
__group_837__new_moment_2 (optional) - T4: New averaged squared gradients
__group_837__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_838__new_weights (optional) - T2: New weights
__group_838__new_gradients (optional) - T3: New gradients
__group_838__new_moment_1 (optional) - T4: New averaged gradients
__group_838__new_moment_2 (optional) - T4: New averaged squared gradients
__group_838__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_839__new_weights (optional) - T2: New weights
__group_839__new_gradients (optional) - T3: New gradients
__group_839__new_moment_1 (optional) - T4: New averaged gradients
__group_839__new_moment_2 (optional) - T4: New averaged squared gradients
__group_839__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_840__new_weights (optional) - T2: New weights
__group_840__new_gradients (optional) - T3: New gradients
__group_840__new_moment_1 (optional) - T4: New averaged gradients
__group_840__new_moment_2 (optional) - T4: New averaged squared gradients
__group_840__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_841__new_weights (optional) - T2: New weights
__group_841__new_gradients (optional) - T3: New gradients
__group_841__new_moment_1 (optional) - T4: New averaged gradients
__group_841__new_moment_2 (optional) - T4: New averaged squared gradients
__group_841__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_842__new_weights (optional) - T2: New weights
__group_842__new_gradients (optional) - T3: New gradients
__group_842__new_moment_1 (optional) - T4: New averaged gradients
__group_842__new_moment_2 (optional) - T4: New averaged squared gradients
__group_842__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_843__new_weights (optional) - T2: New weights
__group_843__new_gradients (optional) - T3: New gradients
__group_843__new_moment_1 (optional) - T4: New averaged gradients
__group_843__new_moment_2 (optional) - T4: New averaged squared gradients
__group_843__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_844__new_weights (optional) - T2: New weights
__group_844__new_gradients (optional) - T3: New gradients
__group_844__new_moment_1 (optional) - T4: New averaged gradients
__group_844__new_moment_2 (optional) - T4: New averaged squared gradients
__group_844__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_845__new_weights (optional) - T2: New weights
__group_845__new_gradients (optional) - T3: New gradients
__group_845__new_moment_1 (optional) - T4: New averaged gradients
__group_845__new_moment_2 (optional) - T4: New averaged squared gradients
__group_845__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_846__new_weights (optional) - T2: New weights
__group_846__new_gradients (optional) - T3: New gradients
__group_846__new_moment_1 (optional) - T4: New averaged gradients
__group_846__new_moment_2 (optional) - T4: New averaged squared gradients
__group_846__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_847__new_weights (optional) - T2: New weights
__group_847__new_gradients (optional) - T3: New gradients
__group_847__new_moment_1 (optional) - T4: New averaged gradients
__group_847__new_moment_2 (optional) - T4: New averaged squared gradients
__group_847__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_848__new_weights (optional) - T2: New weights
__group_848__new_gradients (optional) - T3: New gradients
__group_848__new_moment_1 (optional) - T4: New averaged gradients
__group_848__new_moment_2 (optional) - T4: New averaged squared gradients
__group_848__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_849__new_weights (optional) - T2: New weights
__group_849__new_gradients (optional) - T3: New gradients
__group_849__new_moment_1 (optional) - T4: New averaged gradients
__group_849__new_moment_2 (optional) - T4: New averaged squared gradients
__group_849__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_850__new_weights (optional) - T2: New weights
__group_850__new_gradients (optional) - T3: New gradients
__group_850__new_moment_1 (optional) - T4: New averaged gradients
__group_850__new_moment_2 (optional) - T4: New averaged squared gradients
__group_850__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_851__new_weights (optional) - T2: New weights
__group_851__new_gradients (optional) - T3: New gradients
__group_851__new_moment_1 (optional) - T4: New averaged gradients
__group_851__new_moment_2 (optional) - T4: New averaged squared gradients
__group_851__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_852__new_weights (optional) - T2: New weights
__group_852__new_gradients (optional) - T3: New gradients
__group_852__new_moment_1 (optional) - T4: New averaged gradients
__group_852__new_moment_2 (optional) - T4: New averaged squared gradients
__group_852__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_853__new_weights (optional) - T2: New weights
__group_853__new_gradients (optional) - T3: New gradients
__group_853__new_moment_1 (optional) - T4: New averaged gradients
__group_853__new_moment_2 (optional) - T4: New averaged squared gradients
__group_853__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_854__new_weights (optional) - T2: New weights
__group_854__new_gradients (optional) - T3: New gradients
__group_854__new_moment_1 (optional) - T4: New averaged gradients
__group_854__new_moment_2 (optional) - T4: New averaged squared gradients
__group_854__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_855__new_weights (optional) - T2: New weights
__group_855__new_gradients (optional) - T3: New gradients
__group_855__new_moment_1 (optional) - T4: New averaged gradients
__group_855__new_moment_2 (optional) - T4: New averaged squared gradients
__group_855__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_856__new_weights (optional) - T2: New weights
__group_856__new_gradients (optional) - T3: New gradients
__group_856__new_moment_1 (optional) - T4: New averaged gradients
__group_856__new_moment_2 (optional) - T4: New averaged squared gradients
__group_856__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_857__new_weights (optional) - T2: New weights
__group_857__new_gradients (optional) - T3: New gradients
__group_857__new_moment_1 (optional) - T4: New averaged gradients
__group_857__new_moment_2 (optional) - T4: New averaged squared gradients
__group_857__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_858__new_weights (optional) - T2: New weights
__group_858__new_gradients (optional) - T3: New gradients
__group_858__new_moment_1 (optional) - T4: New averaged gradients
__group_858__new_moment_2 (optional) - T4: New averaged squared gradients
__group_858__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_859__new_weights (optional) - T2: New weights
__group_859__new_gradients (optional) - T3: New gradients
__group_859__new_moment_1 (optional) - T4: New averaged gradients
__group_859__new_moment_2 (optional) - T4: New averaged squared gradients
__group_859__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_860__new_weights (optional) - T2: New weights
__group_860__new_gradients (optional) - T3: New gradients
__group_860__new_moment_1 (optional) - T4: New averaged gradients
__group_860__new_moment_2 (optional) - T4: New averaged squared gradients
__group_860__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_861__new_weights (optional) - T2: New weights
__group_861__new_gradients (optional) - T3: New gradients
__group_861__new_moment_1 (optional) - T4: New averaged gradients
__group_861__new_moment_2 (optional) - T4: New averaged squared gradients
__group_861__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_862__new_weights (optional) - T2: New weights
__group_862__new_gradients (optional) - T3: New gradients
__group_862__new_moment_1 (optional) - T4: New averaged gradients
__group_862__new_moment_2 (optional) - T4: New averaged squared gradients
__group_862__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_863__new_weights (optional) - T2: New weights
__group_863__new_gradients (optional) - T3: New gradients
__group_863__new_moment_1 (optional) - T4: New averaged gradients
__group_863__new_moment_2 (optional) - T4: New averaged squared gradients
__group_863__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_864__new_weights (optional) - T2: New weights
__group_864__new_gradients (optional) - T3: New gradients
__group_864__new_moment_1 (optional) - T4: New averaged gradients
__group_864__new_moment_2 (optional) - T4: New averaged squared gradients
__group_864__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_865__new_weights (optional) - T2: New weights
__group_865__new_gradients (optional) - T3: New gradients
__group_865__new_moment_1 (optional) - T4: New averaged gradients
__group_865__new_moment_2 (optional) - T4: New averaged squared gradients
__group_865__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_866__new_weights (optional) - T2: New weights
__group_866__new_gradients (optional) - T3: New gradients
__group_866__new_moment_1 (optional) - T4: New averaged gradients
__group_866__new_moment_2 (optional) - T4: New averaged squared gradients
__group_866__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_867__new_weights (optional) - T2: New weights
__group_867__new_gradients (optional) - T3: New gradients
__group_867__new_moment_1 (optional) - T4: New averaged gradients
__group_867__new_moment_2 (optional) - T4: New averaged squared gradients
__group_867__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_868__new_weights (optional) - T2: New weights
__group_868__new_gradients (optional) - T3: New gradients
__group_868__new_moment_1 (optional) - T4: New averaged gradients
__group_868__new_moment_2 (optional) - T4: New averaged squared gradients
__group_868__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_869__new_weights (optional) - T2: New weights
__group_869__new_gradients (optional) - T3: New gradients
__group_869__new_moment_1 (optional) - T4: New averaged gradients
__group_869__new_moment_2 (optional) - T4: New averaged squared gradients
__group_869__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_870__new_weights (optional) - T2: New weights
__group_870__new_gradients (optional) - T3: New gradients
__group_870__new_moment_1 (optional) - T4: New averaged gradients
__group_870__new_moment_2 (optional) - T4: New averaged squared gradients
__group_870__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_871__new_weights (optional) - T2: New weights
__group_871__new_gradients (optional) - T3: New gradients
__group_871__new_moment_1 (optional) - T4: New averaged gradients
__group_871__new_moment_2 (optional) - T4: New averaged squared gradients
__group_871__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_872__new_weights (optional) - T2: New weights
__group_872__new_gradients (optional) - T3: New gradients
__group_872__new_moment_1 (optional) - T4: New averaged gradients
__group_872__new_moment_2 (optional) - T4: New averaged squared gradients
__group_872__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_873__new_weights (optional) - T2: New weights
__group_873__new_gradients (optional) - T3: New gradients
__group_873__new_moment_1 (optional) - T4: New averaged gradients
__group_873__new_moment_2 (optional) - T4: New averaged squared gradients
__group_873__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_874__new_weights (optional) - T2: New weights
__group_874__new_gradients (optional) - T3: New gradients
__group_874__new_moment_1 (optional) - T4: New averaged gradients
__group_874__new_moment_2 (optional) - T4: New averaged squared gradients
__group_874__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_875__new_weights (optional) - T2: New weights
__group_875__new_gradients (optional) - T3: New gradients
__group_875__new_moment_1 (optional) - T4: New averaged gradients
__group_875__new_moment_2 (optional) - T4: New averaged squared gradients
__group_875__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_876__new_weights (optional) - T2: New weights
__group_876__new_gradients (optional) - T3: New gradients
__group_876__new_moment_1 (optional) - T4: New averaged gradients
__group_876__new_moment_2 (optional) - T4: New averaged squared gradients
__group_876__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_877__new_weights (optional) - T2: New weights
__group_877__new_gradients (optional) - T3: New gradients
__group_877__new_moment_1 (optional) - T4: New averaged gradients
__group_877__new_moment_2 (optional) - T4: New averaged squared gradients
__group_877__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_878__new_weights (optional) - T2: New weights
__group_878__new_gradients (optional) - T3: New gradients
__group_878__new_moment_1 (optional) - T4: New averaged gradients
__group_878__new_moment_2 (optional) - T4: New averaged squared gradients
__group_878__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_879__new_weights (optional) - T2: New weights
__group_879__new_gradients (optional) - T3: New gradients
__group_879__new_moment_1 (optional) - T4: New averaged gradients
__group_879__new_moment_2 (optional) - T4: New averaged squared gradients
__group_879__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_880__new_weights (optional) - T2: New weights
__group_880__new_gradients (optional) - T3: New gradients
__group_880__new_moment_1 (optional) - T4: New averaged gradients
__group_880__new_moment_2 (optional) - T4: New averaged squared gradients
__group_880__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_881__new_weights (optional) - T2: New weights
__group_881__new_gradients (optional) - T3: New gradients
__group_881__new_moment_1 (optional) - T4: New averaged gradients
__group_881__new_moment_2 (optional) - T4: New averaged squared gradients
__group_881__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_882__new_weights (optional) - T2: New weights
__group_882__new_gradients (optional) - T3: New gradients
__group_882__new_moment_1 (optional) - T4: New averaged gradients
__group_882__new_moment_2 (optional) - T4: New averaged squared gradients
__group_882__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_883__new_weights (optional) - T2: New weights
__group_883__new_gradients (optional) - T3: New gradients
__group_883__new_moment_1 (optional) - T4: New averaged gradients
__group_883__new_moment_2 (optional) - T4: New averaged squared gradients
__group_883__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_884__new_weights (optional) - T2: New weights
__group_884__new_gradients (optional) - T3: New gradients
__group_884__new_moment_1 (optional) - T4: New averaged gradients
__group_884__new_moment_2 (optional) - T4: New averaged squared gradients
__group_884__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_885__new_weights (optional) - T2: New weights
__group_885__new_gradients (optional) - T3: New gradients
__group_885__new_moment_1 (optional) - T4: New averaged gradients
__group_885__new_moment_2 (optional) - T4: New averaged squared gradients
__group_885__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_886__new_weights (optional) - T2: New weights
__group_886__new_gradients (optional) - T3: New gradients
__group_886__new_moment_1 (optional) - T4: New averaged gradients
__group_886__new_moment_2 (optional) - T4: New averaged squared gradients
__group_886__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_887__new_weights (optional) - T2: New weights
__group_887__new_gradients (optional) - T3: New gradients
__group_887__new_moment_1 (optional) - T4: New averaged gradients
__group_887__new_moment_2 (optional) - T4: New averaged squared gradients
__group_887__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_888__new_weights (optional) - T2: New weights
__group_888__new_gradients (optional) - T3: New gradients
__group_888__new_moment_1 (optional) - T4: New averaged gradients
__group_888__new_moment_2 (optional) - T4: New averaged squared gradients
__group_888__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_889__new_weights (optional) - T2: New weights
__group_889__new_gradients (optional) - T3: New gradients
__group_889__new_moment_1 (optional) - T4: New averaged gradients
__group_889__new_moment_2 (optional) - T4: New averaged squared gradients
__group_889__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_890__new_weights (optional) - T2: New weights
__group_890__new_gradients (optional) - T3: New gradients
__group_890__new_moment_1 (optional) - T4: New averaged gradients
__group_890__new_moment_2 (optional) - T4: New averaged squared gradients
__group_890__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_891__new_weights (optional) - T2: New weights
__group_891__new_gradients (optional) - T3: New gradients
__group_891__new_moment_1 (optional) - T4: New averaged gradients
__group_891__new_moment_2 (optional) - T4: New averaged squared gradients
__group_891__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_892__new_weights (optional) - T2: New weights
__group_892__new_gradients (optional) - T3: New gradients
__group_892__new_moment_1 (optional) - T4: New averaged gradients
__group_892__new_moment_2 (optional) - T4: New averaged squared gradients
__group_892__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_893__new_weights (optional) - T2: New weights
__group_893__new_gradients (optional) - T3: New gradients
__group_893__new_moment_1 (optional) - T4: New averaged gradients
__group_893__new_moment_2 (optional) - T4: New averaged squared gradients
__group_893__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_894__new_weights (optional) - T2: New weights
__group_894__new_gradients (optional) - T3: New gradients
__group_894__new_moment_1 (optional) - T4: New averaged gradients
__group_894__new_moment_2 (optional) - T4: New averaged squared gradients
__group_894__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_895__new_weights (optional) - T2: New weights
__group_895__new_gradients (optional) - T3: New gradients
__group_895__new_moment_1 (optional) - T4: New averaged gradients
__group_895__new_moment_2 (optional) - T4: New averaged squared gradients
__group_895__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_896__new_weights (optional) - T2: New weights
__group_896__new_gradients (optional) - T3: New gradients
__group_896__new_moment_1 (optional) - T4: New averaged gradients
__group_896__new_moment_2 (optional) - T4: New averaged squared gradients
__group_896__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_897__new_weights (optional) - T2: New weights
__group_897__new_gradients (optional) - T3: New gradients
__group_897__new_moment_1 (optional) - T4: New averaged gradients
__group_897__new_moment_2 (optional) - T4: New averaged squared gradients
__group_897__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_898__new_weights (optional) - T2: New weights
__group_898__new_gradients (optional) - T3: New gradients
__group_898__new_moment_1 (optional) - T4: New averaged gradients
__group_898__new_moment_2 (optional) - T4: New averaged squared gradients
__group_898__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_899__new_weights (optional) - T2: New weights
__group_899__new_gradients (optional) - T3: New gradients
__group_899__new_moment_1 (optional) - T4: New averaged gradients
__group_899__new_moment_2 (optional) - T4: New averaged squared gradients
__group_899__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_900__new_weights (optional) - T2: New weights
__group_900__new_gradients (optional) - T3: New gradients
__group_900__new_moment_1 (optional) - T4: New averaged gradients
__group_900__new_moment_2 (optional) - T4: New averaged squared gradients
__group_900__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_901__new_weights (optional) - T2: New weights
__group_901__new_gradients (optional) - T3: New gradients
__group_901__new_moment_1 (optional) - T4: New averaged gradients
__group_901__new_moment_2 (optional) - T4: New averaged squared gradients
__group_901__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_902__new_weights (optional) - T2: New weights
__group_902__new_gradients (optional) - T3: New gradients
__group_902__new_moment_1 (optional) - T4: New averaged gradients
__group_902__new_moment_2 (optional) - T4: New averaged squared gradients
__group_902__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_903__new_weights (optional) - T2: New weights
__group_903__new_gradients (optional) - T3: New gradients
__group_903__new_moment_1 (optional) - T4: New averaged gradients
__group_903__new_moment_2 (optional) - T4: New averaged squared gradients
__group_903__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_904__new_weights (optional) - T2: New weights
__group_904__new_gradients (optional) - T3: New gradients
__group_904__new_moment_1 (optional) - T4: New averaged gradients
__group_904__new_moment_2 (optional) - T4: New averaged squared gradients
__group_904__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_905__new_weights (optional) - T2: New weights
__group_905__new_gradients (optional) - T3: New gradients
__group_905__new_moment_1 (optional) - T4: New averaged gradients
__group_905__new_moment_2 (optional) - T4: New averaged squared gradients
__group_905__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_906__new_weights (optional) - T2: New weights
__group_906__new_gradients (optional) - T3: New gradients
__group_906__new_moment_1 (optional) - T4: New averaged gradients
__group_906__new_moment_2 (optional) - T4: New averaged squared gradients
__group_906__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_907__new_weights (optional) - T2: New weights
__group_907__new_gradients (optional) - T3: New gradients
__group_907__new_moment_1 (optional) - T4: New averaged gradients
__group_907__new_moment_2 (optional) - T4: New averaged squared gradients
__group_907__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_908__new_weights (optional) - T2: New weights
__group_908__new_gradients (optional) - T3: New gradients
__group_908__new_moment_1 (optional) - T4: New averaged gradients
__group_908__new_moment_2 (optional) - T4: New averaged squared gradients
__group_908__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_909__new_weights (optional) - T2: New weights
__group_909__new_gradients (optional) - T3: New gradients
__group_909__new_moment_1 (optional) - T4: New averaged gradients
__group_909__new_moment_2 (optional) - T4: New averaged squared gradients
__group_909__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_910__new_weights (optional) - T2: New weights
__group_910__new_gradients (optional) - T3: New gradients
__group_910__new_moment_1 (optional) - T4: New averaged gradients
__group_910__new_moment_2 (optional) - T4: New averaged squared gradients
__group_910__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_911__new_weights (optional) - T2: New weights
__group_911__new_gradients (optional) - T3: New gradients
__group_911__new_moment_1 (optional) - T4: New averaged gradients
__group_911__new_moment_2 (optional) - T4: New averaged squared gradients
__group_911__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_912__new_weights (optional) - T2: New weights
__group_912__new_gradients (optional) - T3: New gradients
__group_912__new_moment_1 (optional) - T4: New averaged gradients
__group_912__new_moment_2 (optional) - T4: New averaged squared gradients
__group_912__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_913__new_weights (optional) - T2: New weights
__group_913__new_gradients (optional) - T3: New gradients
__group_913__new_moment_1 (optional) - T4: New averaged gradients
__group_913__new_moment_2 (optional) - T4: New averaged squared gradients
__group_913__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_914__new_weights (optional) - T2: New weights
__group_914__new_gradients (optional) - T3: New gradients
__group_914__new_moment_1 (optional) - T4: New averaged gradients
__group_914__new_moment_2 (optional) - T4: New averaged squared gradients
__group_914__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_915__new_weights (optional) - T2: New weights
__group_915__new_gradients (optional) - T3: New gradients
__group_915__new_moment_1 (optional) - T4: New averaged gradients
__group_915__new_moment_2 (optional) - T4: New averaged squared gradients
__group_915__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_916__new_weights (optional) - T2: New weights
__group_916__new_gradients (optional) - T3: New gradients
__group_916__new_moment_1 (optional) - T4: New averaged gradients
__group_916__new_moment_2 (optional) - T4: New averaged squared gradients
__group_916__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_917__new_weights (optional) - T2: New weights
__group_917__new_gradients (optional) - T3: New gradients
__group_917__new_moment_1 (optional) - T4: New averaged gradients
__group_917__new_moment_2 (optional) - T4: New averaged squared gradients
__group_917__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_918__new_weights (optional) - T2: New weights
__group_918__new_gradients (optional) - T3: New gradients
__group_918__new_moment_1 (optional) - T4: New averaged gradients
__group_918__new_moment_2 (optional) - T4: New averaged squared gradients
__group_918__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_919__new_weights (optional) - T2: New weights
__group_919__new_gradients (optional) - T3: New gradients
__group_919__new_moment_1 (optional) - T4: New averaged gradients
__group_919__new_moment_2 (optional) - T4: New averaged squared gradients
__group_919__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_920__new_weights (optional) - T2: New weights
__group_920__new_gradients (optional) - T3: New gradients
__group_920__new_moment_1 (optional) - T4: New averaged gradients
__group_920__new_moment_2 (optional) - T4: New averaged squared gradients
__group_920__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_921__new_weights (optional) - T2: New weights
__group_921__new_gradients (optional) - T3: New gradients
__group_921__new_moment_1 (optional) - T4: New averaged gradients
__group_921__new_moment_2 (optional) - T4: New averaged squared gradients
__group_921__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_922__new_weights (optional) - T2: New weights
__group_922__new_gradients (optional) - T3: New gradients
__group_922__new_moment_1 (optional) - T4: New averaged gradients
__group_922__new_moment_2 (optional) - T4: New averaged squared gradients
__group_922__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_923__new_weights (optional) - T2: New weights
__group_923__new_gradients (optional) - T3: New gradients
__group_923__new_moment_1 (optional) - T4: New averaged gradients
__group_923__new_moment_2 (optional) - T4: New averaged squared gradients
__group_923__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_924__new_weights (optional) - T2: New weights
__group_924__new_gradients (optional) - T3: New gradients
__group_924__new_moment_1 (optional) - T4: New averaged gradients
__group_924__new_moment_2 (optional) - T4: New averaged squared gradients
__group_924__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_925__new_weights (optional) - T2: New weights
__group_925__new_gradients (optional) - T3: New gradients
__group_925__new_moment_1 (optional) - T4: New averaged gradients
__group_925__new_moment_2 (optional) - T4: New averaged squared gradients
__group_925__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_926__new_weights (optional) - T2: New weights
__group_926__new_gradients (optional) - T3: New gradients
__group_926__new_moment_1 (optional) - T4: New averaged gradients
__group_926__new_moment_2 (optional) - T4: New averaged squared gradients
__group_926__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_927__new_weights (optional) - T2: New weights
__group_927__new_gradients (optional) - T3: New gradients
__group_927__new_moment_1 (optional) - T4: New averaged gradients
__group_927__new_moment_2 (optional) - T4: New averaged squared gradients
__group_927__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_928__new_weights (optional) - T2: New weights
__group_928__new_gradients (optional) - T3: New gradients
__group_928__new_moment_1 (optional) - T4: New averaged gradients
__group_928__new_moment_2 (optional) - T4: New averaged squared gradients
__group_928__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_929__new_weights (optional) - T2: New weights
__group_929__new_gradients (optional) - T3: New gradients
__group_929__new_moment_1 (optional) - T4: New averaged gradients
__group_929__new_moment_2 (optional) - T4: New averaged squared gradients
__group_929__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_930__new_weights (optional) - T2: New weights
__group_930__new_gradients (optional) - T3: New gradients
__group_930__new_moment_1 (optional) - T4: New averaged gradients
__group_930__new_moment_2 (optional) - T4: New averaged squared gradients
__group_930__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_931__new_weights (optional) - T2: New weights
__group_931__new_gradients (optional) - T3: New gradients
__group_931__new_moment_1 (optional) - T4: New averaged gradients
__group_931__new_moment_2 (optional) - T4: New averaged squared gradients
__group_931__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_932__new_weights (optional) - T2: New weights
__group_932__new_gradients (optional) - T3: New gradients
__group_932__new_moment_1 (optional) - T4: New averaged gradients
__group_932__new_moment_2 (optional) - T4: New averaged squared gradients
__group_932__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_933__new_weights (optional) - T2: New weights
__group_933__new_gradients (optional) - T3: New gradients
__group_933__new_moment_1 (optional) - T4: New averaged gradients
__group_933__new_moment_2 (optional) - T4: New averaged squared gradients
__group_933__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_934__new_weights (optional) - T2: New weights
__group_934__new_gradients (optional) - T3: New gradients
__group_934__new_moment_1 (optional) - T4: New averaged gradients
__group_934__new_moment_2 (optional) - T4: New averaged squared gradients
__group_934__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_935__new_weights (optional) - T2: New weights
__group_935__new_gradients (optional) - T3: New gradients
__group_935__new_moment_1 (optional) - T4: New averaged gradients
__group_935__new_moment_2 (optional) - T4: New averaged squared gradients
__group_935__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_936__new_weights (optional) - T2: New weights
__group_936__new_gradients (optional) - T3: New gradients
__group_936__new_moment_1 (optional) - T4: New averaged gradients
__group_936__new_moment_2 (optional) - T4: New averaged squared gradients
__group_936__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_937__new_weights (optional) - T2: New weights
__group_937__new_gradients (optional) - T3: New gradients
__group_937__new_moment_1 (optional) - T4: New averaged gradients
__group_937__new_moment_2 (optional) - T4: New averaged squared gradients
__group_937__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_938__new_weights (optional) - T2: New weights
__group_938__new_gradients (optional) - T3: New gradients
__group_938__new_moment_1 (optional) - T4: New averaged gradients
__group_938__new_moment_2 (optional) - T4: New averaged squared gradients
__group_938__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_939__new_weights (optional) - T2: New weights
__group_939__new_gradients (optional) - T3: New gradients
__group_939__new_moment_1 (optional) - T4: New averaged gradients
__group_939__new_moment_2 (optional) - T4: New averaged squared gradients
__group_939__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_940__new_weights (optional) - T2: New weights
__group_940__new_gradients (optional) - T3: New gradients
__group_940__new_moment_1 (optional) - T4: New averaged gradients
__group_940__new_moment_2 (optional) - T4: New averaged squared gradients
__group_940__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_941__new_weights (optional) - T2: New weights
__group_941__new_gradients (optional) - T3: New gradients
__group_941__new_moment_1 (optional) - T4: New averaged gradients
__group_941__new_moment_2 (optional) - T4: New averaged squared gradients
__group_941__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_942__new_weights (optional) - T2: New weights
__group_942__new_gradients (optional) - T3: New gradients
__group_942__new_moment_1 (optional) - T4: New averaged gradients
__group_942__new_moment_2 (optional) - T4: New averaged squared gradients
__group_942__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_943__new_weights (optional) - T2: New weights
__group_943__new_gradients (optional) - T3: New gradients
__group_943__new_moment_1 (optional) - T4: New averaged gradients
__group_943__new_moment_2 (optional) - T4: New averaged squared gradients
__group_943__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_944__new_weights (optional) - T2: New weights
__group_944__new_gradients (optional) - T3: New gradients
__group_944__new_moment_1 (optional) - T4: New averaged gradients
__group_944__new_moment_2 (optional) - T4: New averaged squared gradients
__group_944__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_945__new_weights (optional) - T2: New weights
__group_945__new_gradients (optional) - T3: New gradients
__group_945__new_moment_1 (optional) - T4: New averaged gradients
__group_945__new_moment_2 (optional) - T4: New averaged squared gradients
__group_945__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_946__new_weights (optional) - T2: New weights
__group_946__new_gradients (optional) - T3: New gradients
__group_946__new_moment_1 (optional) - T4: New averaged gradients
__group_946__new_moment_2 (optional) - T4: New averaged squared gradients
__group_946__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_947__new_weights (optional) - T2: New weights
__group_947__new_gradients (optional) - T3: New gradients
__group_947__new_moment_1 (optional) - T4: New averaged gradients
__group_947__new_moment_2 (optional) - T4: New averaged squared gradients
__group_947__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_948__new_weights (optional) - T2: New weights
__group_948__new_gradients (optional) - T3: New gradients
__group_948__new_moment_1 (optional) - T4: New averaged gradients
__group_948__new_moment_2 (optional) - T4: New averaged squared gradients
__group_948__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_949__new_weights (optional) - T2: New weights
__group_949__new_gradients (optional) - T3: New gradients
__group_949__new_moment_1 (optional) - T4: New averaged gradients
__group_949__new_moment_2 (optional) - T4: New averaged squared gradients
__group_949__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_950__new_weights (optional) - T2: New weights
__group_950__new_gradients (optional) - T3: New gradients
__group_950__new_moment_1 (optional) - T4: New averaged gradients
__group_950__new_moment_2 (optional) - T4: New averaged squared gradients
__group_950__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_951__new_weights (optional) - T2: New weights
__group_951__new_gradients (optional) - T3: New gradients
__group_951__new_moment_1 (optional) - T4: New averaged gradients
__group_951__new_moment_2 (optional) - T4: New averaged squared gradients
__group_951__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_952__new_weights (optional) - T2: New weights
__group_952__new_gradients (optional) - T3: New gradients
__group_952__new_moment_1 (optional) - T4: New averaged gradients
__group_952__new_moment_2 (optional) - T4: New averaged squared gradients
__group_952__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_953__new_weights (optional) - T2: New weights
__group_953__new_gradients (optional) - T3: New gradients
__group_953__new_moment_1 (optional) - T4: New averaged gradients
__group_953__new_moment_2 (optional) - T4: New averaged squared gradients
__group_953__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_954__new_weights (optional) - T2: New weights
__group_954__new_gradients (optional) - T3: New gradients
__group_954__new_moment_1 (optional) - T4: New averaged gradients
__group_954__new_moment_2 (optional) - T4: New averaged squared gradients
__group_954__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_955__new_weights (optional) - T2: New weights
__group_955__new_gradients (optional) - T3: New gradients
__group_955__new_moment_1 (optional) - T4: New averaged gradients
__group_955__new_moment_2 (optional) - T4: New averaged squared gradients
__group_955__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_956__new_weights (optional) - T2: New weights
__group_956__new_gradients (optional) - T3: New gradients
__group_956__new_moment_1 (optional) - T4: New averaged gradients
__group_956__new_moment_2 (optional) - T4: New averaged squared gradients
__group_956__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_957__new_weights (optional) - T2: New weights
__group_957__new_gradients (optional) - T3: New gradients
__group_957__new_moment_1 (optional) - T4: New averaged gradients
__group_957__new_moment_2 (optional) - T4: New averaged squared gradients
__group_957__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_958__new_weights (optional) - T2: New weights
__group_958__new_gradients (optional) - T3: New gradients
__group_958__new_moment_1 (optional) - T4: New averaged gradients
__group_958__new_moment_2 (optional) - T4: New averaged squared gradients
__group_958__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_959__new_weights (optional) - T2: New weights
__group_959__new_gradients (optional) - T3: New gradients
__group_959__new_moment_1 (optional) - T4: New averaged gradients
__group_959__new_moment_2 (optional) - T4: New averaged squared gradients
__group_959__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_960__new_weights (optional) - T2: New weights
__group_960__new_gradients (optional) - T3: New gradients
__group_960__new_moment_1 (optional) - T4: New averaged gradients
__group_960__new_moment_2 (optional) - T4: New averaged squared gradients
__group_960__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_961__new_weights (optional) - T2: New weights
__group_961__new_gradients (optional) - T3: New gradients
__group_961__new_moment_1 (optional) - T4: New averaged gradients
__group_961__new_moment_2 (optional) - T4: New averaged squared gradients
__group_961__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_962__new_weights (optional) - T2: New weights
__group_962__new_gradients (optional) - T3: New gradients
__group_962__new_moment_1 (optional) - T4: New averaged gradients
__group_962__new_moment_2 (optional) - T4: New averaged squared gradients
__group_962__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_963__new_weights (optional) - T2: New weights
__group_963__new_gradients (optional) - T3: New gradients
__group_963__new_moment_1 (optional) - T4: New averaged gradients
__group_963__new_moment_2 (optional) - T4: New averaged squared gradients
__group_963__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_964__new_weights (optional) - T2: New weights
__group_964__new_gradients (optional) - T3: New gradients
__group_964__new_moment_1 (optional) - T4: New averaged gradients
__group_964__new_moment_2 (optional) - T4: New averaged squared gradients
__group_964__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_965__new_weights (optional) - T2: New weights
__group_965__new_gradients (optional) - T3: New gradients
__group_965__new_moment_1 (optional) - T4: New averaged gradients
__group_965__new_moment_2 (optional) - T4: New averaged squared gradients
__group_965__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_966__new_weights (optional) - T2: New weights
__group_966__new_gradients (optional) - T3: New gradients
__group_966__new_moment_1 (optional) - T4: New averaged gradients
__group_966__new_moment_2 (optional) - T4: New averaged squared gradients
__group_966__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_967__new_weights (optional) - T2: New weights
__group_967__new_gradients (optional) - T3: New gradients
__group_967__new_moment_1 (optional) - T4: New averaged gradients
__group_967__new_moment_2 (optional) - T4: New averaged squared gradients
__group_967__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_968__new_weights (optional) - T2: New weights
__group_968__new_gradients (optional) - T3: New gradients
__group_968__new_moment_1 (optional) - T4: New averaged gradients
__group_968__new_moment_2 (optional) - T4: New averaged squared gradients
__group_968__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_969__new_weights (optional) - T2: New weights
__group_969__new_gradients (optional) - T3: New gradients
__group_969__new_moment_1 (optional) - T4: New averaged gradients
__group_969__new_moment_2 (optional) - T4: New averaged squared gradients
__group_969__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_970__new_weights (optional) - T2: New weights
__group_970__new_gradients (optional) - T3: New gradients
__group_970__new_moment_1 (optional) - T4: New averaged gradients
__group_970__new_moment_2 (optional) - T4: New averaged squared gradients
__group_970__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_971__new_weights (optional) - T2: New weights
__group_971__new_gradients (optional) - T3: New gradients
__group_971__new_moment_1 (optional) - T4: New averaged gradients
__group_971__new_moment_2 (optional) - T4: New averaged squared gradients
__group_971__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_972__new_weights (optional) - T2: New weights
__group_972__new_gradients (optional) - T3: New gradients
__group_972__new_moment_1 (optional) - T4: New averaged gradients
__group_972__new_moment_2 (optional) - T4: New averaged squared gradients
__group_972__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_973__new_weights (optional) - T2: New weights
__group_973__new_gradients (optional) - T3: New gradients
__group_973__new_moment_1 (optional) - T4: New averaged gradients
__group_973__new_moment_2 (optional) - T4: New averaged squared gradients
__group_973__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_974__new_weights (optional) - T2: New weights
__group_974__new_gradients (optional) - T3: New gradients
__group_974__new_moment_1 (optional) - T4: New averaged gradients
__group_974__new_moment_2 (optional) - T4: New averaged squared gradients
__group_974__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_975__new_weights (optional) - T2: New weights
__group_975__new_gradients (optional) - T3: New gradients
__group_975__new_moment_1 (optional) - T4: New averaged gradients
__group_975__new_moment_2 (optional) - T4: New averaged squared gradients
__group_975__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_976__new_weights (optional) - T2: New weights
__group_976__new_gradients (optional) - T3: New gradients
__group_976__new_moment_1 (optional) - T4: New averaged gradients
__group_976__new_moment_2 (optional) - T4: New averaged squared gradients
__group_976__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_977__new_weights (optional) - T2: New weights
__group_977__new_gradients (optional) - T3: New gradients
__group_977__new_moment_1 (optional) - T4: New averaged gradients
__group_977__new_moment_2 (optional) - T4: New averaged squared gradients
__group_977__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_978__new_weights (optional) - T2: New weights
__group_978__new_gradients (optional) - T3: New gradients
__group_978__new_moment_1 (optional) - T4: New averaged gradients
__group_978__new_moment_2 (optional) - T4: New averaged squared gradients
__group_978__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_979__new_weights (optional) - T2: New weights
__group_979__new_gradients (optional) - T3: New gradients
__group_979__new_moment_1 (optional) - T4: New averaged gradients
__group_979__new_moment_2 (optional) - T4: New averaged squared gradients
__group_979__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_980__new_weights (optional) - T2: New weights
__group_980__new_gradients (optional) - T3: New gradients
__group_980__new_moment_1 (optional) - T4: New averaged gradients
__group_980__new_moment_2 (optional) - T4: New averaged squared gradients
__group_980__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_981__new_weights (optional) - T2: New weights
__group_981__new_gradients (optional) - T3: New gradients
__group_981__new_moment_1 (optional) - T4: New averaged gradients
__group_981__new_moment_2 (optional) - T4: New averaged squared gradients
__group_981__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_982__new_weights (optional) - T2: New weights
__group_982__new_gradients (optional) - T3: New gradients
__group_982__new_moment_1 (optional) - T4: New averaged gradients
__group_982__new_moment_2 (optional) - T4: New averaged squared gradients
__group_982__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_983__new_weights (optional) - T2: New weights
__group_983__new_gradients (optional) - T3: New gradients
__group_983__new_moment_1 (optional) - T4: New averaged gradients
__group_983__new_moment_2 (optional) - T4: New averaged squared gradients
__group_983__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_984__new_weights (optional) - T2: New weights
__group_984__new_gradients (optional) - T3: New gradients
__group_984__new_moment_1 (optional) - T4: New averaged gradients
__group_984__new_moment_2 (optional) - T4: New averaged squared gradients
__group_984__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_985__new_weights (optional) - T2: New weights
__group_985__new_gradients (optional) - T3: New gradients
__group_985__new_moment_1 (optional) - T4: New averaged gradients
__group_985__new_moment_2 (optional) - T4: New averaged squared gradients
__group_985__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_986__new_weights (optional) - T2: New weights
__group_986__new_gradients (optional) - T3: New gradients
__group_986__new_moment_1 (optional) - T4: New averaged gradients
__group_986__new_moment_2 (optional) - T4: New averaged squared gradients
__group_986__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_987__new_weights (optional) - T2: New weights
__group_987__new_gradients (optional) - T3: New gradients
__group_987__new_moment_1 (optional) - T4: New averaged gradients
__group_987__new_moment_2 (optional) - T4: New averaged squared gradients
__group_987__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_988__new_weights (optional) - T2: New weights
__group_988__new_gradients (optional) - T3: New gradients
__group_988__new_moment_1 (optional) - T4: New averaged gradients
__group_988__new_moment_2 (optional) - T4: New averaged squared gradients
__group_988__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_989__new_weights (optional) - T2: New weights
__group_989__new_gradients (optional) - T3: New gradients
__group_989__new_moment_1 (optional) - T4: New averaged gradients
__group_989__new_moment_2 (optional) - T4: New averaged squared gradients
__group_989__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_990__new_weights (optional) - T2: New weights
__group_990__new_gradients (optional) - T3: New gradients
__group_990__new_moment_1 (optional) - T4: New averaged gradients
__group_990__new_moment_2 (optional) - T4: New averaged squared gradients
__group_990__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_991__new_weights (optional) - T2: New weights
__group_991__new_gradients (optional) - T3: New gradients
__group_991__new_moment_1 (optional) - T4: New averaged gradients
__group_991__new_moment_2 (optional) - T4: New averaged squared gradients
__group_991__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_992__new_weights (optional) - T2: New weights
__group_992__new_gradients (optional) - T3: New gradients
__group_992__new_moment_1 (optional) - T4: New averaged gradients
__group_992__new_moment_2 (optional) - T4: New averaged squared gradients
__group_992__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_993__new_weights (optional) - T2: New weights
__group_993__new_gradients (optional) - T3: New gradients
__group_993__new_moment_1 (optional) - T4: New averaged gradients
__group_993__new_moment_2 (optional) - T4: New averaged squared gradients
__group_993__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_994__new_weights (optional) - T2: New weights
__group_994__new_gradients (optional) - T3: New gradients
__group_994__new_moment_1 (optional) - T4: New averaged gradients
__group_994__new_moment_2 (optional) - T4: New averaged squared gradients
__group_994__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_995__new_weights (optional) - T2: New weights
__group_995__new_gradients (optional) - T3: New gradients
__group_995__new_moment_1 (optional) - T4: New averaged gradients
__group_995__new_moment_2 (optional) - T4: New averaged squared gradients
__group_995__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_996__new_weights (optional) - T2: New weights
__group_996__new_gradients (optional) - T3: New gradients
__group_996__new_moment_1 (optional) - T4: New averaged gradients
__group_996__new_moment_2 (optional) - T4: New averaged squared gradients
__group_996__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_997__new_weights (optional) - T2: New weights
__group_997__new_gradients (optional) - T3: New gradients
__group_997__new_moment_1 (optional) - T4: New averaged gradients
__group_997__new_moment_2 (optional) - T4: New averaged squared gradients
__group_997__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_998__new_weights (optional) - T2: New weights
__group_998__new_gradients (optional) - T3: New gradients
__group_998__new_moment_1 (optional) - T4: New averaged gradients
__group_998__new_moment_2 (optional) - T4: New averaged squared gradients
__group_998__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_999__new_weights (optional) - T2: New weights
__group_999__new_gradients (optional) - T3: New gradients
__group_999__new_moment_1 (optional) - T4: New averaged gradients
__group_999__new_moment_2 (optional) - T4: New averaged squared gradients
__group_999__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1000__new_weights (optional) - T2: New weights
__group_1000__new_gradients (optional) - T3: New gradients
__group_1000__new_moment_1 (optional) - T4: New averaged gradients
__group_1000__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1000__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1001__new_weights (optional) - T2: New weights
__group_1001__new_gradients (optional) - T3: New gradients
__group_1001__new_moment_1 (optional) - T4: New averaged gradients
__group_1001__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1001__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1002__new_weights (optional) - T2: New weights
__group_1002__new_gradients (optional) - T3: New gradients
__group_1002__new_moment_1 (optional) - T4: New averaged gradients
__group_1002__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1002__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1003__new_weights (optional) - T2: New weights
__group_1003__new_gradients (optional) - T3: New gradients
__group_1003__new_moment_1 (optional) - T4: New averaged gradients
__group_1003__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1003__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1004__new_weights (optional) - T2: New weights
__group_1004__new_gradients (optional) - T3: New gradients
__group_1004__new_moment_1 (optional) - T4: New averaged gradients
__group_1004__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1004__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1005__new_weights (optional) - T2: New weights
__group_1005__new_gradients (optional) - T3: New gradients
__group_1005__new_moment_1 (optional) - T4: New averaged gradients
__group_1005__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1005__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1006__new_weights (optional) - T2: New weights
__group_1006__new_gradients (optional) - T3: New gradients
__group_1006__new_moment_1 (optional) - T4: New averaged gradients
__group_1006__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1006__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1007__new_weights (optional) - T2: New weights
__group_1007__new_gradients (optional) - T3: New gradients
__group_1007__new_moment_1 (optional) - T4: New averaged gradients
__group_1007__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1007__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1008__new_weights (optional) - T2: New weights
__group_1008__new_gradients (optional) - T3: New gradients
__group_1008__new_moment_1 (optional) - T4: New averaged gradients
__group_1008__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1008__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1009__new_weights (optional) - T2: New weights
__group_1009__new_gradients (optional) - T3: New gradients
__group_1009__new_moment_1 (optional) - T4: New averaged gradients
__group_1009__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1009__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1010__new_weights (optional) - T2: New weights
__group_1010__new_gradients (optional) - T3: New gradients
__group_1010__new_moment_1 (optional) - T4: New averaged gradients
__group_1010__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1010__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1011__new_weights (optional) - T2: New weights
__group_1011__new_gradients (optional) - T3: New gradients
__group_1011__new_moment_1 (optional) - T4: New averaged gradients
__group_1011__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1011__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1012__new_weights (optional) - T2: New weights
__group_1012__new_gradients (optional) - T3: New gradients
__group_1012__new_moment_1 (optional) - T4: New averaged gradients
__group_1012__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1012__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1013__new_weights (optional) - T2: New weights
__group_1013__new_gradients (optional) - T3: New gradients
__group_1013__new_moment_1 (optional) - T4: New averaged gradients
__group_1013__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1013__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1014__new_weights (optional) - T2: New weights
__group_1014__new_gradients (optional) - T3: New gradients
__group_1014__new_moment_1 (optional) - T4: New averaged gradients
__group_1014__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1014__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1015__new_weights (optional) - T2: New weights
__group_1015__new_gradients (optional) - T3: New gradients
__group_1015__new_moment_1 (optional) - T4: New averaged gradients
__group_1015__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1015__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1016__new_weights (optional) - T2: New weights
__group_1016__new_gradients (optional) - T3: New gradients
__group_1016__new_moment_1 (optional) - T4: New averaged gradients
__group_1016__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1016__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1017__new_weights (optional) - T2: New weights
__group_1017__new_gradients (optional) - T3: New gradients
__group_1017__new_moment_1 (optional) - T4: New averaged gradients
__group_1017__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1017__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1018__new_weights (optional) - T2: New weights
__group_1018__new_gradients (optional) - T3: New gradients
__group_1018__new_moment_1 (optional) - T4: New averaged gradients
__group_1018__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1018__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1019__new_weights (optional) - T2: New weights
__group_1019__new_gradients (optional) - T3: New gradients
__group_1019__new_moment_1 (optional) - T4: New averaged gradients
__group_1019__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1019__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1020__new_weights (optional) - T2: New weights
__group_1020__new_gradients (optional) - T3: New gradients
__group_1020__new_moment_1 (optional) - T4: New averaged gradients
__group_1020__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1020__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1021__new_weights (optional) - T2: New weights
__group_1021__new_gradients (optional) - T3: New gradients
__group_1021__new_moment_1 (optional) - T4: New averaged gradients
__group_1021__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1021__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1022__new_weights (optional) - T2: New weights
__group_1022__new_gradients (optional) - T3: New gradients
__group_1022__new_moment_1 (optional) - T4: New averaged gradients
__group_1022__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1022__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
__group_1023__new_weights (optional) - T2: New weights
__group_1023__new_gradients (optional) - T3: New gradients
__group_1023__new_moment_1 (optional) - T4: New averaged gradients
__group_1023__new_moment_2 (optional) - T4: New averaged squared gradients
__group_1023__new_mixed_precision_weights (optional) - T_MIXED_PRECISION_FP: New FP16 or BF16 weights
OnnxComMicrosoftLayerNormalizationGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLayerNormalizationGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
LayerNormalizationGrad
Attributes
axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is
?.
Inputs
Y_grad (heterogeneous) - V: The gradient tensor from output.
X (heterogeneous) - T: Input data tensor from the forward path
scale (heterogeneous) - V: Scale tensor.
mean (heterogeneous) - U: mean of X.
inv_std_dev (heterogeneous) - U: inverse std deviation of X.
Outputs
X_grad (heterogeneous) - T: Gradient of the input.
scale_grad (heterogeneous) - V: Gradient of the scale.
bias_grad (heterogeneous) - V: Gradient of the bias.
OnnxComMicrosoftLayerNormalizationGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLayerNormalizationGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
LayerNormalizationGrad
Attributes
axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is
?.
Inputs
Y_grad (heterogeneous) - V: The gradient tensor from output.
X (heterogeneous) - T: Input data tensor from the forward path
scale (heterogeneous) - V: Scale tensor.
mean (heterogeneous) - U: mean of X.
inv_std_dev (heterogeneous) - U: inverse std deviation of X.
Outputs
X_grad (heterogeneous) - T: Gradient of the input.
scale_grad (heterogeneous) - V: Gradient of the scale.
bias_grad (heterogeneous) - V: Gradient of the bias.
OnnxComMicrosoftLogSoftmaxGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLogSoftmaxGrad(*args, **kwargs)#
Version
name: LogSoftmaxGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftLogSoftmaxGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLogSoftmaxGrad_1(*args, **kwargs)#
Version
name: LogSoftmaxGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftLogSoftmaxGrad_13#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLogSoftmaxGrad_13(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the dimension LogSoftmax will be performed on.Defaults to -1. Negative value means counting dimensions from the back. Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftLogSoftmaxGrad_13_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLogSoftmaxGrad_13_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the dimension LogSoftmax will be performed on.Defaults to -1. Negative value means counting dimensions from the back. Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftLongformerAttention#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLongformerAttention(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Longformer Self Attention with a local context and a global context. Tokens attend locally: Each token attends to its W previous tokens and W succeding tokens with W being the window length. A selected few tokens attend globally to all other tokens.
The attention mask is of shape (batch_size, sequence_length), where sequence_length is a multiple of 2W after padding. Mask value < 0 (like -10000.0) means the token is masked, 0 otherwise.
Global attention flags have value 1 for the tokens attend globally and 0 otherwise.
Attributes
num_heads (required): Number of attention heads Default value is
?.window (required): One sided attention windows length W, or half of total window length Default value is
?.
Inputs
input (heterogeneous) - T: 3D input tensor with shape (batch_size, sequence_length, hidden_size), hidden_size = num_heads * head_size
weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, 3 * hidden_size)
bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
mask (heterogeneous) - T: Attention mask with shape (batch_size, sequence_length)
global_weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, 3 * hidden_size)
global_bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
global (heterogeneous) - G: Global attention flags with shape (batch_size, sequence_length)
Outputs
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
OnnxComMicrosoftLongformerAttention_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftLongformerAttention_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Longformer Self Attention with a local context and a global context. Tokens attend locally: Each token attends to its W previous tokens and W succeding tokens with W being the window length. A selected few tokens attend globally to all other tokens.
The attention mask is of shape (batch_size, sequence_length), where sequence_length is a multiple of 2W after padding. Mask value < 0 (like -10000.0) means the token is masked, 0 otherwise.
Global attention flags have value 1 for the tokens attend globally and 0 otherwise.
Attributes
num_heads (required): Number of attention heads Default value is
?.window (required): One sided attention windows length W, or half of total window length Default value is
?.
Inputs
input (heterogeneous) - T: 3D input tensor with shape (batch_size, sequence_length, hidden_size), hidden_size = num_heads * head_size
weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, 3 * hidden_size)
bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
mask (heterogeneous) - T: Attention mask with shape (batch_size, sequence_length)
global_weight (heterogeneous) - T: 2D input tensor with shape (hidden_size, 3 * hidden_size)
global_bias (heterogeneous) - T: 1D input tensor with shape (3 * hidden_size)
global (heterogeneous) - G: Global attention flags with shape (batch_size, sequence_length)
Outputs
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
OnnxComMicrosoftMatMulInteger16#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMatMulInteger16(*args, **kwargs)#
Version
name: MatMulInteger16 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
- Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.
The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
Inputs
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
Outputs
Y (heterogeneous) - T3: Matrix multiply results from A * B
OnnxComMicrosoftMatMulInteger16_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMatMulInteger16_1(*args, **kwargs)#
Version
name: MatMulInteger16 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
- Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html.
The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
Inputs
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
Outputs
Y (heterogeneous) - T3: Matrix multiply results from A * B
OnnxComMicrosoftMatMulIntegerToFloat#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMatMulIntegerToFloat(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
Between 4 and 7 inputs.
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
a_scale (heterogeneous) - T3: Scale of quantized input ‘A’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘A’.
b_scale (heterogeneous) - T3: Scale of quantized input ‘B’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
a_zero_point (optional, heterogeneous) - T1: Zero point tensor for input ‘A’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘A’.
b_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
bias (optional, heterogeneous) - T3: 1D input tensor, whose dimension is same as B’s last dimension
Outputs
Y (heterogeneous) - T3: Matrix multiply results from A * B
OnnxComMicrosoftMatMulIntegerToFloat_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMatMulIntegerToFloat_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
Between 4 and 7 inputs.
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
a_scale (heterogeneous) - T3: Scale of quantized input ‘A’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘A’.
b_scale (heterogeneous) - T3: Scale of quantized input ‘B’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
a_zero_point (optional, heterogeneous) - T1: Zero point tensor for input ‘A’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘A’.
b_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
bias (optional, heterogeneous) - T3: 1D input tensor, whose dimension is same as B’s last dimension
Outputs
Y (heterogeneous) - T3: Matrix multiply results from A * B
OnnxComMicrosoftMaxpoolWithMask#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMaxpoolWithMask(*args, **kwargs)#
Version
name: MaxpoolWithMask (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
For internal use.
Attributes
auto_pad:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.storage_order:
Default value is
?.strides:
Default value is
?.Inputs
X (heterogeneous) - T:
M (heterogeneous) - tensor(int32): mask
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftMaxpoolWithMask_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMaxpoolWithMask_1(*args, **kwargs)#
Version
name: MaxpoolWithMask (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
For internal use.
Attributes
auto_pad:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.storage_order:
Default value is
?.strides:
Default value is
?.Inputs
X (heterogeneous) - T:
M (heterogeneous) - tensor(int32): mask
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftMegatronF#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMegatronF(*args, **kwargs)#
Version
name: MegatronF (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
input (heterogeneous) - T: The input data as Tensor.
Outputs
output (heterogeneous) - T: The output.
OnnxComMicrosoftMegatronF_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMegatronF_1(*args, **kwargs)#
Version
name: MegatronF (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
input (heterogeneous) - T: The input data as Tensor.
Outputs
output (heterogeneous) - T: The output.
OnnxComMicrosoftMegatronG#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMegatronG(*args, **kwargs)#
Version
name: MegatronG (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - data parallel group, 1 - horizontal parallel group Default value is
?.
Inputs
input (heterogeneous) - T: The input data as Tensor.
Outputs
output (heterogeneous) - T: The output.
OnnxComMicrosoftMegatronG_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMegatronG_1(*args, **kwargs)#
Version
name: MegatronG (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - data parallel group, 1 - horizontal parallel group Default value is
?.
Inputs
input (heterogeneous) - T: The input data as Tensor.
Outputs
output (heterogeneous) - T: The output.
OnnxComMicrosoftMixedPrecisionScale#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMixedPrecisionScale(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
MixedPrecisionScale
Attributes
fuse_outputs: If true, fuse all outputs into one continous buffer. Default value is
?.to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
Between 2 and 2147483647 inputs.
S (heterogeneous) - ScaleT: scale
X (variadic, heterogeneous) - SrcT: inputs
Outputs
Between 1 and 2147483647 outputs.
Y (variadic, heterogeneous) - DstT: output
OnnxComMicrosoftMixedPrecisionScale_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMixedPrecisionScale_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
MixedPrecisionScale
Attributes
fuse_outputs: If true, fuse all outputs into one continous buffer. Default value is
?.to (required): The data type to which the elements of the input tensor are cast. Strictly must be one of the types from DataType enum in TensorProto Default value is
?.
Inputs
Between 2 and 2147483647 inputs.
S (heterogeneous) - ScaleT: scale
X (variadic, heterogeneous) - SrcT: inputs
Outputs
Between 1 and 2147483647 outputs.
Y (variadic, heterogeneous) - DstT: output
OnnxComMicrosoftMulInteger#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMulInteger(*args, **kwargs)#
Version
name: MulInteger (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Performs element-wise binary quantized multiplication (with Numpy-style broadcasting support). “This operator supports multidirectional (i.e., Numpy-style) broadcasting” The output of this op is the int32 accumulated result of the mul operation
C (int32) = (A - A_zero_point) * (B - B_zero_point)
Inputs
Between 3 and 4 inputs.
A (heterogeneous) - T: First operand.
A_zero_point (optional, heterogeneous) - T: Input A zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
B (heterogeneous) - T: Second operand.
B_zero_point (optional, heterogeneous) - T: Input B zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
C (heterogeneous) - T1: Constrain output to 32 bit tensor
OnnxComMicrosoftMulInteger_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMulInteger_1(*args, **kwargs)#
Version
name: MulInteger (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Performs element-wise binary quantized multiplication (with Numpy-style broadcasting support). “This operator supports multidirectional (i.e., Numpy-style) broadcasting” The output of this op is the int32 accumulated result of the mul operation
C (int32) = (A - A_zero_point) * (B - B_zero_point)
Inputs
Between 3 and 4 inputs.
A (heterogeneous) - T: First operand.
A_zero_point (optional, heterogeneous) - T: Input A zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
B (heterogeneous) - T: Second operand.
B_zero_point (optional, heterogeneous) - T: Input B zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
C (heterogeneous) - T1: Constrain output to 32 bit tensor
OnnxComMicrosoftMurmurHash3#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMurmurHash3(*args, **kwargs)#
Version
name: MurmurHash3 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The underlying implementation is MurmurHash3_x86_32 generating low latency 32bits hash suitable for implementing lookup tables, Bloom filters, count min sketch or feature hashing.
Attributes
positive: If value is 1, output type is uint32_t, else int32_t. Default value is 1. Default value is
?.seed: Seed for the hashing algorithm, unsigned 32-bit integer, default to 0. Default value is
?.
Inputs
X (heterogeneous) - T1: An input tensor to hash.
Outputs
Y (heterogeneous) - T2: 32-bit hash value.
OnnxComMicrosoftMurmurHash3_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftMurmurHash3_1(*args, **kwargs)#
Version
name: MurmurHash3 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The underlying implementation is MurmurHash3_x86_32 generating low latency 32bits hash suitable for implementing lookup tables, Bloom filters, count min sketch or feature hashing.
Attributes
positive: If value is 1, output type is uint32_t, else int32_t. Default value is 1. Default value is
?.seed: Seed for the hashing algorithm, unsigned 32-bit integer, default to 0. Default value is
?.
Inputs
X (heterogeneous) - T1: An input tensor to hash.
Outputs
Y (heterogeneous) - T2: 32-bit hash value.
OnnxComMicrosoftNGramRepeatBlock#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNGramRepeatBlock(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Enforce no repetition of n-grams. Scores are set to -inf for tokens that form a repeated n-gram if added to the back of the input_ids.
Attributes
ngram_size (required): The NGram size. Default value is
?.
Inputs
input_ids (heterogeneous) - Tid: 2D input tensor with shape (batch_size, sequence_length)
scores (heterogeneous) - T: 2D input tensor with shape (batch_size, vocab_size)
Outputs
scores_out (heterogeneous) - T: 2D output tensor with shape (batch_size, vocab_size)
OnnxComMicrosoftNGramRepeatBlock_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNGramRepeatBlock_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Enforce no repetition of n-grams. Scores are set to -inf for tokens that form a repeated n-gram if added to the back of the input_ids.
Attributes
ngram_size (required): The NGram size. Default value is
?.
Inputs
input_ids (heterogeneous) - Tid: 2D input tensor with shape (batch_size, sequence_length)
scores (heterogeneous) - T: 2D input tensor with shape (batch_size, vocab_size)
Outputs
scores_out (heterogeneous) - T: 2D output tensor with shape (batch_size, vocab_size)
OnnxComMicrosoftNcclAllGather#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNcclAllGather(*args, **kwargs)#
Version
name: NcclAllGather (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - global parallel group, 1 - data parallel group, 2 - node local data parallel group, 3 - cross node data parallel group, 4 - horozontal parallel, 5 - model parallel. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be sent
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: gathered tensors
OnnxComMicrosoftNcclAllGather_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNcclAllGather_1(*args, **kwargs)#
Version
name: NcclAllGather (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - global parallel group, 1 - data parallel group, 2 - node local data parallel group, 3 - cross node data parallel group, 4 - horozontal parallel, 5 - model parallel. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be sent
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: gathered tensors
OnnxComMicrosoftNcclAllReduce#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNcclAllReduce(*args, **kwargs)#
Version
name: NcclAllReduce (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - global parallel group, 1 - data parallel group, 2 - node local data parallel group, 3 - cross node data parallel group, 4 - horozontal parallel, 5 - model parallel. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be reduced
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: reduced tensors
OnnxComMicrosoftNcclAllReduce_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNcclAllReduce_1(*args, **kwargs)#
Version
name: NcclAllReduce (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - global parallel group, 1 - data parallel group, 2 - node local data parallel group, 3 - cross node data parallel group, 4 - horozontal parallel, 5 - model parallel. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be reduced
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: reduced tensors
OnnxComMicrosoftNcclReduceScatter#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNcclReduceScatter(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - global parallel group, 1 - data parallel group, 2 - node local data parallel group, 3 - cross node data parallel group, 4 - horozontal parallel, 5 - model parallel. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be reduced and scattered
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: reduced tensors
OnnxComMicrosoftNcclReduceScatter_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNcclReduceScatter_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
group_type: 0 - global parallel group, 1 - data parallel group, 2 - node local data parallel group, 3 - cross node data parallel group, 4 - horozontal parallel, 5 - model parallel. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - T: tensors to be reduced and scattered
Outputs
Between 1 and 2147483647 outputs.
output (variadic, heterogeneous) - T: reduced tensors
OnnxComMicrosoftNchwcAveragePool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcAveragePool(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
auto_pad:
Default value is
?.ceil_mode:
Default value is
?.count_include_pad:
Default value is
?.dilations:
Default value is
?.kernel_shape (required):
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcAveragePool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcAveragePool_1(*args, **kwargs)#
Version
name: AveragePool (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
auto_pad:
Default value is
?.ceil_mode:
Default value is
?.count_include_pad:
Default value is
?.dilations:
Default value is
?.kernel_shape (required):
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcConv#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcConv(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
Between 2 and 4 inputs.
X (heterogeneous) - T:
W (heterogeneous) - T:
B (optional, heterogeneous) - T:
Sum (optional, heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcConv_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcConv_1(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
Between 2 and 4 inputs.
X (heterogeneous) - T:
W (heterogeneous) - T:
B (optional, heterogeneous) - T:
Sum (optional, heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcGlobalAveragePool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcGlobalAveragePool(*args, **kwargs)#
Version
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcGlobalAveragePool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcGlobalAveragePool_1(*args, **kwargs)#
Version
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcGlobalMaxPool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcGlobalMaxPool(*args, **kwargs)#
Version
name: GlobalMaxPool (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcGlobalMaxPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcGlobalMaxPool_1(*args, **kwargs)#
Version
name: GlobalMaxPool (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcMaxPool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcMaxPool(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
auto_pad:
Default value is
?.ceil_mode:
Default value is
?.dilations:
Default value is
?.kernel_shape (required):
Default value is
?.pads:
Default value is
?.storage_order:
Default value is
?.strides:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcMaxPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcMaxPool_1(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
auto_pad:
Default value is
?.ceil_mode:
Default value is
?.dilations:
Default value is
?.kernel_shape (required):
Default value is
?.pads:
Default value is
?.storage_order:
Default value is
?.strides:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcReorderInput#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcReorderInput(*args, **kwargs)#
Version
name: ReorderInput (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
channels_last:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcReorderInput_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcReorderInput_1(*args, **kwargs)#
Version
name: ReorderInput (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
channels_last:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcReorderOutput#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcReorderOutput(*args, **kwargs)#
Version
name: ReorderOutput (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
channels:
Default value is
?.channels_last:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcReorderOutput_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcReorderOutput_1(*args, **kwargs)#
Version
name: ReorderOutput (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
channels:
Default value is
?.channels_last:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcUpsample#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcUpsample(*args, **kwargs)#
Version
name: Upsample (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
coordinate_transformation_mode:
Default value is
?.mode:
Default value is
?.scales:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNchwcUpsample_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNchwcUpsample_1(*args, **kwargs)#
Version
name: Upsample (GitHub)
domain: com.microsoft.nchwc
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.nchwc.
Summary
For internal use.
Attributes
coordinate_transformation_mode:
Default value is
?.mode:
Default value is
?.scales:
Default value is
?.Inputs
X (heterogeneous) - T:
Outputs
Y (heterogeneous) - T:
OnnxComMicrosoftNegativeLogLikelihoodLossInternal#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNegativeLogLikelihoodLossInternal(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
NegativeLogLikelihoodLossInternal
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 2 and 4 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
OnnxComMicrosoftNegativeLogLikelihoodLossInternal2#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNegativeLogLikelihoodLossInternal2(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
NegativeLogLikelihoodLossInternal
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 2 and 4 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
OnnxComMicrosoftNegativeLogLikelihoodLossInternal2_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNegativeLogLikelihoodLossInternal2_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
NegativeLogLikelihoodLossInternal
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 2 and 4 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
OnnxComMicrosoftNegativeLogLikelihoodLossInternal_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNegativeLogLikelihoodLossInternal_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
NegativeLogLikelihoodLossInternal
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 2 and 4 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
OnnxComMicrosoftNhwcConv#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNhwcConv(*args, **kwargs)#
Version
name: NhwcConv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis. Default value is
?.group: number of groups input channels and output channels are divided into. Default value is
?.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W. Default value is
?.pads:
Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis. Default value is
?.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
OnnxComMicrosoftNhwcConv_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNhwcConv_1(*args, **kwargs)#
Version
name: NhwcConv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis. Default value is
?.group: number of groups input channels and output channels are divided into. Default value is
?.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W. Default value is
?.pads:
Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis. Default value is
?.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
OnnxComMicrosoftNhwcMaxPool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNhwcMaxPool(*args, **kwargs)#
Version
name: NhwcMaxPool (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.ceil_mode:
Default value is
?.dilations:
Default value is
?.kernel_shape (required):
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
x (heterogeneous) - T:
Outputs
y (heterogeneous) - T:
OnnxComMicrosoftNhwcMaxPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftNhwcMaxPool_1(*args, **kwargs)#
Version
name: NhwcMaxPool (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.ceil_mode:
Default value is
?.dilations:
Default value is
?.kernel_shape (required):
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
x (heterogeneous) - T:
Outputs
y (heterogeneous) - T:
OnnxComMicrosoftPad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPad(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Given data tensor, pads, mode, and value. Example: Insert 0 pads to the beginning of the second dimension. data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7], ]
pads = [0, 2, 0, 0] output = [
[ [0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7], ], ]
Attributes
mode: Three modes: constant`(default) - pads with a given constant value, `reflect - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis, edge - pads with the edge values of array Default value is
?.
Inputs
Between 2 and 3 inputs.
data (heterogeneous) - T: Input tensor.
pads (heterogeneous) - tensor(int64): Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. pads should be a 1D tensor of shape [2 * input_rank] or a 2D tensor of shape [1, 2 * input_rank]. pads format (1D example) should be as follow [x1_begin, x2_begin,…,x1_end, x2_end,…], where xi_begin is the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.
value (optional, heterogeneous) - T: (Optional) A scalar or rank 1 tensor containing a single value to be filled if the mode chosen is constant (by default it is 0.0).
Outputs
output (heterogeneous) - T: Tensor after padding.
OnnxComMicrosoftPad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPad_1(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Given data tensor, pads, mode, and value. Example: Insert 0 pads to the beginning of the second dimension. data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7], ]
pads = [0, 2, 0, 0] output = [
[ [0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7], ], ]
Attributes
mode: Three modes: constant`(default) - pads with a given constant value, `reflect - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis, edge - pads with the edge values of array Default value is
?.
Inputs
Between 2 and 3 inputs.
data (heterogeneous) - T: Input tensor.
pads (heterogeneous) - tensor(int64): Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. pads should be a 1D tensor of shape [2 * input_rank] or a 2D tensor of shape [1, 2 * input_rank]. pads format (1D example) should be as follow [x1_begin, x2_begin,…,x1_end, x2_end,…], where xi_begin is the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.
value (optional, heterogeneous) - T: (Optional) A scalar or rank 1 tensor containing a single value to be filled if the mode chosen is constant (by default it is 0.0).
Outputs
output (heterogeneous) - T: Tensor after padding.
OnnxComMicrosoftPassThrough#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPassThrough(*args, **kwargs)#
Version
name: PassThrough (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Barrier op with value pass through, outputs = inputs
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic) - T: input tensors
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T: output tensors
OnnxComMicrosoftPassThrough_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPassThrough_1(*args, **kwargs)#
Version
name: PassThrough (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Barrier op with value pass through, outputs = inputs
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic) - T: input tensors
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T: output tensors
OnnxComMicrosoftPythonOp#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPythonOp(*args, **kwargs)#
Version
name: PythonOp (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Wrapper of Pytorch’s autograd.Function implementation.
Attributes
inplace: Indicate if the output should reuse input memory. Default value is
?.input_convention (required): input_convention[i]==c means a non-tensor argument. input_convention[i]==d means a tensor. Default value is
?.input_float_scalar_positions:
Default value is
?.input_float_scalars: Python float arguments. Default value is
?.input_float_tuple_begins:
Default value is
?.input_float_tuple_positions:
Default value is
?.input_float_tuples:
Default value is
?.input_int_scalar_positions:
Default value is
?.input_int_scalars: Python int arguments. Default value is
?.input_int_tuple_begins:
Default value is
?.input_int_tuple_positions:
Default value is
?.input_int_tuples: Python int-tuple arguments. Default value is
?.input_pointer_scalar_positions:
Default value is
?.input_pointer_scalars:
Default value is
?.input_requires_grads (required): Flags to indicate whether the torch.autograd.apply’s inputs require gradients (including flags for both tensor and non-tensor inputs) Default value is
?.input_tensor_ranks (required): Input tensors’ ranks of autograd.Function.apply. Default value is
?.input_tensor_types (required): Input types of autograd.Function.apply. Default value is
?.name (required): Name of custom class. Default value is
?.output_tensor_ranks (required): Output tensors’ ranks of autograd.Function.apply. Default value is
?.output_tensor_requires_grads (required): Flags to indicate which output has gradient Default value is
?.output_tensor_types (required): Output types of autograd.Function.apply. Default value is
?.training_mode: Indicate if the model is exported in training_mode, by default, False. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic) - T: Module outputs to be returned to pytorch.
Outputs
Between 2 and 2147483647 outputs.
context (heterogeneous) - TInt64: Address of context created in this operator. It can be used in backward.
outputs (variadic) - T: Outputs returned from pytorch.
OnnxComMicrosoftPythonOpGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPythonOpGrad(*args, **kwargs)#
Version
name: PythonOpGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Wrapper of Pytorch’s autograd.Function’s backward implementaiton.
Attributes
inplace: Indicate if the output should reuse input memory. Todo(pengwa): do we really need it? Default value is
?.input_tensor_ranks: Input ranks of autograd.Function.backward (including only tensor inputs).This attribute is mostly used for input checks for better robustness. Default value is
?.input_tensor_requires_grads (required): Flags to indicate which inputs have gradients (including only tensor inputs).This attribute is mostly used for input checks for better robustness. Default value is
?.input_tensor_types: Input types of autograd.Function.backward (including only tensor inputs).This attribute is mostly used for input checks for better robustnes. Default value is
?.name (required): Name of custom class. Default value is
?.output_convention (required): A string inidicating autograd.Function.backward outputs’s type.value ‘c’ - non-tensor output; value ‘d’ - tensor output. Default value is
?.output_tensor_ranks: Output ranks of autograd.Function.backward outputs (including only tensor outputs). Default value is
?.output_tensor_requires_grads (required): Flags to indicate which outputs have gradients (including only tensor outputs). Default value is
?.output_tensor_types: Output types of autograd.Function.backward outputs (including only tensor outputs). Default value is
?.
Inputs
Between 2 and 2147483647 inputs.
context (heterogeneous) - TInt64: Address of context created in this operator. It should be generated by the corresponding forward.
inputs (variadic) - T: There are 2*N inputs: N gradient inputs (as inputs of autograd.Function.backward) + N forward run activations of autograd.Function.apply.The N forward run inputs are used as control dependency between PythonOpGrad and PythonOp
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T: Outputs returned from pytorch.
OnnxComMicrosoftPythonOpGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPythonOpGrad_1(*args, **kwargs)#
Version
name: PythonOpGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Wrapper of Pytorch’s autograd.Function’s backward implementaiton.
Attributes
inplace: Indicate if the output should reuse input memory. Todo(pengwa): do we really need it? Default value is
?.input_tensor_ranks: Input ranks of autograd.Function.backward (including only tensor inputs).This attribute is mostly used for input checks for better robustness. Default value is
?.input_tensor_requires_grads (required): Flags to indicate which inputs have gradients (including only tensor inputs).This attribute is mostly used for input checks for better robustness. Default value is
?.input_tensor_types: Input types of autograd.Function.backward (including only tensor inputs).This attribute is mostly used for input checks for better robustnes. Default value is
?.name (required): Name of custom class. Default value is
?.output_convention (required): A string inidicating autograd.Function.backward outputs’s type.value ‘c’ - non-tensor output; value ‘d’ - tensor output. Default value is
?.output_tensor_ranks: Output ranks of autograd.Function.backward outputs (including only tensor outputs). Default value is
?.output_tensor_requires_grads (required): Flags to indicate which outputs have gradients (including only tensor outputs). Default value is
?.output_tensor_types: Output types of autograd.Function.backward outputs (including only tensor outputs). Default value is
?.
Inputs
Between 2 and 2147483647 inputs.
context (heterogeneous) - TInt64: Address of context created in this operator. It should be generated by the corresponding forward.
inputs (variadic) - T: There are 2*N inputs: N gradient inputs (as inputs of autograd.Function.backward) + N forward run activations of autograd.Function.apply.The N forward run inputs are used as control dependency between PythonOpGrad and PythonOp
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T: Outputs returned from pytorch.
OnnxComMicrosoftPythonOp_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftPythonOp_1(*args, **kwargs)#
Version
name: PythonOp (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Wrapper of Pytorch’s autograd.Function implementation.
Attributes
inplace: Indicate if the output should reuse input memory. Default value is
?.input_convention (required): input_convention[i]==c means a non-tensor argument. input_convention[i]==d means a tensor. Default value is
?.input_float_scalar_positions:
Default value is
?.input_float_scalars: Python float arguments. Default value is
?.input_float_tuple_begins:
Default value is
?.input_float_tuple_positions:
Default value is
?.input_float_tuples:
Default value is
?.input_int_scalar_positions:
Default value is
?.input_int_scalars: Python int arguments. Default value is
?.input_int_tuple_begins:
Default value is
?.input_int_tuple_positions:
Default value is
?.input_int_tuples: Python int-tuple arguments. Default value is
?.input_pointer_scalar_positions:
Default value is
?.input_pointer_scalars:
Default value is
?.input_requires_grads (required): Flags to indicate whether the torch.autograd.apply’s inputs require gradients (including flags for both tensor and non-tensor inputs) Default value is
?.input_tensor_ranks (required): Input tensors’ ranks of autograd.Function.apply. Default value is
?.input_tensor_types (required): Input types of autograd.Function.apply. Default value is
?.name (required): Name of custom class. Default value is
?.output_tensor_ranks (required): Output tensors’ ranks of autograd.Function.apply. Default value is
?.output_tensor_requires_grads (required): Flags to indicate which output has gradient Default value is
?.output_tensor_types (required): Output types of autograd.Function.apply. Default value is
?.training_mode: Indicate if the model is exported in training_mode, by default, False. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic) - T: Module outputs to be returned to pytorch.
Outputs
Between 2 and 2147483647 outputs.
context (heterogeneous) - TInt64: Address of context created in this operator. It can be used in backward.
outputs (variadic) - T: Outputs returned from pytorch.
OnnxComMicrosoftQAttention#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQAttention(*args, **kwargs)#
Version
name: QAttention (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Quantization of Multi-Head Self Attention.
Attributes
num_heads (required): Number of attention heads Default value is
?.unidirectional: Whether every token can only attend to previous tokens. Default value is 0. Default value is
?.
Inputs
Between 5 and 9 inputs.
input (heterogeneous) - T1: 3D input tensor with shape (batch_size, sequence_length, input_hidden_size)
weight (heterogeneous) - T2: 2D input tensor with shape (input_hidden_size, 3 * hidden_size), hidden_size = num_heads * head_size
bias (heterogeneous) - T3: 1D input tensor with shape (3 * hidden_size)
input_scale (heterogeneous) - T3: scale of quantized input tensor. It’s a scalar, which means a per- tensor/layer quantization.
weight_scale (heterogeneous) - T3: scale of weight scale. It’s a scalar or a 1D tensor, which means a per-tensor/per-column quantization.Its size should be 3 * hidden_size if it is per-column quantization
mask_index (optional, heterogeneous) - T4: Attention mask index with shape (batch_size)
input_zero_point (optional, heterogeneous) - T1: zero point of quantized input tensor.It’s a scalar, which means a per-tensor/layer quantization.
weight_zero_point (optional, heterogeneous) - T2: zero point of quantized weight tensor. It’s a scalar or a 1D tensor, which means a per-tensor/per-column quantization.Its size should be 3 * hidden_size if it is per-column quantization
past (optional, heterogeneous) - T3: past state for key and value with shape (2, batch_size, num_heads, past_sequence_length, head_size).
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T3: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
present (optional, heterogeneous) - T3: present state for key and value with shape (2, batch_size, num_heads, past_sequence_length + sequence_length, head_size)
OnnxComMicrosoftQAttention_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQAttention_1(*args, **kwargs)#
Version
name: QAttention (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Quantization of Multi-Head Self Attention.
Attributes
num_heads (required): Number of attention heads Default value is
?.unidirectional: Whether every token can only attend to previous tokens. Default value is 0. Default value is
?.
Inputs
Between 5 and 9 inputs.
input (heterogeneous) - T1: 3D input tensor with shape (batch_size, sequence_length, input_hidden_size)
weight (heterogeneous) - T2: 2D input tensor with shape (input_hidden_size, 3 * hidden_size), hidden_size = num_heads * head_size
bias (heterogeneous) - T3: 1D input tensor with shape (3 * hidden_size)
input_scale (heterogeneous) - T3: scale of quantized input tensor. It’s a scalar, which means a per- tensor/layer quantization.
weight_scale (heterogeneous) - T3: scale of weight scale. It’s a scalar or a 1D tensor, which means a per-tensor/per-column quantization.Its size should be 3 * hidden_size if it is per-column quantization
mask_index (optional, heterogeneous) - T4: Attention mask index with shape (batch_size)
input_zero_point (optional, heterogeneous) - T1: zero point of quantized input tensor.It’s a scalar, which means a per-tensor/layer quantization.
weight_zero_point (optional, heterogeneous) - T2: zero point of quantized weight tensor. It’s a scalar or a 1D tensor, which means a per-tensor/per-column quantization.Its size should be 3 * hidden_size if it is per-column quantization
past (optional, heterogeneous) - T3: past state for key and value with shape (2, batch_size, num_heads, past_sequence_length, head_size).
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T3: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
present (optional, heterogeneous) - T3: present state for key and value with shape (2, batch_size, num_heads, past_sequence_length + sequence_length, head_size)
OnnxComMicrosoftQEmbedLayerNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQEmbedLayerNormalization(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QEmbedLayerNormalization is the quantized fusion of embedding layer in BERT model, with optional mask processing. The embedding layer takes input_ids (word IDs) and segment_ids (sentence IDs) to look up word_embedding, position_embedding, and segment_emedding; the embeddings are added then applied layer normalization using gamma and beta tensors. The input_ids and segment_ids remain int32. All embeddings, gamma, and beta tensors are converted to int8/uint8. The last input mask is optional. If mask is provided, mask index (that is position of first 0 in mask, or number of words will be calculated.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
?.
Inputs
input_ids (heterogeneous) - T1: 2D words IDs with shape (batch_size, sequence_length)
segment_ids (optional, heterogeneous) - T1: 2D segment IDs with shape (batch_size, sequence_length)
word_embedding_quant (heterogeneous) - T2: 2D with shape (,hidden_size)
position_embedding_quant (heterogeneous) - T2: 2D with shape (, hidden_size)
segment_embedding (optional, heterogeneous) - T2: 2D with shape (, hidden_size)
gamma_quant (heterogeneous) - T2: 1D gamma tensor for layer normalization with shape (hidden_size)
beta_quant (heterogeneous) - T2: 1D beta tensor for layer normalization with shape (hidden_size)
mask (optional, heterogeneous) - T1: Mask
word_embedding_scale (heterogeneous) - T: Scale for word embeddings
position_embedding_scale (heterogeneous) - T: Scale for position embeddings
segment_embedding_scale (optional, heterogeneous) - T: Scale for segment embeddings
gamma_scale (heterogeneous) - T: Scale for 1D gamma tensor
beta_scale (heterogeneous) - T: Scale for 1D beta tensor
word_embedding_zero_point (heterogeneous) - T2: Zero point for word embeddings
position_embedding_zero_point (heterogeneous) - T2: Zero point for position embeddings
segment_embedding_zero_point (optional, heterogeneous) - T2: Zero Point for segment embeddings
gamma_zero_point (heterogeneous) - T2: Zero Point for 1D gamma tensor
beta_zero_point (heterogeneous) - T2: Zero Point for 1D beta tensor
Outputs
layernorm_out (heterogeneous) - T: LayerNorm Output
mask_index_out (heterogeneous) - T1: Mask Index Output
OnnxComMicrosoftQEmbedLayerNormalization_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQEmbedLayerNormalization_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QEmbedLayerNormalization is the quantized fusion of embedding layer in BERT model, with optional mask processing. The embedding layer takes input_ids (word IDs) and segment_ids (sentence IDs) to look up word_embedding, position_embedding, and segment_emedding; the embeddings are added then applied layer normalization using gamma and beta tensors. The input_ids and segment_ids remain int32. All embeddings, gamma, and beta tensors are converted to int8/uint8. The last input mask is optional. If mask is provided, mask index (that is position of first 0 in mask, or number of words will be calculated.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
?.
Inputs
input_ids (heterogeneous) - T1: 2D words IDs with shape (batch_size, sequence_length)
segment_ids (optional, heterogeneous) - T1: 2D segment IDs with shape (batch_size, sequence_length)
word_embedding_quant (heterogeneous) - T2: 2D with shape (,hidden_size)
position_embedding_quant (heterogeneous) - T2: 2D with shape (, hidden_size)
segment_embedding (optional, heterogeneous) - T2: 2D with shape (, hidden_size)
gamma_quant (heterogeneous) - T2: 1D gamma tensor for layer normalization with shape (hidden_size)
beta_quant (heterogeneous) - T2: 1D beta tensor for layer normalization with shape (hidden_size)
mask (optional, heterogeneous) - T1: Mask
word_embedding_scale (heterogeneous) - T: Scale for word embeddings
position_embedding_scale (heterogeneous) - T: Scale for position embeddings
segment_embedding_scale (optional, heterogeneous) - T: Scale for segment embeddings
gamma_scale (heterogeneous) - T: Scale for 1D gamma tensor
beta_scale (heterogeneous) - T: Scale for 1D beta tensor
word_embedding_zero_point (heterogeneous) - T2: Zero point for word embeddings
position_embedding_zero_point (heterogeneous) - T2: Zero point for position embeddings
segment_embedding_zero_point (optional, heterogeneous) - T2: Zero Point for segment embeddings
gamma_zero_point (heterogeneous) - T2: Zero Point for 1D gamma tensor
beta_zero_point (heterogeneous) - T2: Zero Point for 1D beta tensor
Outputs
layernorm_out (heterogeneous) - T: LayerNorm Output
mask_index_out (heterogeneous) - T1: Mask Index Output
OnnxComMicrosoftQGemm#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQGemm(*args, **kwargs)#
Version
name: QGemm (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Quantized Gemm
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
?.transA: Whether A should be transposed Default value is
?.transB: Whether B should be transposed Default value is
?.
Inputs
Between 6 and 9 inputs.
A (heterogeneous) - TA: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
a_scale (heterogeneous) - T: Scale of quantized input ‘A’. It is a scalar,which means a per- tensor quantization.
a_zero_point (heterogeneous) - TA: Zero point tensor for input ‘A’. It is a scalar.
B (heterogeneous) - TB: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
b_scale (heterogeneous) - T: Scale of quantized input ‘B’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
b_zero_point (heterogeneous) - TB: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
C (optional, heterogeneous) - TC: Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N). Its type is int32_t and must be quantized with zero_point = 0 and scale = alpha / beta * a_scale * b_scale.
y_scale (optional, heterogeneous) - T: Scale of output ‘Y’. It is a scalar, which means a per-tensor quantization. It is optional. The output is full precision(float32) if it is not provided. Or the output is quantized.
y_zero_point (optional, heterogeneous) - TYZ: Zero point tensor for output ‘Y’. It is a scalar, which means a per- tensor quantization. It is optional. The output is full precision(float32) if it is not provided. Or the output is quantized.
Outputs
Y (heterogeneous) - TY: Output tensor of shape (M, N).
OnnxComMicrosoftQGemm_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQGemm_1(*args, **kwargs)#
Version
name: QGemm (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Quantized Gemm
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
?.transA: Whether A should be transposed Default value is
?.transB: Whether B should be transposed Default value is
?.
Inputs
Between 6 and 9 inputs.
A (heterogeneous) - TA: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
a_scale (heterogeneous) - T: Scale of quantized input ‘A’. It is a scalar,which means a per- tensor quantization.
a_zero_point (heterogeneous) - TA: Zero point tensor for input ‘A’. It is a scalar.
B (heterogeneous) - TB: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
b_scale (heterogeneous) - T: Scale of quantized input ‘B’. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
b_zero_point (heterogeneous) - TB: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per-tensor or per-column quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of columns of input ‘B’.
C (optional, heterogeneous) - TC: Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N). Its type is int32_t and must be quantized with zero_point = 0 and scale = alpha / beta * a_scale * b_scale.
y_scale (optional, heterogeneous) - T: Scale of output ‘Y’. It is a scalar, which means a per-tensor quantization. It is optional. The output is full precision(float32) if it is not provided. Or the output is quantized.
y_zero_point (optional, heterogeneous) - TYZ: Zero point tensor for output ‘Y’. It is a scalar, which means a per- tensor quantization. It is optional. The output is full precision(float32) if it is not provided. Or the output is quantized.
Outputs
Y (heterogeneous) - TY: Output tensor of shape (M, N).
OnnxComMicrosoftQLinearAdd#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearAdd(*args, **kwargs)#
Version
name: QLinearAdd (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Performs element-wise binary addition on 8 bit data types (with Numpy-style broadcasting support).
C = (A_scale * (A - A_zero_point) + B_scale * (B - B_zero_point))/C_scale + C_zero_point
Inputs
Between 7 and 8 inputs.
A (heterogeneous) - T: First operand.
A_scale (heterogeneous) - tensor(float): Input A’s scale. It’s a scalar, which means a per-tensor/layer quantization.
A_zero_point (optional, heterogeneous) - T: Input A zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
B (heterogeneous) - T: Second operand.
B_scale (heterogeneous) - tensor(float): Input B’s scale. It’s a scalar, which means a per-tensor/layer quantization.
B_zero_point (optional, heterogeneous) - T: Input B zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
C_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
C_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
OnnxComMicrosoftQLinearAdd_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearAdd_1(*args, **kwargs)#
Version
name: QLinearAdd (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Performs element-wise binary addition on 8 bit data types (with Numpy-style broadcasting support).
C = (A_scale * (A - A_zero_point) + B_scale * (B - B_zero_point))/C_scale + C_zero_point
Inputs
Between 7 and 8 inputs.
A (heterogeneous) - T: First operand.
A_scale (heterogeneous) - tensor(float): Input A’s scale. It’s a scalar, which means a per-tensor/layer quantization.
A_zero_point (optional, heterogeneous) - T: Input A zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
B (heterogeneous) - T: Second operand.
B_scale (heterogeneous) - tensor(float): Input B’s scale. It’s a scalar, which means a per-tensor/layer quantization.
B_zero_point (optional, heterogeneous) - T: Input B zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
C_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
C_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
OnnxComMicrosoftQLinearAveragePool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearAveragePool(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearAveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
` output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) `or` output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) `if ceil_mode is enabled` * pad_shape[i] is sum of pads along axis i `auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
` VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) `And pad shape will be following if SAME_UPPER or SAME_LOWER:` pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] `The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero).
Input and output scales and zero points are used to convert the output to a new quantization range. Output = Dequantize(Input) -> AveragePool on fp32 data -> Quantize(output)
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
?.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
?.channels_last: Works on NHWC layout or not? Default not. Default value is
?.count_include_pad: Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad. Default value is
?.kernel_shape (required): The size of the kernel along each axis. Default value is
?.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis. Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis. Default value is
?.
Inputs
Between 4 and 5 inputs.
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
x_scale (heterogeneous) - tensor(float): Input scale. It’s a scalar, which means a per-tensor/layer quantization.
x_zero_point (optional, heterogeneous) - T: Input zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
y_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
y_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
OnnxComMicrosoftQLinearAveragePool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearAveragePool_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearAveragePool consumes an input tensor X and applies average pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. average pooling consisting of computing the average on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
` output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) `or` output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) `if ceil_mode is enabled` * pad_shape[i] is sum of pads along axis i `auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
` VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i]) `And pad shape will be following if SAME_UPPER or SAME_LOWER:` pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i] `The output of each pooling window is divided by the number of elements (exclude pad when attribute count_include_pad is zero).
Input and output scales and zero points are used to convert the output to a new quantization range. Output = Dequantize(Input) -> AveragePool on fp32 data -> Quantize(output)
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
?.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
?.channels_last: Works on NHWC layout or not? Default not. Default value is
?.count_include_pad: Whether include pad pixels when calculating values for the edges. Default is 0, doesn’t count include pad. Default value is
?.kernel_shape (required): The size of the kernel along each axis. Default value is
?.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis. Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis. Default value is
?.
Inputs
Between 4 and 5 inputs.
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
x_scale (heterogeneous) - tensor(float): Input scale. It’s a scalar, which means a per-tensor/layer quantization.
x_zero_point (optional, heterogeneous) - T: Input zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
y_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
y_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
OnnxComMicrosoftQLinearConcat#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearConcat(*args, **kwargs)#
Version
name: QLinearConcat (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Concatenate a list of tensors into a single tensor.All input tensors must have the same shape, except for the dimension size of the axis to concatenate on.
Attributes
axis (required): Which axis to concat on Default value is
?.
Inputs
Between 3 and 2147483647 inputs.
Y_scale (heterogeneous) - TF: Y’s scale.
Y_zero_point (heterogeneous) - T8: Y’s zero point.
inputs (variadic) - TV: List of tensors/scale/zero_point for concatenation
Outputs
Y (heterogeneous) - T8: Concatenated tensor
OnnxComMicrosoftQLinearConcat_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearConcat_1(*args, **kwargs)#
Version
name: QLinearConcat (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Concatenate a list of tensors into a single tensor.All input tensors must have the same shape, except for the dimension size of the axis to concatenate on.
Attributes
axis (required): Which axis to concat on Default value is
?.
Inputs
Between 3 and 2147483647 inputs.
Y_scale (heterogeneous) - TF: Y’s scale.
Y_zero_point (heterogeneous) - T8: Y’s zero point.
inputs (variadic) - TV: List of tensors/scale/zero_point for concatenation
Outputs
Y (heterogeneous) - T8: Concatenated tensor
OnnxComMicrosoftQLinearConv#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearConv(*args, **kwargs)#
Version
name: QLinearConv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.channels_last:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
Between 8 and 9 inputs.
x (heterogeneous) - T1:
x_scale (heterogeneous) - tensor(float):
x_zero_point (heterogeneous) - T1:
w (heterogeneous) - T2:
w_scale (heterogeneous) - tensor(float):
w_zero_point (heterogeneous) - T2:
y_scale (heterogeneous) - tensor(float):
y_zero_point (heterogeneous) - T3:
B (optional, heterogeneous) - T4:
Outputs
y (heterogeneous) - T3:
OnnxComMicrosoftQLinearConv_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearConv_1(*args, **kwargs)#
Version
name: QLinearConv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
auto_pad:
Default value is
?.channels_last:
Default value is
?.dilations:
Default value is
?.group:
Default value is
?.kernel_shape:
Default value is
?.pads:
Default value is
?.strides:
Default value is
?.Inputs
Between 8 and 9 inputs.
x (heterogeneous) - T1:
x_scale (heterogeneous) - tensor(float):
x_zero_point (heterogeneous) - T1:
w (heterogeneous) - T2:
w_scale (heterogeneous) - tensor(float):
w_zero_point (heterogeneous) - T2:
y_scale (heterogeneous) - tensor(float):
y_zero_point (heterogeneous) - T3:
B (optional, heterogeneous) - T4:
Outputs
y (heterogeneous) - T3:
OnnxComMicrosoftQLinearGlobalAveragePool#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearGlobalAveragePool(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearGlobalAveragePool consumes an input tensor X and applies Average pooling across the values in the same channel. This is equivalent to AveragePool with kernel size equal to the spatial dimension of input tensor. Input is of type uint8_t or int8_t.
Attributes
channels_last:
Default value is
?.Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; According to channels_last, dimensions for image case are (N x C x H x W), or (N x H x W x C) where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), or (N x D1 X D2 … Dn x C) where N is the batch size.
x_scale (heterogeneous) - tensor(float): Scale of quantized input ‘X’. It must be a scalar.
x_zero_point (heterogeneous) - T: Zero point tensor for input ‘X’. It must be a scalar.
y_scale (heterogeneous) - tensor(float): Scale of quantized output ‘Y’. It must be a scalar.
y_zero_point (heterogeneous) - T: Zero point tensor for output ‘Y’. It must be a scalar.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. with the N and C value keep it value, while the otherdimensions are all 1.
OnnxComMicrosoftQLinearGlobalAveragePool_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearGlobalAveragePool_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearGlobalAveragePool consumes an input tensor X and applies Average pooling across the values in the same channel. This is equivalent to AveragePool with kernel size equal to the spatial dimension of input tensor. Input is of type uint8_t or int8_t.
Attributes
channels_last:
Default value is
?.Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; According to channels_last, dimensions for image case are (N x C x H x W), or (N x H x W x C) where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), or (N x D1 X D2 … Dn x C) where N is the batch size.
x_scale (heterogeneous) - tensor(float): Scale of quantized input ‘X’. It must be a scalar.
x_zero_point (heterogeneous) - T: Zero point tensor for input ‘X’. It must be a scalar.
y_scale (heterogeneous) - tensor(float): Scale of quantized output ‘Y’. It must be a scalar.
y_zero_point (heterogeneous) - T: Zero point tensor for output ‘Y’. It must be a scalar.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. with the N and C value keep it value, while the otherdimensions are all 1.
OnnxComMicrosoftQLinearLeakyRelu#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearLeakyRelu(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearLeakyRelu takes quantized input data (Tensor), an argument alpha, and quantize parameter for output, and produces one output data (Tensor<T>) where the function f(x) = quantize(alpha * dequantize(x)) for dequantize(x) < 0, f(x) = quantize(dequantize(x)) for dequantize(x) >= 0, is applied to the data tensor elementwise.
Attributes
alpha: Coefficient of leakage. Default value is
?.
Inputs
Between 4 and 5 inputs.
X (heterogeneous) - T: Input tensor
X_scale (heterogeneous) - tensor(float): Input X’s scale. It’s a scalar, which means a per-tensor/layer quantization.
X_zero_point (optional, heterogeneous) - T: Input X’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Y_scale (heterogeneous) - tensor(float): Output Y’s scale. It’s a scalar, which means a per-tensor/layer quantization.
Y_zero_point (optional, heterogeneous) - T: Output Y’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
Y (heterogeneous) - T: Output tensor
OnnxComMicrosoftQLinearLeakyRelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearLeakyRelu_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearLeakyRelu takes quantized input data (Tensor), an argument alpha, and quantize parameter for output, and produces one output data (Tensor<T>) where the function f(x) = quantize(alpha * dequantize(x)) for dequantize(x) < 0, f(x) = quantize(dequantize(x)) for dequantize(x) >= 0, is applied to the data tensor elementwise.
Attributes
alpha: Coefficient of leakage. Default value is
?.
Inputs
Between 4 and 5 inputs.
X (heterogeneous) - T: Input tensor
X_scale (heterogeneous) - tensor(float): Input X’s scale. It’s a scalar, which means a per-tensor/layer quantization.
X_zero_point (optional, heterogeneous) - T: Input X’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Y_scale (heterogeneous) - tensor(float): Output Y’s scale. It’s a scalar, which means a per-tensor/layer quantization.
Y_zero_point (optional, heterogeneous) - T: Output Y’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
Y (heterogeneous) - T: Output tensor
OnnxComMicrosoftQLinearMul#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearMul(*args, **kwargs)#
Version
name: QLinearMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Performs element-wise binary multiplication on 8 bit data types (with Numpy-style broadcasting support).
C = ((A - A_zero_point) * (B - B_zero_point)) * (A_scale * B_scale)/C_scale + C_zero_point
Inputs
Between 7 and 8 inputs.
A (heterogeneous) - T: First operand.
A_scale (heterogeneous) - tensor(float): Input A’s scale. It’s a scalar, which means a per-tensor/layer quantization.
A_zero_point (optional, heterogeneous) - T: Input A zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
B (heterogeneous) - T: Second operand.
B_scale (heterogeneous) - tensor(float): Input B’s scale. It’s a scalar, which means a per-tensor/layer quantization.
B_zero_point (optional, heterogeneous) - T: Input B zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
C_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
C_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
OnnxComMicrosoftQLinearMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearMul_1(*args, **kwargs)#
Version
name: QLinearMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Performs element-wise binary multiplication on 8 bit data types (with Numpy-style broadcasting support).
C = ((A - A_zero_point) * (B - B_zero_point)) * (A_scale * B_scale)/C_scale + C_zero_point
Inputs
Between 7 and 8 inputs.
A (heterogeneous) - T: First operand.
A_scale (heterogeneous) - tensor(float): Input A’s scale. It’s a scalar, which means a per-tensor/layer quantization.
A_zero_point (optional, heterogeneous) - T: Input A zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
B (heterogeneous) - T: Second operand.
B_scale (heterogeneous) - tensor(float): Input B’s scale. It’s a scalar, which means a per-tensor/layer quantization.
B_zero_point (optional, heterogeneous) - T: Input B zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
C_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
C_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
OnnxComMicrosoftQLinearReduceMean#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearReduceMean(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes the mean of the low-precision input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor have the reduced dimension pruned. The above behavior is similar to numpy, with the exception that numpy default keepdims to False instead of True. Input and Output scales and zero points are used to requantize the output in a new range. This helps to improve accuracy as after ReduceMean operation the range of the output is expected to decrease.
"Output = Dequantize(Input) -> ReduceMean on fp32 data -> Quantize(output)",
Attributes
axes (required): A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Default value is
?.keepdims (required): Keep the reduced dimension or not, default 1 mean keep reduced dimension. Default value is
?.
Inputs
Between 4 and 5 inputs.
data (heterogeneous) - T: An input tensor.
data_scale (heterogeneous) - tensor(float): Input scale. It’s a scalar, which means a per-tensor/layer quantization.
data_zero_point (optional, heterogeneous) - T: Input zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
reduced_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
reduced_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
OnnxComMicrosoftQLinearReduceMean_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearReduceMean_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes the mean of the low-precision input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor have the reduced dimension pruned. The above behavior is similar to numpy, with the exception that numpy default keepdims to False instead of True. Input and Output scales and zero points are used to requantize the output in a new range. This helps to improve accuracy as after ReduceMean operation the range of the output is expected to decrease.
"Output = Dequantize(Input) -> ReduceMean on fp32 data -> Quantize(output)",
Attributes
axes (required): A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Default value is
?.keepdims (required): Keep the reduced dimension or not, default 1 mean keep reduced dimension. Default value is
?.
Inputs
Between 4 and 5 inputs.
data (heterogeneous) - T: An input tensor.
data_scale (heterogeneous) - tensor(float): Input scale. It’s a scalar, which means a per-tensor/layer quantization.
data_zero_point (optional, heterogeneous) - T: Input zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
reduced_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
reduced_zero_point (optional, heterogeneous) - T: Output zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
OnnxComMicrosoftQLinearSigmoid#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearSigmoid(*args, **kwargs)#
Version
name: QLinearSigmoid (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearSigmoid takes quantized input data (Tensor), and quantize parameter for output, and produces one output data (Tensor<T>) where the function f(x) = quantize(Sigmoid(dequantize(x))), is applied to the data tensor elementwise. Wwhere the function Sigmoid(x) = 1 / (1 + exp(-x))
Inputs
Between 4 and 5 inputs.
X (heterogeneous) - T: Input tensor
X_scale (heterogeneous) - tensor(float): Input X’s scale. It’s a scalar, which means a per-tensor/layer quantization.
X_zero_point (optional, heterogeneous) - T: Input X’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Y_scale (heterogeneous) - tensor(float): Output Y’s scale. It’s a scalar, which means a per-tensor/layer quantization.
Y_zero_point (optional, heterogeneous) - T: Output Y’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
Y (heterogeneous) - T: Output tensor
OnnxComMicrosoftQLinearSigmoid_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQLinearSigmoid_1(*args, **kwargs)#
Version
name: QLinearSigmoid (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
QLinearSigmoid takes quantized input data (Tensor), and quantize parameter for output, and produces one output data (Tensor<T>) where the function f(x) = quantize(Sigmoid(dequantize(x))), is applied to the data tensor elementwise. Wwhere the function Sigmoid(x) = 1 / (1 + exp(-x))
Inputs
Between 4 and 5 inputs.
X (heterogeneous) - T: Input tensor
X_scale (heterogeneous) - tensor(float): Input X’s scale. It’s a scalar, which means a per-tensor/layer quantization.
X_zero_point (optional, heterogeneous) - T: Input X’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Y_scale (heterogeneous) - tensor(float): Output Y’s scale. It’s a scalar, which means a per-tensor/layer quantization.
Y_zero_point (optional, heterogeneous) - T: Output Y’s zero point. Default value is 0 if it’s not specified. It’s a scalar, which means a per-tensor/layer quantization.
Outputs
Y (heterogeneous) - T: Output tensor
OnnxComMicrosoftQuantizeLinear#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQuantizeLinear(*args, **kwargs)#
Version
name: QuantizeLinear (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The linear quantization operator. It consumes a full precision data, a scale, a zero point to compute the low precision / quantized tensor. The quantization formula is y = saturate ((x / y_scale) + y_zero_point).For saturation, it saturates to [0, 255] if it’s uint8, or [-128, 127] if it’s int8. For (x / y_scale), it’s rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. Scale and zero point must have same shape. They must be either scalar (per tensor) or 1-D tensor (per ‘axis’).
Attributes
axis: The axis along which same quantization parameters are applied. It’s optional.If it’s not specified, it means per-tensor quantization and input ‘x_scale’ and ‘x_zero_point’ must be scalars.If it’s specified, it means per ‘axis’ quantization and input ‘x_scale’ and ‘x_zero_point’ must be 1-D tensors. Default value is
?.
Inputs
x (heterogeneous) - T1: N-D full precision Input tensor to be quantized.
y_scale (heterogeneous) - T1: Scale for doing quantization to get ‘y’. It could be a scalar or a 1-D tensor,which means a per-tensor or per-axis quantization. If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
y_zero_point (heterogeneous) - T2: Zero point for doing quantization to get ‘y’. It could be a scalar or a 1-D tensor, which means a per-tensoror per-axis quantization. If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
Outputs
y (heterogeneous) - T2: N-D quantized output tensor. It has same shape as input ‘x’.
OnnxComMicrosoftQuantizeLinear_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftQuantizeLinear_1(*args, **kwargs)#
Version
name: QuantizeLinear (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The linear quantization operator. It consumes a full precision data, a scale, a zero point to compute the low precision / quantized tensor. The quantization formula is y = saturate ((x / y_scale) + y_zero_point).For saturation, it saturates to [0, 255] if it’s uint8, or [-128, 127] if it’s int8. For (x / y_scale), it’s rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. Scale and zero point must have same shape. They must be either scalar (per tensor) or 1-D tensor (per ‘axis’).
Attributes
axis: The axis along which same quantization parameters are applied. It’s optional.If it’s not specified, it means per-tensor quantization and input ‘x_scale’ and ‘x_zero_point’ must be scalars.If it’s specified, it means per ‘axis’ quantization and input ‘x_scale’ and ‘x_zero_point’ must be 1-D tensors. Default value is
?.
Inputs
x (heterogeneous) - T1: N-D full precision Input tensor to be quantized.
y_scale (heterogeneous) - T1: Scale for doing quantization to get ‘y’. It could be a scalar or a 1-D tensor,which means a per-tensor or per-axis quantization. If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
y_zero_point (heterogeneous) - T2: Zero point for doing quantization to get ‘y’. It could be a scalar or a 1-D tensor, which means a per-tensoror per-axis quantization. If it’s a 1-D tensor, its number of elements should be equal to the dimension value of ‘axis’ dimension of input ‘x’.
Outputs
y (heterogeneous) - T2: N-D quantized output tensor. It has same shape as input ‘x’.
OnnxComMicrosoftRange#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRange(*args, **kwargs)#
Version
name: Range (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Creates a sequence of numbers that begins at start and extends by increments of delta up to but not including limit.
Inputs
Between 2 and 3 inputs.
start (heterogeneous) - T: Tensor(scalar, or dims=[1]). First entry in the range.
limit (heterogeneous) - T: Tensor(scalar, or dims=[1]). Upper limit of sequence, exclusive.
delta (optional, heterogeneous) - T: Tensor(scalar, or dims=[1]). Number that increments start. Defaults to 1.
Outputs
Y (heterogeneous) - T: 1-D Tensor of the range.
OnnxComMicrosoftRange_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRange_1(*args, **kwargs)#
Version
name: Range (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Creates a sequence of numbers that begins at start and extends by increments of delta up to but not including limit.
Inputs
Between 2 and 3 inputs.
start (heterogeneous) - T: Tensor(scalar, or dims=[1]). First entry in the range.
limit (heterogeneous) - T: Tensor(scalar, or dims=[1]). Upper limit of sequence, exclusive.
delta (optional, heterogeneous) - T: Tensor(scalar, or dims=[1]). Number that increments start. Defaults to 1.
Outputs
Y (heterogeneous) - T: 1-D Tensor of the range.
OnnxComMicrosoftRecordEvent#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRecordEvent(*args, **kwargs)#
Version
name: RecordEvent (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Record an event.
Inputs
Between 2 and 2147483647 inputs.
EventIdentifier (heterogeneous) - TInt64: Event identifier to record.
InputData (variadic) - T: Input data.
Outputs
Between 0 and 2147483647 outputs.
OutputData (variadic) - T: Output data.
OnnxComMicrosoftRecordEvent_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRecordEvent_1(*args, **kwargs)#
Version
name: RecordEvent (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Record an event.
Inputs
Between 2 and 2147483647 inputs.
EventIdentifier (heterogeneous) - TInt64: Event identifier to record.
InputData (variadic) - T: Input data.
Outputs
Between 0 and 2147483647 outputs.
OutputData (variadic) - T: Output data.
OnnxComMicrosoftRecv#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRecv(*args, **kwargs)#
Version
name: Recv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Receive a tensor from the the specified source.
Attributes
element_types (required): Element types of the received tensors. Default value is
?.tag (required): The tag of the message carrying Data. Default value is
?.
Inputs
InputSignal (heterogeneous) - TBool: Input control signal. It must be a scalar.
Remote (heterogeneous) - TInt64: Remote src rank. It must be a scalar.
Outputs
Between 2 and 2147483647 outputs.
OutputSignal (heterogeneous) - TBool: Output control signal. It must be a scalar.
Data (variadic) - V: The Received tensors.
OnnxComMicrosoftRecv_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRecv_1(*args, **kwargs)#
Version
name: Recv (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Receive a tensor from the the specified source.
Attributes
element_types (required): Element types of the received tensors. Default value is
?.tag (required): The tag of the message carrying Data. Default value is
?.
Inputs
InputSignal (heterogeneous) - TBool: Input control signal. It must be a scalar.
Remote (heterogeneous) - TInt64: Remote src rank. It must be a scalar.
Outputs
Between 2 and 2147483647 outputs.
OutputSignal (heterogeneous) - TBool: Output control signal. It must be a scalar.
Data (variadic) - V: The Received tensors.
OnnxComMicrosoftReduceAllL2#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReduceAllL2(*args, **kwargs)#
Version
name: ReduceAllL2 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Multi-tensor version of ReduceL2.
Inputs
Between 1 and 2147483647 inputs.
X (variadic, heterogeneous) - TIn: inputs
Outputs
Y (heterogeneous) - TOut: output
OnnxComMicrosoftReduceAllL2_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReduceAllL2_1(*args, **kwargs)#
Version
name: ReduceAllL2 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Multi-tensor version of ReduceL2.
Inputs
Between 1 and 2147483647 inputs.
X (variadic, heterogeneous) - TIn: inputs
Outputs
Y (heterogeneous) - TOut: output
OnnxComMicrosoftReduceSumInteger#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReduceSumInteger(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes the sum of the low-precision input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor have the reduced dimension pruned. The above behavior is similar to numpy, with the exception that numpy default keepdims to False instead of True.
Attributes
axes (required): A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Default value is
?.keepdims (required): Keep the reduced dimension or not, default 1 mean keep reduced dimension. Default value is
?.
Inputs
data (heterogeneous) - T1: An input tensor.
Outputs
reduced (heterogeneous) - T2: Reduced output tensor.
OnnxComMicrosoftReduceSumInteger_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReduceSumInteger_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Computes the sum of the low-precision input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equal 1. If keepdims equal 0, then the resulting tensor have the reduced dimension pruned. The above behavior is similar to numpy, with the exception that numpy default keepdims to False instead of True.
Attributes
axes (required): A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Default value is
?.keepdims (required): Keep the reduced dimension or not, default 1 mean keep reduced dimension. Default value is
?.
Inputs
data (heterogeneous) - T1: An input tensor.
Outputs
reduced (heterogeneous) - T2: Reduced output tensor.
OnnxComMicrosoftReduceSumTraining#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReduceSumTraining(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
ReduceSumTraining
Attributes
keepdims: Keep the reduced dimension or not, default 1 mean keep reduced dimension. Default value is
?.noop_with_empty_axes: Perform reduction or not when axes is empty, default false mean perform reduction.when axes is empty and this attribute is set to true, input tensor will not be reduced,thus output tensor would be equivalent to input tensor. Default value is
?.
Inputs
data (heterogeneous) - T: An input tensor.
axes (heterogeneous) - tensor(int64): A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
OnnxComMicrosoftReduceSumTraining_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReduceSumTraining_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
ReduceSumTraining
Attributes
keepdims: Keep the reduced dimension or not, default 1 mean keep reduced dimension. Default value is
?.noop_with_empty_axes: Perform reduction or not when axes is empty, default false mean perform reduction.when axes is empty and this attribute is set to true, input tensor will not be reduced,thus output tensor would be equivalent to input tensor. Default value is
?.
Inputs
data (heterogeneous) - T: An input tensor.
axes (heterogeneous) - tensor(int64): A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
OnnxComMicrosoftReluGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReluGrad(*args, **kwargs)#
Version
name: ReluGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftReluGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftReluGrad_1(*args, **kwargs)#
Version
name: ReluGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
dY (heterogeneous) - T: Gradient of output Y
X (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftRfft#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRfft(*args, **kwargs)#
Version
name: Rfft (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
normalized:
Default value is
?.onesided:
Default value is
?.signal_ndim:
Default value is
?.Inputs
X (heterogeneous) - T: input tensor
Outputs
Y (heterogeneous) - T: output tensor
OnnxComMicrosoftRfft_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftRfft_1(*args, **kwargs)#
Version
name: Rfft (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
normalized:
Default value is
?.onesided:
Default value is
?.signal_ndim:
Default value is
?.Inputs
X (heterogeneous) - T: input tensor
Outputs
Y (heterogeneous) - T: output tensor
OnnxComMicrosoftSGDOptimizer#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSGDOptimizer(*args, **kwargs)#
Version
name: SGDOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
ETA (heterogeneous) - L: Learning Rate
W (heterogeneous) - T: Original weight(s)
G (heterogeneous) - T: Gradient of Weight(s)
Outputs
Between 0 and 2 outputs.
NW (optional, heterogeneous) - T: Updated weight(s)
NG (optional, heterogeneous) - T: Updated gradients(s)
OnnxComMicrosoftSGDOptimizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSGDOptimizer_1(*args, **kwargs)#
Version
name: SGDOptimizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
ETA (heterogeneous) - L: Learning Rate
W (heterogeneous) - T: Original weight(s)
G (heterogeneous) - T: Gradient of Weight(s)
Outputs
Between 0 and 2 outputs.
NW (optional, heterogeneous) - T: Updated weight(s)
NG (optional, heterogeneous) - T: Updated gradients(s)
OnnxComMicrosoftSampleOp#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSampleOp(*args, **kwargs)#
Version
name: SampleOp (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Sample echo operator.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - T: output
OnnxComMicrosoftSampleOp_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSampleOp_1(*args, **kwargs)#
Version
name: SampleOp (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Sample echo operator.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - T: output
OnnxComMicrosoftScale#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftScale(*args, **kwargs)#
Version
name: Scale (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Scale
Attributes
scale_down: If true, the output tensor is input tensor devided by scale, otherwise, it’s input tensor multiplied by scale. The default value is false. Default value is
?.
Inputs
input (heterogeneous) - T: Input tensor.
scale (heterogeneous) - ScaleT: Scale scalar tensor.
Outputs
output (heterogeneous) - T: The scaled output tensor.
OnnxComMicrosoftScale_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftScale_1(*args, **kwargs)#
Version
name: Scale (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Scale
Attributes
scale_down: If true, the output tensor is input tensor devided by scale, otherwise, it’s input tensor multiplied by scale. The default value is false. Default value is
?.
Inputs
input (heterogeneous) - T: Input tensor.
scale (heterogeneous) - ScaleT: Scale scalar tensor.
Outputs
output (heterogeneous) - T: The scaled output tensor.
OnnxComMicrosoftSend#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSend(*args, **kwargs)#
Version
name: Send (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Send data tensor to the specified destination.
Attributes
element_types (required): Element types of the sent tensors. Default value is
?.tag (required): The tag of the message carrying Data. Default value is
?.
Inputs
Between 3 and 2147483647 inputs.
InputSignal (heterogeneous) - TBool: Input control signal. It must be a scalar.
Remote (heterogeneous) - TInt64: Remote dst rank. It must be a scalar.
Data (variadic) - V: Tensors to send.
Outputs
OutputSignal (heterogeneous) - TBool: Output control signal. It must be a scalar.
OnnxComMicrosoftSend_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSend_1(*args, **kwargs)#
Version
name: Send (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Send data tensor to the specified destination.
Attributes
element_types (required): Element types of the sent tensors. Default value is
?.tag (required): The tag of the message carrying Data. Default value is
?.
Inputs
Between 3 and 2147483647 inputs.
InputSignal (heterogeneous) - TBool: Input control signal. It must be a scalar.
Remote (heterogeneous) - TInt64: Remote dst rank. It must be a scalar.
Data (variadic) - V: Tensors to send.
Outputs
OutputSignal (heterogeneous) - TBool: Output control signal. It must be a scalar.
OnnxComMicrosoftSigmoidGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSigmoidGrad(*args, **kwargs)#
Version
name: SigmoidGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SigmoidGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
Y (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftSigmoidGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSigmoidGrad_1(*args, **kwargs)#
Version
name: SigmoidGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SigmoidGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
Y (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftSimplifiedLayerNormalizationGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSimplifiedLayerNormalizationGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SimplifiedLayerNormalizationGrad
Attributes
axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is
?.
Inputs
Y_grad (heterogeneous) - V: The gradient tensor from output.
X (heterogeneous) - T: Input data tensor from the forward path
scale (heterogeneous) - V: Scale tensor.
inv_std_var (heterogeneous) - U: inverse std variance of X.
Outputs
X_grad (heterogeneous) - T: Gradient of the input.
scale_grad (heterogeneous) - V: Gradient of the scale.
OnnxComMicrosoftSimplifiedLayerNormalizationGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSimplifiedLayerNormalizationGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SimplifiedLayerNormalizationGrad
Attributes
axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is
?.
Inputs
Y_grad (heterogeneous) - V: The gradient tensor from output.
X (heterogeneous) - T: Input data tensor from the forward path
scale (heterogeneous) - V: Scale tensor.
inv_std_var (heterogeneous) - U: inverse std variance of X.
Outputs
X_grad (heterogeneous) - T: Gradient of the input.
scale_grad (heterogeneous) - V: Gradient of the scale.
OnnxComMicrosoftSkipLayerNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSkipLayerNormalization(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Skip and Layer Normalization Fusion
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
?.
Inputs
Between 3 and 5 inputs.
input (heterogeneous) - T: 3D input tensor with shape (batch_size, sequence_length, hidden_size)
skip (heterogeneous) - T: 3D skip tensor with shape (batch_size, sequence_length, hidden_size)
gamma (heterogeneous) - T: 1D input tensor with shape (hidden_size)
beta (optional, heterogeneous) - T: 1D skip tensor with shape (hidden_size
bias (optional, heterogeneous) - T: 1D bias tensor with shape (hidden_size
Outputs
Between 1 and 3 outputs.
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
mean (optional, heterogeneous) - U: Saved mean used during training to speed up gradient computation
inv_std_var (optional, heterogeneous) - U: Saved inverse standard variance used during training to speed up gradient computation.
OnnxComMicrosoftSkipLayerNormalization_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSkipLayerNormalization_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Skip and Layer Normalization Fusion
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
?.
Inputs
Between 3 and 5 inputs.
input (heterogeneous) - T: 3D input tensor with shape (batch_size, sequence_length, hidden_size)
skip (heterogeneous) - T: 3D skip tensor with shape (batch_size, sequence_length, hidden_size)
gamma (heterogeneous) - T: 1D input tensor with shape (hidden_size)
beta (optional, heterogeneous) - T: 1D skip tensor with shape (hidden_size
bias (optional, heterogeneous) - T: 1D bias tensor with shape (hidden_size
Outputs
Between 1 and 3 outputs.
output (heterogeneous) - T: 3D output tensor with shape (batch_size, sequence_length, hidden_size)
mean (optional, heterogeneous) - U: Saved mean used during training to speed up gradient computation
inv_std_var (optional, heterogeneous) - U: Saved inverse standard variance used during training to speed up gradient computation.
OnnxComMicrosoftSliceGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSliceGrad(*args, **kwargs)#
Version
name: SliceGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
Between 4 and 6 inputs.
dY (heterogeneous) - T: Gradient of output
shape (heterogeneous) - I: Shape of the Slice input X.
starts (heterogeneous) - Tind: Tensor of starting indices of corresponding axis in axes
ends (heterogeneous) - Tind: Tensor of starting indices of corresponding axis in ‘axes’
axes (optional, heterogeneous) - Tind: Tensor of axes that starts and ends apply to
steps (optional, heterogeneous) - Tind: Tensor of slice step of corresponding axis in axes
Outputs
dX (heterogeneous) - T: Gradient of input
OnnxComMicrosoftSliceGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSliceGrad_1(*args, **kwargs)#
Version
name: SliceGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Inputs
Between 4 and 6 inputs.
dY (heterogeneous) - T: Gradient of output
shape (heterogeneous) - I: Shape of the Slice input X.
starts (heterogeneous) - Tind: Tensor of starting indices of corresponding axis in axes
ends (heterogeneous) - Tind: Tensor of starting indices of corresponding axis in ‘axes’
axes (optional, heterogeneous) - Tind: Tensor of axes that starts and ends apply to
steps (optional, heterogeneous) - Tind: Tensor of slice step of corresponding axis in axes
Outputs
dX (heterogeneous) - T: Gradient of input
OnnxComMicrosoftSnpe#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSnpe(*args, **kwargs)#
Version
name: Snpe (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Onnx node for SNPE.
Attributes
DLC (required): payload of the SNPE DLC file. Default value is
?.notes: (Optional) Some notes for the model Default value is
?.snpe_version: (Optional) SNPE version used to convert the model. Default value is
?.target_device: (Optional) Target device like CPU, DSP, etc. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for SNPE DLC input
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs, list of tensors for DLC output
OnnxComMicrosoftSnpe_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSnpe_1(*args, **kwargs)#
Version
name: Snpe (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Onnx node for SNPE.
Attributes
DLC (required): payload of the SNPE DLC file. Default value is
?.notes: (Optional) Some notes for the model Default value is
?.snpe_version: (Optional) SNPE version used to convert the model. Default value is
?.target_device: (Optional) Target device like CPU, DSP, etc. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for SNPE DLC input
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs, list of tensors for DLC output
OnnxComMicrosoftSoftmaxCrossEntropy#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropy(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropy
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
logits (heterogeneous) - T: Unscaled log probabilities, N-D input of shape (-1, num_classes).
label (heterogeneous) - T: The onehot label is N-D input with the same shape as logits.
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: loss.
log_prob (optional, heterogeneous) - T: logsoftmax(logits)
OnnxComMicrosoftSoftmaxCrossEntropyGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyGrad
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
dY (heterogeneous) - T: gradient of Y
log_prob (heterogeneous) - T: logsoftmax(logits), N-D input of shape (-1, num_classes).
label (heterogeneous) - T: The onehot label is N-D input with the same shape as logits.
Outputs
d_logits (heterogeneous) - T: gradient of logits
OnnxComMicrosoftSoftmaxCrossEntropyGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyGrad
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
dY (heterogeneous) - T: gradient of Y
log_prob (heterogeneous) - T: logsoftmax(logits), N-D input of shape (-1, num_classes).
label (heterogeneous) - T: The onehot label is N-D input with the same shape as logits.
Outputs
d_logits (heterogeneous) - T: gradient of logits
OnnxComMicrosoftSoftmaxCrossEntropyLossGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyLossGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyLossGrad
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. Default value is
?.reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 3 and 4 inputs.
dY (heterogeneous) - T: gradient of Y
log_prob (heterogeneous) - T: logsoftmax(logits), (N+1)-D input of shape (batch_size).
label (heterogeneous) - Tind: label is N-D input whose shape should match that of logits. It is a tensor of nonnegative integers, where each element is the nonnegative integer label for the element of the batch.
weight (optional, heterogeneous) - T: weight for each sample. The shape is 1-D tensor.
Outputs
d_logits (heterogeneous) - T: gradient of logits
OnnxComMicrosoftSoftmaxCrossEntropyLossGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyLossGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyLossGrad
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. Default value is
?.reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 3 and 4 inputs.
dY (heterogeneous) - T: gradient of Y
log_prob (heterogeneous) - T: logsoftmax(logits), (N+1)-D input of shape (batch_size).
label (heterogeneous) - Tind: label is N-D input whose shape should match that of logits. It is a tensor of nonnegative integers, where each element is the nonnegative integer label for the element of the batch.
weight (optional, heterogeneous) - T: weight for each sample. The shape is 1-D tensor.
Outputs
d_logits (heterogeneous) - T: gradient of logits
OnnxComMicrosoftSoftmaxCrossEntropyLossInternal#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyLossInternal(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyLossInternal
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 2 and 4 inputs.
scores (heterogeneous) - T: The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , …, Dk], where K is the number of dimensions.
labels (heterogeneous) - Tind: The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, …, Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
weights (optional, heterogeneous) - T: A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
output (heterogeneous) - T: Weighted loss float Tensor. If reduction is ‘none’, this has the shape of [batch_size], or [batch_size, D1, D2, …, Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
log_prob (heterogeneous) - T: Log probability tensor. If the output of softmax is prob, its value is log(prob).
OnnxComMicrosoftSoftmaxCrossEntropyLossInternalGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyLossInternalGrad(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyLossInternalGrad
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 3 and 5 inputs.
dY (heterogeneous) - T: gradient of Y
log_prob (heterogeneous) - T: logsoftmax(logits), (N+1)-D input of shape (batch_size).
label (heterogeneous) - Tind: label is N-D input whose shape should match that of logits. It is a tensor of nonnegative integers, where each element is the nonnegative integer label for the element of the batch.
weight (optional, heterogeneous) - T: weight for each sample. The shape is 1-D tensor.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
d_logits (heterogeneous) - T: gradient of logits
OnnxComMicrosoftSoftmaxCrossEntropyLossInternalGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyLossInternalGrad_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyLossInternalGrad
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 3 and 5 inputs.
dY (heterogeneous) - T: gradient of Y
log_prob (heterogeneous) - T: logsoftmax(logits), (N+1)-D input of shape (batch_size).
label (heterogeneous) - Tind: label is N-D input whose shape should match that of logits. It is a tensor of nonnegative integers, where each element is the nonnegative integer label for the element of the batch.
weight (optional, heterogeneous) - T: weight for each sample. The shape is 1-D tensor.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
d_logits (heterogeneous) - T: gradient of logits
OnnxComMicrosoftSoftmaxCrossEntropyLossInternal_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropyLossInternal_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropyLossInternal
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
Between 2 and 4 inputs.
scores (heterogeneous) - T: The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , …, Dk], where K is the number of dimensions.
labels (heterogeneous) - Tind: The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, …, Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
weights (optional, heterogeneous) - T: A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
ignore_index (optional, heterogeneous) - I: Scalar tensor to specify a target value that is ignored and does not contribute to the input gradient.
Outputs
output (heterogeneous) - T: Weighted loss float Tensor. If reduction is ‘none’, this has the shape of [batch_size], or [batch_size, D1, D2, …, Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
log_prob (heterogeneous) - T: Log probability tensor. If the output of softmax is prob, its value is log(prob).
OnnxComMicrosoftSoftmaxCrossEntropy_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxCrossEntropy_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SoftmaxCrossEntropy
Attributes
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: the output is the loss for each sample in the batch.’sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the batch_size. Default value is
?.
Inputs
logits (heterogeneous) - T: Unscaled log probabilities, N-D input of shape (-1, num_classes).
label (heterogeneous) - T: The onehot label is N-D input with the same shape as logits.
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: loss.
log_prob (optional, heterogeneous) - T: logsoftmax(logits)
OnnxComMicrosoftSoftmaxGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxGrad(*args, **kwargs)#
Version
name: SoftmaxGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
Y (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftSoftmaxGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxGrad_1(*args, **kwargs)#
Version
name: SoftmaxGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
Y (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftSoftmaxGrad_13#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxGrad_13(*args, **kwargs)#
Version
name: SoftmaxGrad_13 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the dimension Softmax will be performed on.Defaults to -1. Negative value means counting dimensions from the back. Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
Y (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftSoftmaxGrad_13_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSoftmaxGrad_13_1(*args, **kwargs)#
Version
name: SoftmaxGrad_13 (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
axis: Describes the dimension Softmax will be performed on.Defaults to -1. Negative value means counting dimensions from the back. Default value is
?.
Inputs
dY (heterogeneous) - T: Gradient of output Y
Y (heterogeneous) - T: Input tensor
Outputs
dX (heterogeneous) - T: Gradient of input X
OnnxComMicrosoftSparseToDenseMatMul#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSparseToDenseMatMul(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
alpha: Scalar multiplier for the product of the input tensors. Default value is
?.transA: Whether A should be transposed on the last two dimensions before doing multiplication Default value is
?.transB: Whether B should be transposed on the last two dimensions before doing multiplication Default value is
?.
Inputs
A (heterogeneous) - T: 2-dimensional sparse matrix A. Either COO or CSR format
B (heterogeneous) - T1: N-dimensional dense matrix B
Outputs
Y (heterogeneous) - T1: Matrix multiply results
OnnxComMicrosoftSparseToDenseMatMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSparseToDenseMatMul_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Attributes
alpha: Scalar multiplier for the product of the input tensors. Default value is
?.transA: Whether A should be transposed on the last two dimensions before doing multiplication Default value is
?.transB: Whether B should be transposed on the last two dimensions before doing multiplication Default value is
?.
Inputs
A (heterogeneous) - T: 2-dimensional sparse matrix A. Either COO or CSR format
B (heterogeneous) - T1: N-dimensional dense matrix B
Outputs
Y (heterogeneous) - T1: Matrix multiply results
OnnxComMicrosoftSplitTraining#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSplitTraining(*args, **kwargs)#
Version
name: SplitTraining (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SplitTraining
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input). Default value is
?.
Inputs
input (heterogeneous) - T: The tensor to split
split (heterogeneous) - tensor(int64): length of each output
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
OnnxComMicrosoftSplitTraining_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSplitTraining_1(*args, **kwargs)#
Version
name: SplitTraining (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SplitTraining
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input). Default value is
?.
Inputs
input (heterogeneous) - T: The tensor to split
split (heterogeneous) - tensor(int64): length of each output
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
OnnxComMicrosoftSummaryHistogram#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryHistogram(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryHistogram
Attributes
tag (required): The tag corresponding to the histogram data. Default value is
?.
Inputs
input (heterogeneous) - T: The scalar tensor to produce a histogram over.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryHistogram_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryHistogram_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryHistogram
Attributes
tag (required): The tag corresponding to the histogram data. Default value is
?.
Inputs
input (heterogeneous) - T: The scalar tensor to produce a histogram over.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryMerge#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryMerge(*args, **kwargs)#
Version
name: SummaryMerge (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryMerge
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - S: One or more serialized Tensorboard Summary tensors to merge into a single Summary.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryMerge_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryMerge_1(*args, **kwargs)#
Version
name: SummaryMerge (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryMerge
Inputs
Between 1 and 2147483647 inputs.
input (variadic, heterogeneous) - S: One or more serialized Tensorboard Summary tensors to merge into a single Summary.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryScalar#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryScalar(*args, **kwargs)#
Version
name: SummaryScalar (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryScalar
Attributes
tags (required): The tags corresponding to each input scalar. Default value is
?.
Inputs
input (heterogeneous) - T: The scalar tensor to summarize as simple values.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryScalar_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryScalar_1(*args, **kwargs)#
Version
name: SummaryScalar (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryScalar
Attributes
tags (required): The tags corresponding to each input scalar. Default value is
?.
Inputs
input (heterogeneous) - T: The scalar tensor to summarize as simple values.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryText#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryText(*args, **kwargs)#
Version
name: SummaryText (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryText
Attributes
tag (required): The tag corresponding to the text data. Default value is
?.
Inputs
input (heterogeneous) - S: The string tensor to render in the Tensorboard Text dashboard.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftSummaryText_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftSummaryText_1(*args, **kwargs)#
Version
name: SummaryText (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
SummaryText
Attributes
tag (required): The tag corresponding to the text data. Default value is
?.
Inputs
input (heterogeneous) - S: The string tensor to render in the Tensorboard Text dashboard.
Outputs
summary (heterogeneous) - S: The serialized Tensorboard Summary.
OnnxComMicrosoftTanhGrad#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTanhGrad(*args, **kwargs)#
Version
name: TanhGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
TanhGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
Y (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftTanhGrad_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTanhGrad_1(*args, **kwargs)#
Version
name: TanhGrad (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
TanhGrad
Inputs
dY (heterogeneous) - T: The gradient tensor from output.
Y (heterogeneous) - T: The input tensor.
Outputs
dX (heterogeneous) - T: Gradient of the input.
OnnxComMicrosoftTokenizer#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTokenizer(*args, **kwargs)#
Version
name: Tokenizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Tokenizer divides each string in X into a vector of strings along the last axis. Allowed input shapes are [C] and [N, C]. If the maximum number of tokens found per input string is D, the output shape would be [N, C, D] when input shape is [N, C]. Similarly, if input shape is [C] then the output should be [C, D]. Tokenizer has two different operation modes. The first mode is selected when “tokenexp” is not set and “separators” is set. If “tokenexp” is set and “separators” is not set, the second mode will be used. The first mode breaks each input string into tokens by matching and removing separators. “separators” is a list of strings which are regular expressions. “tokenexp” is a single regular expression. Let’s assume “separators” is [” “] and consider an example. If input is [“Hello World”, “I love computer science !”] whose shape is [2], then the output would be
[[“Hello”, “World”, padvalue, padvalue, padvalue], [“I”, “love”, “computer”, “science”, “!”]] whose shape is [2, 5] because you can find at most 5 tokens per input string. Note that the input at most can have two axes, so 3-D and higher dimension are not supported. If “separators” contains a single empty string, the Tokenizer will enter into character tokenezation mode. This means all strings will be broken part into individual characters. For each input string, the second mode searches matches of “tokenexp” and each match will be a token in Y. The matching of “tokenexp” is conducted greedily (i.e., a match should be as long as possible). This operator searches for the first match starting from the beginning of the considered string, and then launches another search starting from the first remained character after the first matched token. If no match found, this operator will remove the first character from the remained string and do another search. This procedure will be repeated until reaching the end of the considered string.
Let’s consider another example to illustrate the effect of setting “mark” to true. If input is [“Hello”, “World”], then the corresponding output would be [0x02, “Hello”, “World”, 0x03]. This implies that if mark is true, [C]/[N, C] - input’s output shape becomes [C, D+2]/[N, C, D+2].
If tokenizer removes the entire content of [C]-input, it will produce [[]]. I.e. the output shape should be [C][0] or [N][C][0] if input shape was [N][C]. If the tokenizer receives empty input of [0] then the output is [0] if empty input of [N, 0] then [N, 0].
Attributes
mark (required): Boolean whether to mark the beginning/end character with start of text character (0x02)/end of text character (0x03). Default value is
?.mincharnum (required): Minimum number of characters allowed in the output. For example, if mincharnum is 2, tokens such as “A” and “B” would be ignored Default value is
?.pad_value (required): The string used to pad output tensors when the tokens extracted doesn’t match the maximum number of tokens found. If start/end markers are needed, padding will appear outside the markers. Default value is
?.separators: an optional list of strings attribute that contains a list of separators - regular expressions to match separators Two consecutive segments in X connected by a separator would be divided into two tokens. For example, if the input is “Hello World!” and this attribute contains only one space character, the corresponding output would be [“Hello”, “World!”]. To achieve character-level tokenization, one should set the ‘separators’ to [“”], which contains an empty string. Default value is
?.tokenexp: An optional string. Token’s regular expression in basic POSIX format (pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap09.html#ta g_09_03). If set, tokenizer may produce tokens matching the specified pattern. Note that one and only of ‘tokenexp’ and ‘separators’ should be set. Default value is
?.
Inputs
X (heterogeneous) - T: Strings to tokenize
Outputs
Y (heterogeneous) - T: Tokenized strings
OnnxComMicrosoftTokenizer_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTokenizer_1(*args, **kwargs)#
Version
name: Tokenizer (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Tokenizer divides each string in X into a vector of strings along the last axis. Allowed input shapes are [C] and [N, C]. If the maximum number of tokens found per input string is D, the output shape would be [N, C, D] when input shape is [N, C]. Similarly, if input shape is [C] then the output should be [C, D]. Tokenizer has two different operation modes. The first mode is selected when “tokenexp” is not set and “separators” is set. If “tokenexp” is set and “separators” is not set, the second mode will be used. The first mode breaks each input string into tokens by matching and removing separators. “separators” is a list of strings which are regular expressions. “tokenexp” is a single regular expression. Let’s assume “separators” is [” “] and consider an example. If input is [“Hello World”, “I love computer science !”] whose shape is [2], then the output would be
[[“Hello”, “World”, padvalue, padvalue, padvalue], [“I”, “love”, “computer”, “science”, “!”]] whose shape is [2, 5] because you can find at most 5 tokens per input string. Note that the input at most can have two axes, so 3-D and higher dimension are not supported. If “separators” contains a single empty string, the Tokenizer will enter into character tokenezation mode. This means all strings will be broken part into individual characters. For each input string, the second mode searches matches of “tokenexp” and each match will be a token in Y. The matching of “tokenexp” is conducted greedily (i.e., a match should be as long as possible). This operator searches for the first match starting from the beginning of the considered string, and then launches another search starting from the first remained character after the first matched token. If no match found, this operator will remove the first character from the remained string and do another search. This procedure will be repeated until reaching the end of the considered string.
Let’s consider another example to illustrate the effect of setting “mark” to true. If input is [“Hello”, “World”], then the corresponding output would be [0x02, “Hello”, “World”, 0x03]. This implies that if mark is true, [C]/[N, C] - input’s output shape becomes [C, D+2]/[N, C, D+2].
If tokenizer removes the entire content of [C]-input, it will produce [[]]. I.e. the output shape should be [C][0] or [N][C][0] if input shape was [N][C]. If the tokenizer receives empty input of [0] then the output is [0] if empty input of [N, 0] then [N, 0].
Attributes
mark (required): Boolean whether to mark the beginning/end character with start of text character (0x02)/end of text character (0x03). Default value is
?.mincharnum (required): Minimum number of characters allowed in the output. For example, if mincharnum is 2, tokens such as “A” and “B” would be ignored Default value is
?.pad_value (required): The string used to pad output tensors when the tokens extracted doesn’t match the maximum number of tokens found. If start/end markers are needed, padding will appear outside the markers. Default value is
?.separators: an optional list of strings attribute that contains a list of separators - regular expressions to match separators Two consecutive segments in X connected by a separator would be divided into two tokens. For example, if the input is “Hello World!” and this attribute contains only one space character, the corresponding output would be [“Hello”, “World!”]. To achieve character-level tokenization, one should set the ‘separators’ to [“”], which contains an empty string. Default value is
?.tokenexp: An optional string. Token’s regular expression in basic POSIX format (pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap09.html#ta g_09_03). If set, tokenizer may produce tokens matching the specified pattern. Note that one and only of ‘tokenexp’ and ‘separators’ should be set. Default value is
?.
Inputs
X (heterogeneous) - T: Strings to tokenize
Outputs
Y (heterogeneous) - T: Tokenized strings
OnnxComMicrosoftTorchEmbedding#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTorchEmbedding(*args, **kwargs)#
Version
name: TorchEmbedding (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
- Based on Torch operator Embedding, creates a lookup table of embedding vectors of fixed size,
for a dictionary of fixed size.
Inputs
Between 2 and 4 inputs.
weight (heterogeneous) - T: The embedding matrix of size N x M. ‘N’ is equal to the maximum possible index + 1, and ‘M’ is equal to the embedding size
indices (heterogeneous) - tensor(int64): Long tensor containing the indices to extract from embedding matrix.
padding_idx (optional, heterogeneous) - tensor(int64): A 0-D scalar tensor. If specified, the entries at padding_idx do not contribute to the gradient; therefore, the embedding vector at padding_idx is not updated during training, i.e. it remains as a fixed pad.
scale_grad_by_freq (optional, heterogeneous) - tensor(bool): A 0-D bool tensor. If given, this will scale gradients by the inverse of frequency of the indices (words) in the mini-batch. Default is
False
Outputs
Y (heterogeneous) - T: Output tensor of the same type as the input tensor. Shape of the output is * x M, where ‘*’ is the shape of input indices, and ‘M’ is the embedding size.
OnnxComMicrosoftTorchEmbedding_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTorchEmbedding_1(*args, **kwargs)#
Version
name: TorchEmbedding (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
- Based on Torch operator Embedding, creates a lookup table of embedding vectors of fixed size,
for a dictionary of fixed size.
Inputs
Between 2 and 4 inputs.
weight (heterogeneous) - T: The embedding matrix of size N x M. ‘N’ is equal to the maximum possible index + 1, and ‘M’ is equal to the embedding size
indices (heterogeneous) - tensor(int64): Long tensor containing the indices to extract from embedding matrix.
padding_idx (optional, heterogeneous) - tensor(int64): A 0-D scalar tensor. If specified, the entries at padding_idx do not contribute to the gradient; therefore, the embedding vector at padding_idx is not updated during training, i.e. it remains as a fixed pad.
scale_grad_by_freq (optional, heterogeneous) - tensor(bool): A 0-D bool tensor. If given, this will scale gradients by the inverse of frequency of the indices (words) in the mini-batch. Default is
False
Outputs
Y (heterogeneous) - T: Output tensor of the same type as the input tensor. Shape of the output is * x M, where ‘*’ is the shape of input indices, and ‘M’ is the embedding size.
OnnxComMicrosoftTransposeMatMul#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTransposeMatMul(*args, **kwargs)#
Version
name: TransposeMatMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Duplicate of FusedMatMul. Going forward FusedMatMul should be used. This OP will be supported for backward compatibility. Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Attributes
alpha: Scalar multiplier for the product of the input tensors. Default value is
?.transA: Whether A should be transposed on the last two dimensions before doing multiplication Default value is
?.transB: Whether B should be transposed on the last two dimensions before doing multiplication Default value is
?.
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results
OnnxComMicrosoftTransposeMatMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTransposeMatMul_1(*args, **kwargs)#
Version
name: TransposeMatMul (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Duplicate of FusedMatMul. Going forward FusedMatMul should be used. This OP will be supported for backward compatibility. Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Attributes
alpha: Scalar multiplier for the product of the input tensors. Default value is
?.transA: Whether A should be transposed on the last two dimensions before doing multiplication Default value is
?.transB: Whether B should be transposed on the last two dimensions before doing multiplication Default value is
?.
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results
OnnxComMicrosoftTrilu#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTrilu(*args, **kwargs)#
Version
name: Trilu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Returns the upper or lower triangular part of a 2-D matrix, or batches of 2-D matrices. If the attribute “upper” is set to true, the upper triangular matrix is retained. Lower triangular matrix is retained otherwise. Default value for upper is true. Trilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists of the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal. All other elements in the matrix are set to zero. If k = 0, the triangular part on and above/below the main diagonal is retained. If upper is set to true, a positive k retains the upper triangular matrix excluding k diagonals above the main diagonal. A negative k value includes as many diagonals below the main diagonal. If upper is set to false, a positive k retains the lower triangular matrix including k diagonals above the main diagonal. A negative k value excludes as many diagonals below the main diagonal.
Attributes
upper: Boolean. Indicates whether upper or lower part of matrix is retained. Default is true. Default value is
?.
Inputs
Between 1 and 2 inputs.
X (heterogeneous) - T: Input tensor of rank 2 or higher.
k (optional, heterogeneous) - tensor(int64): A 0-D tensor containing a single value corresponding to the number diagonals above or the main diagonal to exclude or include.Default value is 0 if it’s not specified.
Outputs
Y (heterogeneous) - T: Output tensor of the same type and shape as the input tensor.
OnnxComMicrosoftTrilu_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftTrilu_1(*args, **kwargs)#
Version
name: Trilu (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Returns the upper or lower triangular part of a 2-D matrix, or batches of 2-D matrices. If the attribute “upper” is set to true, the upper triangular matrix is retained. Lower triangular matrix is retained otherwise. Default value for upper is true. Trilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists of the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal. All other elements in the matrix are set to zero. If k = 0, the triangular part on and above/below the main diagonal is retained. If upper is set to true, a positive k retains the upper triangular matrix excluding k diagonals above the main diagonal. A negative k value includes as many diagonals below the main diagonal. If upper is set to false, a positive k retains the lower triangular matrix including k diagonals above the main diagonal. A negative k value excludes as many diagonals below the main diagonal.
Attributes
upper: Boolean. Indicates whether upper or lower part of matrix is retained. Default is true. Default value is
?.
Inputs
Between 1 and 2 inputs.
X (heterogeneous) - T: Input tensor of rank 2 or higher.
k (optional, heterogeneous) - tensor(int64): A 0-D tensor containing a single value corresponding to the number diagonals above or the main diagonal to exclude or include.Default value is 0 if it’s not specified.
Outputs
Y (heterogeneous) - T: Output tensor of the same type and shape as the input tensor.
OnnxComMicrosoftUnique#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftUnique(*args, **kwargs)#
Version
name: Unique (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Finds all the unique values (deduped list) present in the given input tensor. This operator returns 3 outputs. The first output tensor ‘uniques’ contains all of the unique elements of the input, sorted in the same order that they occur in the input. The second output tensor ‘idx’ is the same size as the input and it contains the index of each value of the input in ‘uniques’. The third output tensor ‘counts’ contains the count of each element of ‘uniques’ in the input. Example:
input_x = [2, 1, 1, 3, 4, 3] output_uniques = [2, 1, 3, 4] output_idx = [0, 1, 1, 2, 3, 2] output_counts = [1, 2, 2, 1]
Inputs
x (heterogeneous) - T: A 1-D input tensor that is to be processed.
Outputs
y (heterogeneous) - T: A 1-D tensor of the same type as ‘x’ containing all the unique values in ‘x’ sorted in the same order that they occur in the input ‘x’
idx (heterogeneous) - tensor(int64): A 1-D INT64 tensor of the same size as ‘x’ containing the indices for each value in ‘x’ in the output ‘uniques’
counts (heterogeneous) - tensor(int64): A 1-D INT64 tensor containing the the count of each element of ‘uniques’ in the input ‘x’
OnnxComMicrosoftUnique_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftUnique_1(*args, **kwargs)#
Version
name: Unique (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Finds all the unique values (deduped list) present in the given input tensor. This operator returns 3 outputs. The first output tensor ‘uniques’ contains all of the unique elements of the input, sorted in the same order that they occur in the input. The second output tensor ‘idx’ is the same size as the input and it contains the index of each value of the input in ‘uniques’. The third output tensor ‘counts’ contains the count of each element of ‘uniques’ in the input. Example:
input_x = [2, 1, 1, 3, 4, 3] output_uniques = [2, 1, 3, 4] output_idx = [0, 1, 1, 2, 3, 2] output_counts = [1, 2, 2, 1]
Inputs
x (heterogeneous) - T: A 1-D input tensor that is to be processed.
Outputs
y (heterogeneous) - T: A 1-D tensor of the same type as ‘x’ containing all the unique values in ‘x’ sorted in the same order that they occur in the input ‘x’
idx (heterogeneous) - tensor(int64): A 1-D INT64 tensor of the same size as ‘x’ containing the indices for each value in ‘x’ in the output ‘uniques’
counts (heterogeneous) - tensor(int64): A 1-D INT64 tensor containing the the count of each element of ‘uniques’ in the input ‘x’
OnnxComMicrosoftView#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftView(*args, **kwargs)#
Version
name: View (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
View. The output tensors are views of the input, according to the shapes provided.
Inputs
Between 2 and 2147483647 inputs.
input (heterogeneous) - T: Input tensor.
shapes (variadic, heterogeneous) - tensor(int64): Shapes of each view output. The shapes must adds up to the input buffer size.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: Output tensors viewed according the shapes input. It has a one to one mapping to the shapes input
OnnxComMicrosoftView_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftView_1(*args, **kwargs)#
Version
name: View (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
View. The output tensors are views of the input, according to the shapes provided.
Inputs
Between 2 and 2147483647 inputs.
input (heterogeneous) - T: Input tensor.
shapes (variadic, heterogeneous) - tensor(int64): Shapes of each view output. The shapes must adds up to the input buffer size.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: Output tensors viewed according the shapes input. It has a one to one mapping to the shapes input
OnnxComMicrosoftWaitEvent#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftWaitEvent(*args, **kwargs)#
Version
name: WaitEvent (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Wait for an event to be recorded.
Inputs
Between 2 and 2147483647 inputs.
EventIdentifier (heterogeneous) - TInt64: Event identifier to record.
InputData (variadic) - T: Input data.
Outputs
Between 1 and 2147483647 outputs.
OutputData (variadic) - T: Output data.
OnnxComMicrosoftWaitEvent_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftWaitEvent_1(*args, **kwargs)#
Version
name: WaitEvent (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Wait for an event to be recorded.
Inputs
Between 2 and 2147483647 inputs.
EventIdentifier (heterogeneous) - TInt64: Event identifier to record.
InputData (variadic) - T: Input data.
Outputs
Between 1 and 2147483647 outputs.
OutputData (variadic) - T: Output data.
OnnxComMicrosoftWordConvEmbedding#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftWordConvEmbedding(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The WordConvEmbedding takes in a batch of sequence words and embed each word to a vector.
Attributes
char_embedding_size: Integer representing the embedding vector size for each char.If not provide, use the char embedding size of embedding vector. Default value is
?.conv_window_size: This operator applies convolution to word from left to right with window equal to conv_window_size and stride to 1.Take word ‘example’ for example, with conv_window_size equal to 2, conv is applied to [ex],[xa], [am], [mp]…If not provide, use the first dimension of conv kernal shape. Default value is
?.embedding_size: Integer representing the embedding vector size for each word.If not provide, use the fileter size of conv weight Default value is
?.
Inputs
Sequence (heterogeneous) - T: Specify batchs of sequence words to embedding
W (heterogeneous) - T1: Specify weights of conv
B (heterogeneous) - T1: Specify bias of conv
C (heterogeneous) - T1: Specify embedding vector of char
Outputs
Y (heterogeneous) - T1: output
OnnxComMicrosoftWordConvEmbedding_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftWordConvEmbedding_1(*args, **kwargs)#
Version
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
The WordConvEmbedding takes in a batch of sequence words and embed each word to a vector.
Attributes
char_embedding_size: Integer representing the embedding vector size for each char.If not provide, use the char embedding size of embedding vector. Default value is
?.conv_window_size: This operator applies convolution to word from left to right with window equal to conv_window_size and stride to 1.Take word ‘example’ for example, with conv_window_size equal to 2, conv is applied to [ex],[xa], [am], [mp]…If not provide, use the first dimension of conv kernal shape. Default value is
?.embedding_size: Integer representing the embedding vector size for each word.If not provide, use the fileter size of conv weight Default value is
?.
Inputs
Sequence (heterogeneous) - T: Specify batchs of sequence words to embedding
W (heterogeneous) - T1: Specify weights of conv
B (heterogeneous) - T1: Specify bias of conv
C (heterogeneous) - T1: Specify embedding vector of char
Outputs
Y (heterogeneous) - T1: output
OnnxComMicrosoftYieldOp#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftYieldOp(*args, **kwargs)#
Version
name: YieldOp (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Yield Op.
Attributes
full_shape_outputs (required): The indices of the module outputs that must have full shape. Default value is
?.non_differentiable_outputs: The indices of the module outputs that doesn’t have a gradient. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
module_outputs (variadic) - T: Module outputs to be returned to pytorch.
Outputs
Between 0 and 2147483647 outputs.
module_outputs_grad (variadic) - T: Gradient of module outputs returned from pytorch.
OnnxComMicrosoftYieldOp_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftYieldOp_1(*args, **kwargs)#
Version
name: YieldOp (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
Yield Op.
Attributes
full_shape_outputs (required): The indices of the module outputs that must have full shape. Default value is
?.non_differentiable_outputs: The indices of the module outputs that doesn’t have a gradient. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
module_outputs (variadic) - T: Module outputs to be returned to pytorch.
Outputs
Between 0 and 2147483647 outputs.
module_outputs_grad (variadic) - T: Gradient of module outputs returned from pytorch.
OnnxComMicrosoftZeroGradient#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftZeroGradient(*args, **kwargs)#
Version
name: ZeroGradient (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
reset the accumulator for gradient
Inputs
old_gradient (heterogeneous) - T1: historical result of accumulated gradient
reset_signal (heterogeneous) - T2: if this input is available, it is ready to reset the accumulator
Outputs
zero_gradient (heterogeneous) - T1: reset the gradient
OnnxComMicrosoftZeroGradient_1#
- class mlprodict.npy.xop_auto_import_.OnnxComMicrosoftZeroGradient_1(*args, **kwargs)#
Version
name: ZeroGradient (GitHub)
domain: com.microsoft
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain com.microsoft.
Summary
reset the accumulator for gradient
Inputs
old_gradient (heterogeneous) - T1: historical result of accumulated gradient
reset_signal (heterogeneous) - T2: if this input is available, it is ready to reset the accumulator
Outputs
zero_gradient (heterogeneous) - T1: reset the gradient
OnnxComMsInternalNhwcConv#
- class mlprodict.npy.xop_auto_import_.OnnxComMsInternalNhwcConv(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: com.ms.internal.nhwc
since_version: 11
function:
support_level:
shape inference:
This version of the operator has been available since version 11 of domain com.ms.internal.nhwc.
Summary
The convolution operator consumes an input tensor and a filter, and computes the output.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
?.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis. Default value is
?.group: number of groups input channels and output channels are divided into. Default value is
?.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W. Default value is
?.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis. Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis. Default value is
?.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
OnnxComMsInternalNhwcConv_11#
- class mlprodict.npy.xop_auto_import_.OnnxComMsInternalNhwcConv_11(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: com.ms.internal.nhwc
since_version: 11
function:
support_level:
shape inference:
This version of the operator has been available since version 11 of domain com.ms.internal.nhwc.
Summary
The convolution operator consumes an input tensor and a filter, and computes the output.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
?.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis. Default value is
?.group: number of groups input channels and output channels are divided into. Default value is
?.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W. Default value is
?.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis. Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis. Default value is
?.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
OnnxComMsInternalNhwcMaxPool#
- class mlprodict.npy.xop_auto_import_.OnnxComMsInternalNhwcMaxPool(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: com.ms.internal.nhwc
since_version: 11
function:
support_level:
shape inference:
This version of the operator has been available since version 11 of domain com.ms.internal.nhwc.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
?.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
?.dilations: Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis. Default value is
?.kernel_shape (required): The size of the kernel along each axis. Default value is
?.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis. Default value is
?.storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis. Default value is
?.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
OnnxComMsInternalNhwcMaxPool_11#
- class mlprodict.npy.xop_auto_import_.OnnxComMsInternalNhwcMaxPool_11(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: com.ms.internal.nhwc
since_version: 11
function:
support_level:
shape inference:
This version of the operator has been available since version 11 of domain com.ms.internal.nhwc.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
activation:
Default value is
?.activation_params:
Default value is
?.auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
?.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
?.dilations: Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis. Default value is
?.kernel_shape (required): The size of the kernel along each axis. Default value is
?.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis. Default value is
?.storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. Default value is
?.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis. Default value is
?.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
OnnxCompress#
- class mlprodict.npy.xop_auto_import_.OnnxCompress(*args, **kwargs)#
Version
name: Compress (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Selects slices from an input tensor along a given axis where condition evaluates to True for each axis index. In case axis is not provided, input is flattened before elements are selected. Compress behaves like numpy.compress: https://docs.scipy.org/doc/numpy/reference/generated/numpy.compress.html
Attributes
axis: (Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
Inputs
input (heterogeneous) - T: Tensor of rank r >= 1.
condition (heterogeneous) - T1: Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length along the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.
Outputs
output (heterogeneous) - T: Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
T1 in ( tensor(bool) ): Constrain to boolean tensors.
OnnxCompress_11#
- class mlprodict.npy.xop_auto_import_.OnnxCompress_11(*args, **kwargs)#
Version
name: Compress (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Selects slices from an input tensor along a given axis where condition evaluates to True for each axis index. In case axis is not provided, input is flattened before elements are selected. Compress behaves like numpy.compress: https://docs.scipy.org/doc/numpy/reference/generated/numpy.compress.html
Attributes
axis: (Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
Inputs
input (heterogeneous) - T: Tensor of rank r >= 1.
condition (heterogeneous) - T1: Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length along the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.
Outputs
output (heterogeneous) - T: Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
T1 in ( tensor(bool) ): Constrain to boolean tensors.
OnnxCompress_9#
- class mlprodict.npy.xop_auto_import_.OnnxCompress_9(*args, **kwargs)#
Version
name: Compress (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 9.
Summary
Selects slices from an input tensor along a given axis where condition evaluates to True for each axis index. In case axis is not provided, input is flattened before elements are selected. Compress behaves like numpy.compress: https://docs.scipy.org/doc/numpy/reference/generated/numpy.compress.html
Attributes
axis: (Optional) Axis along which to take slices. If not specified, input is flattened before elements being selected.
Inputs
input (heterogeneous) - T: Tensor of rank r >= 1.
condition (heterogeneous) - T1: Rank 1 tensor of booleans to indicate which slices or data elements to be selected. Its length can be less than the input length alone the axis or the flattened input size if axis is not specified. In such cases data slices or elements exceeding the condition length are discarded.
Outputs
output (heterogeneous) - T: Tensor of rank r if axis is specified. Otherwise output is a Tensor of rank 1.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
T1 in ( tensor(bool) ): Constrain to boolean tensors.
OnnxConcat#
- class mlprodict.npy.xop_auto_import_.OnnxConcat(*args, **kwargs)#
Version
name: Concat (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Concatenate a list of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on.
Attributes
axis (required): Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to any tensor type.
OnnxConcatFromSequence#
- class mlprodict.npy.xop_auto_import_.OnnxConcatFromSequence(*args, **kwargs)#
Version
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Concatenate a sequence of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. By default ‘new_axis’ is 0, the behavior is similar to numpy.concatenate. When ‘new_axis’ is 1, the behavior is similar to numpy.stack.
Attributes
axis (required): Which axis to concat on. Accepted range in [-r, r - 1], where r is the rank of input tensors. When new_axis is 1, accepted range is [-r - 1, r].
new_axis: Insert and concatenate on a new axis or not, default 0 means do not insert new axis. Default value is
0.
Inputs
input_sequence (heterogeneous) - S: Sequence of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain input types to any tensor type.
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to any tensor type.
OnnxConcatFromSequence_11#
- class mlprodict.npy.xop_auto_import_.OnnxConcatFromSequence_11(*args, **kwargs)#
Version
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Concatenate a sequence of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on. By default ‘new_axis’ is 0, the behavior is similar to numpy.concatenate. When ‘new_axis’ is 1, the behavior is similar to numpy.stack.
Attributes
axis (required): Which axis to concat on. Accepted range in [-r, r - 1], where r is the rank of input tensors. When new_axis is 1, accepted range is [-r - 1, r].
new_axis: Insert and concatenate on a new axis or not, default 0 means do not insert new axis. Default value is
0.
Inputs
input_sequence (heterogeneous) - S: Sequence of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain input types to any tensor type.
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to any tensor type.
OnnxConcat_1#
- class mlprodict.npy.xop_auto_import_.OnnxConcat_1(*args, **kwargs)#
Version
name: Concat (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Concatenate a list of tensors into a single tensor
Attributes
axis: Which axis to concat on. Default value is 1.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxConcat_11#
- class mlprodict.npy.xop_auto_import_.OnnxConcat_11(*args, **kwargs)#
Version
name: Concat (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Concatenate a list of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on.
Attributes
axis (required): Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to any tensor type.
OnnxConcat_13#
- class mlprodict.npy.xop_auto_import_.OnnxConcat_13(*args, **kwargs)#
Version
name: Concat (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Concatenate a list of tensors into a single tensor. All input tensors must have the same shape, except for the dimension size of the axis to concatenate on.
Attributes
axis (required): Which axis to concat on. A negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(inputs)..
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to any tensor type.
OnnxConcat_4#
- class mlprodict.npy.xop_auto_import_.OnnxConcat_4(*args, **kwargs)#
Version
name: Concat (GitHub)
domain: main
since_version: 4
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 4.
Summary
Concatenate a list of tensors into a single tensor
Attributes
axis (required): Which axis to concat on
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: List of tensors for concatenation
Outputs
concat_result (heterogeneous) - T: Concatenated tensor
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to any tensor type.
OnnxConstant#
- class mlprodict.npy.xop_auto_import_.OnnxConstant(*args, **kwargs)#
Version
name: Constant (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, or value_* must be specified.
Attributes
sparse_value: The value for the elements of the output tensor in sparse format.
value: The value for the elements of the output tensor.
value_float: The value for the sole element for the scalar, float32, output tensor.
value_floats: The values for the elements for the 1D, float32, output tensor.
value_int: The value for the sole element for the scalar, int64, output tensor.
value_ints: The values for the elements for the 1D, int64, output tensor.
value_string: The value for the sole element for the scalar, UTF-8 string, output tensor.
value_strings: The values for the elements for the 1D, UTF-8 string, output tensor.
Outputs
output (heterogeneous) - T: Output tensor containing the same value of the provided tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxConstantOfShape#
- class mlprodict.npy.xop_auto_import_.OnnxConstantOfShape(*args, **kwargs)#
Version
name: ConstantOfShape (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Generate a tensor with given value and shape.
Attributes
value: (Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
Inputs
input (heterogeneous) - T1: 1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
Outputs
output (heterogeneous) - T2: Output tensor of shape specified by ‘input’.If attribute ‘value’ is specified, the value and datatype of the output tensor is taken from ‘value’.If attribute ‘value’ is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
Type Constraints
T1 in ( tensor(int64) ): Constrain input types.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to be numerics.
OnnxConstantOfShape_9#
- class mlprodict.npy.xop_auto_import_.OnnxConstantOfShape_9(*args, **kwargs)#
Version
name: ConstantOfShape (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Generate a tensor with given value and shape.
Attributes
value: (Optional) The value of the output elements.Should be a one-element tensor. If not specified, it defaults to a tensor of value 0 and datatype float32
Inputs
input (heterogeneous) - T1: 1D tensor. The shape of the expected output tensor. If empty tensor is given, the output would be a scalar. All values must be >= 0.
Outputs
output (heterogeneous) - T2: Output tensor of shape specified by ‘input’.If attribute ‘value’ is specified, the value and datatype of the output tensor is taken from ‘value’.If attribute ‘value’ is not specified, the value in the output defaults to 0, and the datatype defaults to float32.
Type Constraints
T1 in ( tensor(int64) ): Constrain input types.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to be numerics.
OnnxConstant_1#
- class mlprodict.npy.xop_auto_import_.OnnxConstant_1(*args, **kwargs)#
Version
name: Constant (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
A constant tensor.
Attributes
value (required): The value for the elements of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor containing the same value of the provided tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxConstant_11#
- class mlprodict.npy.xop_auto_import_.OnnxConstant_11(*args, **kwargs)#
Version
name: Constant (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
A constant tensor. Exactly one of the two attributes, either value or sparse_value, must be specified.
Attributes
sparse_value: The value for the elements of the output tensor in sparse format.
value: The value for the elements of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor containing the same value of the provided tensor.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxConstant_12#
- class mlprodict.npy.xop_auto_import_.OnnxConstant_12(*args, **kwargs)#
Version
name: Constant (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, or value_* must be specified.
Attributes
sparse_value: The value for the elements of the output tensor in sparse format.
value: The value for the elements of the output tensor.
value_float: The value for the sole element for the scalar, float32, output tensor.
value_floats: The values for the elements for the 1D, float32, output tensor.
value_int: The value for the sole element for the scalar, int64, output tensor.
value_ints: The values for the elements for the 1D, int64, output tensor.
value_string: The value for the sole element for the scalar, UTF-8 string, output tensor.
value_strings: The values for the elements for the 1D, UTF-8 string, output tensor.
Outputs
output (heterogeneous) - T: Output tensor containing the same value of the provided tensor.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxConstant_13#
- class mlprodict.npy.xop_auto_import_.OnnxConstant_13(*args, **kwargs)#
Version
name: Constant (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
This operator produces a constant tensor. Exactly one of the provided attributes, either value, sparse_value, or value_* must be specified.
Attributes
sparse_value: The value for the elements of the output tensor in sparse format.
value: The value for the elements of the output tensor.
value_float: The value for the sole element for the scalar, float32, output tensor.
value_floats: The values for the elements for the 1D, float32, output tensor.
value_int: The value for the sole element for the scalar, int64, output tensor.
value_ints: The values for the elements for the 1D, int64, output tensor.
value_string: The value for the sole element for the scalar, UTF-8 string, output tensor.
value_strings: The values for the elements for the 1D, UTF-8 string, output tensor.
Outputs
output (heterogeneous) - T: Output tensor containing the same value of the provided tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxConstant_9#
- class mlprodict.npy.xop_auto_import_.OnnxConstant_9(*args, **kwargs)#
Version
name: Constant (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
A constant tensor.
Attributes
value (required): The value for the elements of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor containing the same value of the provided tensor.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxConv#
- class mlprodict.npy.xop_auto_import_.OnnxConv(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
The convolution operator consumes an input tensor and a filter, and computes the output.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
group: number of groups input channels and output channels are divided into. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxConvInteger#
- class mlprodict.npy.xop_auto_import_.OnnxConvInteger(*args, **kwargs)#
Version
name: ConvInteger (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
The integer convolution operator consumes an input tensor, its zero-point, a filter, and its zero-point, and computes the output. The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each axis.
group: number of groups input channels and output channels are divided into. default is 1. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input ‘w’.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number ofpixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each axis.
Inputs
Between 2 and 4 inputs.
x (heterogeneous) - T1: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
w (heterogeneous) - T2: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
x_zero_point (optional, heterogeneous) - T1: Zero point tensor for input ‘x’. It’s optional and default value is 0. It’s a scalar, which means a per-tensor/layer quantization.
w_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘w’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per- tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M)
Outputs
y (heterogeneous) - T3: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input x and its zero point data type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain input w and its zero point data type to 8-bit integer tensor.
T3 in ( tensor(int32) ): Constrain output y data type to 32-bit integer tensor.
OnnxConvInteger_10#
- class mlprodict.npy.xop_auto_import_.OnnxConvInteger_10(*args, **kwargs)#
Version
name: ConvInteger (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
The integer convolution operator consumes an input tensor, its zero-point, a filter, and its zero-point, and computes the output. The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each axis.
group: number of groups input channels and output channels are divided into. default is 1. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input ‘w’.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number ofpixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each axis.
Inputs
Between 2 and 4 inputs.
x (heterogeneous) - T1: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
w (heterogeneous) - T2: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
x_zero_point (optional, heterogeneous) - T1: Zero point tensor for input ‘x’. It’s optional and default value is 0. It’s a scalar, which means a per-tensor/layer quantization.
w_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘w’. It’s optional and default value is 0. It could be a scalar or a 1-D tensor, which means a per- tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M)
Outputs
y (heterogeneous) - T3: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input x and its zero point data type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain input w and its zero point data type to 8-bit integer tensor.
T3 in ( tensor(int32) ): Constrain output y data type to 32-bit integer tensor.
OnnxConvTranspose#
- class mlprodict.npy.xop_auto_import_.OnnxConvTranspose(*args, **kwargs)#
Version
name: ConvTranspose (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
The convolution transpose operator consumes an input tensor and a filter, and computes the output.
If the pads parameter is provided the shape of the output is calculated via the following equation:
output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i]
output_shape can also be explicitly specified in which case pads values are auto generated using these equations:
total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] If (auto_pads == SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = input_shape[i] * strides[i] for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
group: number of groups input channels and output channels are divided into. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W.
output_padding: Additional elements added to the side with higher coordinate indices in the output. Each padding value in “output_padding” must be less than the corresponding stride/dilation dimension. By default, this attribute is a zero vector. Note that this attribute doesn’t directly affect the computed output values. It only controls the selection of the computed values, so changing this attribute only adds or removes output elements. If “output_shape” is explicitly provided, “output_padding” does not contribute additional size to “output_shape” but participates in the computation of the needed padding amount. This is also called adjs or adjustment in some frameworks.
output_shape: The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn)
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x … x kn), where (k1 x k2 x … x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxConvTranspose_1#
- class mlprodict.npy.xop_auto_import_.OnnxConvTranspose_1(*args, **kwargs)#
Version
name: ConvTranspose (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
The convolution transpose operator consumes an input tensor and a filter, and computes the output.
If the pads parameter is provided the shape of the output is calculated via the following equation:
output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i]
output_shape can also be explicitly specified in which case pads values are auto generated using these equations:
total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] If (auto_pads != SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter.
group: number of groups input channels and output channels are divided into. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W.
output_padding: The zero-padding added to one side of the output. This is also called adjs/adjustment in some frameworks.
output_shape: The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn)
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x … x kn), where (k1 x k2 x … x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxConvTranspose_11#
- class mlprodict.npy.xop_auto_import_.OnnxConvTranspose_11(*args, **kwargs)#
Version
name: ConvTranspose (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
The convolution transpose operator consumes an input tensor and a filter, and computes the output.
If the pads parameter is provided the shape of the output is calculated via the following equation:
output_shape[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - pads[start_i] - pads[end_i]
output_shape can also be explicitly specified in which case pads values are auto generated using these equations:
total_padding[i] = stride[i] * (input_size[i] - 1) + output_padding[i] + ((kernel_shape[i] - 1) * dilations[i] + 1) - output_shape[i] If (auto_pads == SAME_UPPER): pads[start_i] = total_padding[i]/2; pads[end_i] = total_padding[i] - (total_padding[i]/2) Else: pads[start_i] = total_padding[i] - (total_padding[i]/2); pads[end_i] = (total_padding[i]/2).
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = input_shape[i] * strides[i] for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
group: number of groups input channels and output channels are divided into. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W.
output_padding: Additional elements added to the side with higher coordinate indices in the output. Each padding value in “output_padding” must be less than the corresponding stride/dilation dimension. By default, this attribute is a zero vector. Note that this attribute doesn’t directly affect the computed output values. It only controls the selection of the computed values, so changing this attribute only adds or removes output elements. If “output_shape” is explicitly provided, “output_padding” does not contribute additional size to “output_shape” but participates in the computation of the needed padding amount. This is also called adjs or adjustment in some frameworks.
output_shape: The shape of the output can be explicitly set which will cause pads values to be auto generated. If output_shape is specified pads values are ignored. See doc for details for equations to generate pads
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn)
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (C x M/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the weight shape will be (C x M/group x k1 x k2 x … x kn), where (k1 x k2 x … x kn) is the dimension of the kernel. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, pad lengths and group count. The number of channels in the output should be equal to W.shape[1] * group (assuming zero based indices of the shape array)
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxConv_1#
- class mlprodict.npy.xop_auto_import_.OnnxConv_1(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
The convolution operator consumes an input tensor and a filter, and computes the output.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter.
group: number of groups input channels and output channels are divided into. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxConv_11#
- class mlprodict.npy.xop_auto_import_.OnnxConv_11(*args, **kwargs)#
Version
name: Conv (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
The convolution operator consumes an input tensor and a filter, and computes the output.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
group: number of groups input channels and output channels are divided into. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input W.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults is 1 along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
W (heterogeneous) - T: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.
B (optional, heterogeneous) - T: Optional 1D bias to be added to the convolution, has size of M.
Outputs
Y (heterogeneous) - T: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCos#
- class mlprodict.npy.xop_auto_import_.OnnxCos(*args, **kwargs)#
Version
name: Cos (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the cosine of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The cosine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCos_7#
- class mlprodict.npy.xop_auto_import_.OnnxCos_7(*args, **kwargs)#
Version
name: Cos (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the cosine of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The cosine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCosh#
- class mlprodict.npy.xop_auto_import_.OnnxCosh(*args, **kwargs)#
Version
name: Cosh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic cosine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic cosine values of the input tensor computed element- wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCosh_9#
- class mlprodict.npy.xop_auto_import_.OnnxCosh_9(*args, **kwargs)#
Version
name: Cosh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic cosine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic cosine values of the input tensor computed element- wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxCumSum#
- class mlprodict.npy.xop_auto_import_.OnnxCumSum(*args, **kwargs)#
Version
name: CumSum (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs cumulative sum of the input elements along the given axis. By default, it will do the sum inclusively meaning the first element is copied as is. Through an exclusive attribute, this behavior can change to exclude the first element. It can also perform summation in the opposite direction of the axis. For that, set reverse attribute to 1.
Example:
input_x = [1, 2, 3] axis=0 output = [1, 3, 6] exclusive=1 output = [0, 1, 3] exclusive=0 reverse=1 output = [6, 5, 3] exclusive=1 reverse=1 output = [5, 3, 0]
Attributes
exclusive: If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements. Default value is
0.reverse: If set to 1 will perform the sums in reverse direction. Default value is
0.
Inputs
x (heterogeneous) - T: An input tensor that is to be processed.
axis (heterogeneous) - T2: A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
Outputs
y (heterogeneous) - T: Output tensor of the same type as ‘x’ with cumulative sums of the x’s elements
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
T2 in ( tensor(int32), tensor(int64) ): axis tensor can be int32 or int64 only
OnnxCumSum_11#
- class mlprodict.npy.xop_auto_import_.OnnxCumSum_11(*args, **kwargs)#
Version
name: CumSum (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Performs cumulative sum of the input elements along the given axis. By default, it will do the sum inclusively meaning the first element is copied as is. Through an exclusive attribute, this behavior can change to exclude the first element. It can also perform summation in the opposite direction of the axis. For that, set reverse attribute to 1.
Example:
input_x = [1, 2, 3] axis=0 output = [1, 3, 6] exclusive=1 output = [0, 1, 3] exclusive=0 reverse=1 output = [6, 5, 3] exclusive=1 reverse=1 output = [5, 3, 0]
Attributes
exclusive: If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements. Default value is
0.reverse: If set to 1 will perform the sums in reverse direction. Default value is
0.
Inputs
x (heterogeneous) - T: An input tensor that is to be processed.
axis (heterogeneous) - T2: A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
Outputs
y (heterogeneous) - T: Output tensor of the same type as ‘x’ with cumulative sums of the x’s elements
Type Constraints
T in ( tensor(double), tensor(float), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Input can be of any tensor type.
T2 in ( tensor(int32), tensor(int64) ): axis tensor can be int32 or int64 only
OnnxCumSum_14#
- class mlprodict.npy.xop_auto_import_.OnnxCumSum_14(*args, **kwargs)#
Version
name: CumSum (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs cumulative sum of the input elements along the given axis. By default, it will do the sum inclusively meaning the first element is copied as is. Through an exclusive attribute, this behavior can change to exclude the first element. It can also perform summation in the opposite direction of the axis. For that, set reverse attribute to 1.
Example:
input_x = [1, 2, 3] axis=0 output = [1, 3, 6] exclusive=1 output = [0, 1, 3] exclusive=0 reverse=1 output = [6, 5, 3] exclusive=1 reverse=1 output = [5, 3, 0]
Attributes
exclusive: If set to 1 will return exclusive sum in which the top element is not included. In other terms, if set to 1, the j-th output element would be the sum of the first (j-1) elements. Otherwise, it would be the sum of the first j elements. Default value is
0.reverse: If set to 1 will perform the sums in reverse direction. Default value is
0.
Inputs
x (heterogeneous) - T: An input tensor that is to be processed.
axis (heterogeneous) - T2: A 0-D tensor. Must be in the range [-rank(x), rank(x)-1]. Negative value means counting dimensions from the back.
Outputs
y (heterogeneous) - T: Output tensor of the same type as ‘x’ with cumulative sums of the x’s elements
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
T2 in ( tensor(int32), tensor(int64) ): axis tensor can be int32 or int64 only
OnnxDFT#
- class mlprodict.npy.xop_auto_import_.OnnxDFT(*args, **kwargs)#
Version
name: DFT (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Computes the discrete Fourier transform of input.
Attributes
axis: The axis on which to perform the DFT. By default this value is set to 1, which corresponds to the first dimension after the batch index. Default value is
1.inverse: Whether to perform the inverse discrete fourier transform. By default this value is set to 0, which corresponds to false. Default value is
0.onesided: If onesided is 1, only values for w in [0, 1, 2, …, floor(n_fft/2) + 1] are returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m,w]=X[m,n_fft-w]*. Note if the input or window tensors are complex, then onesided output is not possible. Enabling onesided with real inputs performs a Real-valued fast Fourier transform (RFFT). When invoked with real or complex valued input, the default value is 0. Values can be 0 or 1. Default value is
0.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T1: For real input, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[signal_dimN][1]. For complex input, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[signal_dimN][2]. The first dimension is the batch dimension. The following N dimentions correspond to the signal’s dimensions. The final dimension represents the real and imaginary parts of the value in that order.
dft_length (optional, heterogeneous) - T2: The length of the signal.If greater than the axis dimension, the signal will be zero-padded up to dft_length. If less than the axis dimension, only the first dft_length values will be used as the signal. It’s an optional value.
Outputs
output (heterogeneous) - T1: The Fourier Transform of the input vector.If onesided is 0, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[signal_dimN][2]. If axis=0 and onesided is 1, the following shape is expected: [batch_idx][floor(signal_dim1/2)+1][signal_dim2]…[signal_dimN][2]. If axis=1 and onesided is 1, the following shape is expected: [batch_idx][signal_dim1][floor(signal_dim2/2)+1]…[signal_dimN][2]. If axis=N-1 and onesided is 1, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[floor(signal_dimN/2)+1][2]. The signal_dim at the specified axis is equal to the dft_length.
Type Constraints
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T2 in ( tensor(int32), tensor(int64) ): Constrain scalar length types to int64_t.
OnnxDFT_17#
- class mlprodict.npy.xop_auto_import_.OnnxDFT_17(*args, **kwargs)#
Version
name: DFT (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Computes the discrete Fourier transform of input.
Attributes
axis: The axis on which to perform the DFT. By default this value is set to 1, which corresponds to the first dimension after the batch index. Default value is
1.inverse: Whether to perform the inverse discrete fourier transform. By default this value is set to 0, which corresponds to false. Default value is
0.onesided: If onesided is 1, only values for w in [0, 1, 2, …, floor(n_fft/2) + 1] are returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m,w]=X[m,n_fft-w]*. Note if the input or window tensors are complex, then onesided output is not possible. Enabling onesided with real inputs performs a Real-valued fast Fourier transform (RFFT). When invoked with real or complex valued input, the default value is 0. Values can be 0 or 1. Default value is
0.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T1: For real input, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[signal_dimN][1]. For complex input, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[signal_dimN][2]. The first dimension is the batch dimension. The following N dimentions correspond to the signal’s dimensions. The final dimension represents the real and imaginary parts of the value in that order.
dft_length (optional, heterogeneous) - T2: The length of the signal.If greater than the axis dimension, the signal will be zero-padded up to dft_length. If less than the axis dimension, only the first dft_length values will be used as the signal. It’s an optional value.
Outputs
output (heterogeneous) - T1: The Fourier Transform of the input vector.If onesided is 0, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[signal_dimN][2]. If axis=0 and onesided is 1, the following shape is expected: [batch_idx][floor(signal_dim1/2)+1][signal_dim2]…[signal_dimN][2]. If axis=1 and onesided is 1, the following shape is expected: [batch_idx][signal_dim1][floor(signal_dim2/2)+1]…[signal_dimN][2]. If axis=N-1 and onesided is 1, the following shape is expected: [batch_idx][signal_dim1][signal_dim2]…[floor(signal_dimN/2)+1][2]. The signal_dim at the specified axis is equal to the dft_length.
Type Constraints
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T2 in ( tensor(int32), tensor(int64) ): Constrain scalar length types to int64_t.
OnnxDepthToSpace#
- class mlprodict.npy.xop_auto_import_.OnnxDepthToSpace(*args, **kwargs)#
Version
name: DepthToSpace (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of the input tensor where values from the depth dimension are moved in spatial blocks to the height and width dimensions. By default, mode = DCR. In the DCR mode, elements along the depth dimension from the input tensor are rearranged in the following order: depth, column, and then row. The output y is computed from the input x as below:
b, c, h, w = x.shape
tmp = np.reshape(x, [b, blocksize, blocksize, c // (blocksize**2), h, w])
tmp = np.transpose(tmp, [0, 3, 4, 1, 5, 2])
y = np.reshape(tmp, [b, c // (blocksize**2), h * blocksize, w * blocksize])
In the CRD mode, elements along the depth dimension from the input tensor are rearranged in the following order: column, row, and the depth. The output y is computed from the input x as below:
b, c, h, w = x.shape
tmp = np.reshape(x, [b, c // (blocksize ** 2), blocksize, blocksize, h, w])
tmp = np.transpose(tmp, [0, 1, 4, 2, 5, 3])
y = np.reshape(tmp, [b, c // (blocksize ** 2), h * blocksize, w * blocksize])
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
mode: DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order. Default value is
'DCR'.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxDepthToSpace_1#
- class mlprodict.npy.xop_auto_import_.OnnxDepthToSpace_1(*args, **kwargs)#
Version
name: DepthToSpace (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of the input tensor where values from the depth dimension are moved in spatial blocks to the height and width dimensions.
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxDepthToSpace_11#
- class mlprodict.npy.xop_auto_import_.OnnxDepthToSpace_11(*args, **kwargs)#
Version
name: DepthToSpace (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of the input tensor where values from the depth dimension are moved in spatial blocks to the height and width dimensions. By default, mode = DCR. In the DCR mode, elements along the depth dimension from the input tensor are rearranged in the following order: depth, column, and then row. The output y is computed from the input x as below:
b, c, h, w = x.shape
tmp = np.reshape(x, [b, blocksize, blocksize, c // (blocksize**2), h, w])
tmp = np.transpose(tmp, [0, 3, 4, 1, 5, 2])
y = np.reshape(tmp, [b, c // (blocksize**2), h * blocksize, w * blocksize])
In the CRD mode, elements along the depth dimension from the input tensor are rearranged in the following order: column, row, and the depth. The output y is computed from the input x as below:
b, c, h, w = x.shape
tmp = np.reshape(x, [b, c // (blocksize ** 2), blocksize, blocksize, h, w])
tmp = np.transpose(tmp, [0, 1, 4, 2, 5, 3])
y = np.reshape(tmp, [b, c // (blocksize ** 2), h * blocksize, w * blocksize])
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
mode: DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order. Default value is
'DCR'.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxDepthToSpace_13#
- class mlprodict.npy.xop_auto_import_.OnnxDepthToSpace_13(*args, **kwargs)#
Version
name: DepthToSpace (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
DepthToSpace rearranges (permutes) data from depth into blocks of spatial data. This is the reverse transformation of SpaceToDepth. More specifically, this op outputs a copy of the input tensor where values from the depth dimension are moved in spatial blocks to the height and width dimensions. By default, mode = DCR. In the DCR mode, elements along the depth dimension from the input tensor are rearranged in the following order: depth, column, and then row. The output y is computed from the input x as below:
b, c, h, w = x.shape
tmp = np.reshape(x, [b, blocksize, blocksize, c // (blocksize**2), h, w])
tmp = np.transpose(tmp, [0, 3, 4, 1, 5, 2])
y = np.reshape(tmp, [b, c // (blocksize**2), h * blocksize, w * blocksize])
In the CRD mode, elements along the depth dimension from the input tensor are rearranged in the following order: column, row, and the depth. The output y is computed from the input x as below:
b, c, h, w = x.shape
tmp = np.reshape(x, [b, c // (blocksize ** 2), blocksize, blocksize, h, w])
tmp = np.transpose(tmp, [0, 1, 4, 2, 5, 3])
y = np.reshape(tmp, [b, c // (blocksize ** 2), h * blocksize, w * blocksize])
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
mode: DCR (default) for depth-column-row order re-arrangement. Use CRD for column-row-depth order. Default value is
'DCR'.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C/(blocksize * blocksize), H * blocksize, W * blocksize].
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxDequantizeLinear#
- class mlprodict.npy.xop_auto_import_.OnnxDequantizeLinear(*args, **kwargs)#
Version
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the full precision tensor. The dequantization formula is y = (x - x_zero_point) * x_scale. ‘x_scale’ and ‘x_zero_point’ must have same shape, and can be either a scalar for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. ‘x_zero_point’ and ‘x’ must have same type. ‘x’ and ‘y’ must have same shape. In the case of dequantizing int32, there’s no zero point (zero point is supposed to be 0).
Attributes
axis: (Optional) The axis of the dequantizing dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
Between 2 and 3 inputs.
x (heterogeneous) - T: N-D quantized input tensor to be de-quantized.
x_scale (heterogeneous) - tensor(float): Scale for input ‘x’. It can be a scalar, which means a per- tensor/layer dequantization, or a 1-D tensor for per-axis dequantization.
x_zero_point (optional, heterogeneous) - T: Zero point for input ‘x’. Shape must match x_scale. It’s optional. Zero point is 0 when it’s not specified.
Outputs
y (heterogeneous) - tensor(float): N-D full precision output tensor. It has same shape as input ‘x’.
Type Constraints
T in ( tensor(int32), tensor(int8), tensor(uint8) ): Constrain ‘x_zero_point’ and ‘x’ to 8-bit/32-bit integer tensor.
OnnxDequantizeLinear_10#
- class mlprodict.npy.xop_auto_import_.OnnxDequantizeLinear_10(*args, **kwargs)#
Version
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
The linear dequantization operator. It consumes a quantized tensor, a scale, a zero point to compute the full precision tensor. The dequantization formula is y = (x - x_zero_point) * x_scale. ‘x_scale’ and ‘x_zero_point’ are both scalars. ‘x_zero_point’ and ‘x’ must have same type. ‘x’ and ‘y’ must have same shape. In the case of dequantizing int32, there’s no zero point (zero point is supposed to be 0).
Inputs
Between 2 and 3 inputs.
x (heterogeneous) - T: N-D quantized input tensor to be de-quantized.
x_scale (heterogeneous) - tensor(float): Scale for input ‘x’. It’s a scalar, which means a per-tensor/layer quantization.
x_zero_point (optional, heterogeneous) - T: Zero point for input ‘x’. It’s a scalar, which means a per- tensor/layer quantization. It’s optional. 0 is the default value when it’s not specified.
Outputs
y (heterogeneous) - tensor(float): N-D full precision output tensor. It has same shape as input ‘x’.
Type Constraints
T in ( tensor(int32), tensor(int8), tensor(uint8) ): Constrain ‘x_zero_point’ and ‘x’ to 8-bit/32-bit integer tensor.
OnnxDequantizeLinear_13#
- class mlprodict.npy.xop_auto_import_.OnnxDequantizeLinear_13(*args, **kwargs)#
Version
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The linear dequantization operator. It consumes a quantized tensor, a scale, and a zero point to compute the full precision tensor. The dequantization formula is y = (x - x_zero_point) * x_scale. ‘x_scale’ and ‘x_zero_point’ must have same shape, and can be either a scalar for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. ‘x_zero_point’ and ‘x’ must have same type. ‘x’ and ‘y’ must have same shape. In the case of dequantizing int32, there’s no zero point (zero point is supposed to be 0).
Attributes
axis: (Optional) The axis of the dequantizing dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
Between 2 and 3 inputs.
x (heterogeneous) - T: N-D quantized input tensor to be de-quantized.
x_scale (heterogeneous) - tensor(float): Scale for input ‘x’. It can be a scalar, which means a per- tensor/layer dequantization, or a 1-D tensor for per-axis dequantization.
x_zero_point (optional, heterogeneous) - T: Zero point for input ‘x’. Shape must match x_scale. It’s optional. Zero point is 0 when it’s not specified.
Outputs
y (heterogeneous) - tensor(float): N-D full precision output tensor. It has same shape as input ‘x’.
Type Constraints
T in ( tensor(int32), tensor(int8), tensor(uint8) ): Constrain ‘x_zero_point’ and ‘x’ to 8-bit/32-bit integer tensor.
OnnxDet#
- class mlprodict.npy.xop_auto_import_.OnnxDet(*args, **kwargs)#
Version
name: Det (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Det calculates determinant of a square matrix or batches of square matrices. Det takes one input tensor of shape [*, M, M], where * is zero or more batch dimensions, and the inner-most 2 dimensions form square matrices. The output is a tensor of shape [*], containing the determinants of all input submatrices. e.g., When the input is 2-D, the output is a scalar(shape is empty: []).
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to floating-point tensors.
OnnxDet_11#
- class mlprodict.npy.xop_auto_import_.OnnxDet_11(*args, **kwargs)#
Version
name: Det (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Det calculates determinant of a square matrix or batches of square matrices. Det takes one input tensor of shape [*, M, M], where * is zero or more batch dimensions, and the inner-most 2 dimensions form square matrices. The output is a tensor of shape [*], containing the determinants of all input submatrices. e.g., When the input is 2-D, the output is a scalar(shape is empty: []).
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to floating-point tensors.
OnnxDiv#
- class mlprodict.npy.xop_auto_import_.OnnxDiv(*args, **kwargs)#
Version
name: Div (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary division (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxDiv_1#
- class mlprodict.npy.xop_auto_import_.OnnxDiv_1(*args, **kwargs)#
Version
name: Div (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Performs element-wise binary division (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.consumed_inputs: legacy optimization attribute.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxDiv_13#
- class mlprodict.npy.xop_auto_import_.OnnxDiv_13(*args, **kwargs)#
Version
name: Div (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Performs element-wise binary division (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxDiv_14#
- class mlprodict.npy.xop_auto_import_.OnnxDiv_14(*args, **kwargs)#
Version
name: Div (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary division (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxDiv_6#
- class mlprodict.npy.xop_auto_import_.OnnxDiv_6(*args, **kwargs)#
Version
name: Div (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Performs element-wise binary division (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxDiv_7#
- class mlprodict.npy.xop_auto_import_.OnnxDiv_7(*args, **kwargs)#
Version
name: Div (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Performs element-wise binary division (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxDropout#
- class mlprodict.npy.xop_auto_import_.OnnxDropout(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, output (floating-point tensor) and mask (optional Tensor<bool>). If training_mode is true then the output Y will be a random dropout; Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, the user can simply not pass training_mode input or set it to false.
output = scale * data * mask,
where
scale = 1. / (1. - ratio).
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
Between 1 and 3 inputs.
data (heterogeneous) - T: The input data as Tensor.
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T2: The output mask.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input ‘ratio’ types to float tensors.
T2 in ( tensor(bool) ): Constrain output ‘mask’ types to boolean tensors.
OnnxDropout_1#
- class mlprodict.npy.xop_auto_import_.OnnxDropout_1(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Dropout takes one input data (Tensor<float>) and produces two Tensor outputs, output (Tensor<float>) and mask (Tensor<bool>). Depending on whether it is in test mode or not, the output Y will either be a random dropout, or a simple copy of the input. Note that our implementation of Dropout does scaling in the training phase, so during testing nothing needs to be done.
Attributes
consumed_inputs: legacy optimization attribute.
is_test: (int, default 0) if nonzero, run dropout in test mode where the output is simply Y = X. Default value is
0.ratio: (float, default 0.5) the ratio of random dropout Default value is
0.5.
Inputs
data (heterogeneous) - T: The input data as Tensor.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T: The output mask. If is_test is nonzero, this output is not filled.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxDropout_10#
- class mlprodict.npy.xop_auto_import_.OnnxDropout_10(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Dropout takes one input floating tensor and produces two tensor outputs, output (floating tensor) and mask (Tensor<bool>). Depending on whether it is in test mode or not, the output Y will either be a random dropout, or a simple copy of the input. Note that our implementation of Dropout does scaling in the training phase, so during testing nothing needs to be done. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
ratio: The ratio of random dropout Default value is
0.5.
Inputs
data (heterogeneous) - T: The input data as Tensor.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T1: The output mask.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(bool) ): Constrain output mask types to boolean tensors.
OnnxDropout_12#
- class mlprodict.npy.xop_auto_import_.OnnxDropout_12(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, output (floating-point tensor) and mask (optional Tensor<bool>). If training_mode is true then the output Y will be a random dropout; Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, the user can simply not pass training_mode input or set it to false.
output = scale * data * mask,
where
scale = 1. / (1. - ratio).
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
Between 1 and 3 inputs.
data (heterogeneous) - T: The input data as Tensor.
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T2: The output mask.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input ‘ratio’ types to float tensors.
T2 in ( tensor(bool) ): Constrain output ‘mask’ types to boolean tensors.
OnnxDropout_13#
- class mlprodict.npy.xop_auto_import_.OnnxDropout_13(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Dropout takes an input floating-point tensor, an optional input ratio (floating-point scalar) and an optional input training_mode (boolean scalar). It produces two tensor outputs, output (floating-point tensor) and mask (optional Tensor<bool>). If training_mode is true then the output Y will be a random dropout; Note that this Dropout scales the masked input data by the following equation, so to convert the trained model into inference mode, the user can simply not pass training_mode input or set it to false.
output = scale * data * mask,
where
scale = 1. / (1. - ratio).
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
Between 1 and 3 inputs.
data (heterogeneous) - T: The input data as Tensor.
ratio (optional, heterogeneous) - T1: The ratio of random dropout, with value in [0, 1). If this input was not set, or if it was set to 0, the output would be a simple copy of the input. If it’s non-zero, output will be a random dropout of the scaled input, which is typically the case during training. It is an optional value, if not specified it will default to 0.5.
training_mode (optional, heterogeneous) - T2: If set to true then it indicates dropout is being used for training. It is an optional value hence unless specified explicitly, it is false. If it is false, ratio is ignored and the operation mimics inference mode where nothing will be dropped from the input data and if mask is requested as output it will contain all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T2: The output mask.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input ‘ratio’ types to float tensors.
T2 in ( tensor(bool) ): Constrain output ‘mask’ types to boolean tensors.
OnnxDropout_6#
- class mlprodict.npy.xop_auto_import_.OnnxDropout_6(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Dropout takes one input data (Tensor<float>) and produces two Tensor outputs, output (Tensor<float>) and mask (Tensor<bool>). Depending on whether it is in test mode or not, the output Y will either be a random dropout, or a simple copy of the input. Note that our implementation of Dropout does scaling in the training phase, so during testing nothing needs to be done.
Attributes
is_test: (int, default 0) if nonzero, run dropout in test mode where the output is simply Y = X. Default value is
0.ratio: (float, default 0.5) the ratio of random dropout Default value is
0.5.
Inputs
data (heterogeneous) - T: The input data as Tensor.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T: The output mask. If is_test is nonzero, this output is not filled.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxDropout_7#
- class mlprodict.npy.xop_auto_import_.OnnxDropout_7(*args, **kwargs)#
Version
name: Dropout (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Dropout takes one input data (Tensor<float>) and produces two Tensor outputs, output (Tensor<float>) and mask (Tensor<bool>). Depending on whether it is in test mode or not, the output Y will either be a random dropout, or a simple copy of the input. Note that our implementation of Dropout does scaling in the training phase, so during testing nothing needs to be done. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
ratio: The ratio of random dropout Default value is
0.5.
Inputs
data (heterogeneous) - T: The input data as Tensor.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: The output.
mask (optional, heterogeneous) - T: The output mask.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxDynamicQuantizeLinear#
- class mlprodict.npy.xop_auto_import_.OnnxDynamicQuantizeLinear(*args, **kwargs)#
Version
domain: main
since_version: 11
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
A Function to fuse calculation for Scale, Zero Point and FP32->8Bit convertion of FP32 Input data. Outputs Scale, ZeroPoint and Quantized Input for a given FP32 Input. Scale is calculated as:
y_scale = (max(x) - min(x))/(qmax - qmin) * where qmax and qmin are max and min values for quantization range .i.e [0, 255] in case of uint8 * data range is adjusted to include 0.
Zero point is calculated as:
intermediate_zero_point = qmin - min(x)/y_scale y_zero_point = cast(round(saturate(itermediate_zero_point))) * where qmax and qmin are max and min values for quantization range .i.e [0, 255] in case of uint8 * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. * rounding to nearest ties to even.
Data quantization formula is:
y = saturate (round (x / y_scale) + y_zero_point) * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. * rounding to nearest ties to even.
Inputs
x (heterogeneous) - T1: Input tensor
Outputs
y (heterogeneous) - T2: Quantized output tensor
y_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
y_zero_point (heterogeneous) - T2: Output zero point. It’s a scalar, which means a per-tensor/layer quantization.
Type Constraints
T1 in ( tensor(float) ): Constrain ‘x’ to float tensor.
T2 in ( tensor(uint8) ): Constrain ‘y_zero_point’ and ‘y’ to 8-bit unsigned integer tensor.
OnnxDynamicQuantizeLinear_11#
- class mlprodict.npy.xop_auto_import_.OnnxDynamicQuantizeLinear_11(*args, **kwargs)#
Version
domain: main
since_version: 11
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
A Function to fuse calculation for Scale, Zero Point and FP32->8Bit convertion of FP32 Input data. Outputs Scale, ZeroPoint and Quantized Input for a given FP32 Input. Scale is calculated as:
y_scale = (max(x) - min(x))/(qmax - qmin) * where qmax and qmin are max and min values for quantization range .i.e [0, 255] in case of uint8 * data range is adjusted to include 0.
Zero point is calculated as:
intermediate_zero_point = qmin - min(x)/y_scale y_zero_point = cast(round(saturate(itermediate_zero_point))) * where qmax and qmin are max and min values for quantization range .i.e [0, 255] in case of uint8 * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. * rounding to nearest ties to even.
Data quantization formula is:
y = saturate (round (x / y_scale) + y_zero_point) * for saturation, it saturates to [0, 255] if it's uint8, or [-127, 127] if it's int8. Right now only uint8 is supported. * rounding to nearest ties to even.
Inputs
x (heterogeneous) - T1: Input tensor
Outputs
y (heterogeneous) - T2: Quantized output tensor
y_scale (heterogeneous) - tensor(float): Output scale. It’s a scalar, which means a per-tensor/layer quantization.
y_zero_point (heterogeneous) - T2: Output zero point. It’s a scalar, which means a per-tensor/layer quantization.
Type Constraints
T1 in ( tensor(float) ): Constrain ‘x’ to float tensor.
T2 in ( tensor(uint8) ): Constrain ‘y_zero_point’ and ‘y’ to 8-bit unsigned integer tensor.
OnnxEinsum#
- class mlprodict.npy.xop_auto_import_.OnnxEinsum(*args, **kwargs)#
Version
name: Einsum (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
An einsum of the form
`term1, term2 -> output-term`produces an output tensor using the following equationwhere the reduce-sum performs a summation over all the indices occurring in the input terms (term1, term2) that do not occur in the output-term. The Einsum operator evaluates algebraic tensor operations on a sequence of tensors, using the Einstein summation convention. The equation string contains a comma-separated sequence of lower case letters. Each term corresponds to an operand tensor, and the characters within the terms correspond to operands dimensions. This sequence may be followed by "->" to separate the left and right hand side of the equation. If the equation contains "->" followed by the right-hand side, the explicit (not classical) form of the Einstein summation is performed, and the right-hand side indices indicate output tensor dimensions. In other cases, output indices are (implicitly) set to the alphabetically sorted sequence of indices appearing exactly once in the equation. When a dimension character is repeated in the left-hand side, it represents summation along the dimension. The equation may contain ellipsis ("...") to enable broadcasting. Ellipsis must indicate a fixed number of dimensions. Specifically, every occurrence of ellipsis in the equation must represent the same number of dimensions. The right-hand side may contain exactly one ellipsis. In implicit mode, the ellipsis dimensions are set to the beginning of the output. The equation string may contain space (U+0020) character.
Attributes
equation (required): Einsum expression string.
Inputs
Between 1 and 2147483647 inputs.
Inputs (variadic, heterogeneous) - T: Operands
Outputs
Output (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numerical tensor types.
OnnxEinsum_12#
- class mlprodict.npy.xop_auto_import_.OnnxEinsum_12(*args, **kwargs)#
Version
name: Einsum (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
An einsum of the form
`term1, term2 -> output-term`produces an output tensor using the following equationwhere the reduce-sum performs a summation over all the indices occurring in the input terms (term1, term2) that do not occur in the output-term. The Einsum operator evaluates algebraic tensor operations on a sequence of tensors, using the Einstein summation convention. The equation string contains a comma-separated sequence of lower case letters. Each term corresponds to an operand tensor, and the characters within the terms correspond to operands dimensions. This sequence may be followed by "->" to separate the left and right hand side of the equation. If the equation contains "->" followed by the right-hand side, the explicit (not classical) form of the Einstein summation is performed, and the right-hand side indices indicate output tensor dimensions. In other cases, output indices are (implicitly) set to the alphabetically sorted sequence of indices appearing exactly once in the equation. When a dimension character is repeated in the left-hand side, it represents summation along the dimension. The equation may contain ellipsis ("...") to enable broadcasting. Ellipsis must indicate a fixed number of dimensions. Specifically, every occurrence of ellipsis in the equation must represent the same number of dimensions. The right-hand side may contain exactly one ellipsis. In implicit mode, the ellipsis dimensions are set to the beginning of the output. The equation string may contain space (U+0020) character.
Attributes
equation (required): Einsum expression string.
Inputs
Between 1 and 2147483647 inputs.
Inputs (variadic, heterogeneous) - T: Operands
Outputs
Output (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numerical tensor types.
OnnxElu#
- class mlprodict.npy.xop_auto_import_.OnnxElu(*args, **kwargs)#
Version
name: Elu (GitHub)
domain: main
since_version: 6
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Elu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the function f(x) = alpha * (exp(x) - 1.) for x < 0, f(x) = x for x >= 0., is applied to the tensor elementwise.
Attributes
alpha: Coefficient of ELU. Default value is
1.0.
Inputs
X (heterogeneous) - T: 1D input tensor
Outputs
Y (heterogeneous) - T: 1D output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxElu_1#
- class mlprodict.npy.xop_auto_import_.OnnxElu_1(*args, **kwargs)#
Version
name: Elu (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Elu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the function f(x) = alpha * (exp(x) - 1.) for x < 0, f(x) = x for x >= 0., is applied to the tensor elementwise.
Attributes
alpha: Coefficient of ELU default to 1.0. Default value is
1.0.consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: 1D input tensor
Outputs
Y (heterogeneous) - T: 1D input tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxElu_6#
- class mlprodict.npy.xop_auto_import_.OnnxElu_6(*args, **kwargs)#
Version
name: Elu (GitHub)
domain: main
since_version: 6
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Elu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the function f(x) = alpha * (exp(x) - 1.) for x < 0, f(x) = x for x >= 0., is applied to the tensor elementwise.
Attributes
alpha: Coefficient of ELU. Default value is
1.0.
Inputs
X (heterogeneous) - T: 1D input tensor
Outputs
Y (heterogeneous) - T: 1D output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxEqual#
- class mlprodict.npy.xop_auto_import_.OnnxEqual(*args, **kwargs)#
Version
name: Equal (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the tensor resulted from performing the equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxEqual_1#
- class mlprodict.npy.xop_auto_import_.OnnxEqual_1(*args, **kwargs)#
Version
name: Equal (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the tensor resulted from performing the equal logical operation elementwise on the input tensors A and B.
If broadcasting is enabled, the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. See the doc of Add for a detailed description of the broadcasting rules.
Attributes
axis: If set, defines the broadcast dimensions.
broadcast: Enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: Left input tensor for the logical operator.
B (heterogeneous) - T: Right input tensor for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool), tensor(int32), tensor(int64) ): Constrain input to integral tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxEqual_11#
- class mlprodict.npy.xop_auto_import_.OnnxEqual_11(*args, **kwargs)#
Version
name: Equal (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Returns the tensor resulted from performing the equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxEqual_13#
- class mlprodict.npy.xop_auto_import_.OnnxEqual_13(*args, **kwargs)#
Version
name: Equal (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the tensor resulted from performing the equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxEqual_7#
- class mlprodict.npy.xop_auto_import_.OnnxEqual_7(*args, **kwargs)#
Version
name: Equal (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool), tensor(int32), tensor(int64) ): Constrain input to integral tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxErf#
- class mlprodict.npy.xop_auto_import_.OnnxErf(*args, **kwargs)#
Version
name: Erf (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the error function of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The error function of the input tensor computed element-wise. It has the same shape and type of the input.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxErf_13#
- class mlprodict.npy.xop_auto_import_.OnnxErf_13(*args, **kwargs)#
Version
name: Erf (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the error function of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The error function of the input tensor computed element-wise. It has the same shape and type of the input.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxErf_9#
- class mlprodict.npy.xop_auto_import_.OnnxErf_9(*args, **kwargs)#
Version
name: Erf (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Computes the error function of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The error function of the input tensor computed element-wise. It has the same shape and type of the input.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxExp#
- class mlprodict.npy.xop_auto_import_.OnnxExp(*args, **kwargs)#
Version
name: Exp (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculates the exponential of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The exponential of the input tensor computed element-wise
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxExp_1#
- class mlprodict.npy.xop_auto_import_.OnnxExp_1(*args, **kwargs)#
Version
name: Exp (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Calculates the exponential of the given input tensor, element-wise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The exponential of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxExp_13#
- class mlprodict.npy.xop_auto_import_.OnnxExp_13(*args, **kwargs)#
Version
name: Exp (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculates the exponential of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The exponential of the input tensor computed element-wise
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxExp_6#
- class mlprodict.npy.xop_auto_import_.OnnxExp_6(*args, **kwargs)#
Version
name: Exp (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Calculates the exponential of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The exponential of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxExpand#
- class mlprodict.npy.xop_auto_import_.OnnxExpand(*args, **kwargs)#
Version
name: Expand (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Broadcast the input tensor following the given shape and the broadcast rule. The broadcast rule is similar to numpy.array(input) * numpy.ones(shape): Dimensions are right alignment; Two corresponding dimensions must have the same value, or one of them is equal to 1. Also, this operator is similar to numpy.broadcast_to(input, shape), but the major difference is numpy.broadcast_to() does not allow shape to be smaller than input.size(). It is possible that the output.shape is not equal to shape, when some dimensions in shape is equal to 1, or the shape.ndim < input.shape.ndim.
Inputs
input (heterogeneous) - T: Input tensor
shape (heterogeneous) - tensor(int64): A 1-D tensor indicates the shape you want to expand to, following the broadcast rule
Outputs
output (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensors.
OnnxExpand_13#
- class mlprodict.npy.xop_auto_import_.OnnxExpand_13(*args, **kwargs)#
Version
name: Expand (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Broadcast the input tensor following the given shape and the broadcast rule. The broadcast rule is similar to numpy.array(input) * numpy.ones(shape): Dimensions are right alignment; Two corresponding dimensions must have the same value, or one of them is equal to 1. Also, this operator is similar to numpy.broadcast_to(input, shape), but the major difference is numpy.broadcast_to() does not allow shape to be smaller than input.size(). It is possible that the output.shape is not equal to shape, when some dimensions in shape is equal to 1, or the shape.ndim < input.shape.ndim.
Inputs
input (heterogeneous) - T: Input tensor
shape (heterogeneous) - tensor(int64): A 1-D tensor indicates the shape you want to expand to, following the broadcast rule
Outputs
output (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensors.
OnnxExpand_8#
- class mlprodict.npy.xop_auto_import_.OnnxExpand_8(*args, **kwargs)#
Version
name: Expand (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
Broadcast the input tensor following the given shape and the broadcast rule. The broadcast rule is similar to numpy.array(input) * numpy.ones(shape): Dimensions are right alignment; Two corresponding dimensions must have the same value, or one of them is equal to 1. Also, this operator is similar to numpy.broadcast_to(input, shape), but the major difference is numpy.broadcast_to() does not allow shape to be smaller than input.size(). It is possible that the output.shape is not equal to shape, when some dimensions in shape is equal to 1, or the shape.ndim < input.shape.ndim.
Inputs
input (heterogeneous) - T: Input tensor
shape (heterogeneous) - tensor(int64): A 1-D tensor indicates the shape you want to expand to, following the broadcast rule
Outputs
output (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensors.
OnnxEyeLike#
- class mlprodict.npy.xop_auto_import_.OnnxEyeLike(*args, **kwargs)#
Version
name: EyeLike (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere else. Only 2D tensors are supported, i.e. input T1 must be of rank 2. The shape of the output tensor is the same as the input tensor. The data type can be specified by the ‘dtype’ argument. If ‘dtype’ is not specified, then the type of input tensor is used. By default, the main diagonal is populated with ones, but attribute ‘k’ can be used to populate upper or lower diagonals. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message and be valid as an output type.
Attributes
dtype: (Optional) The data type for the elements of the output tensor. If not specified,the data type of the input tensor T1 is used. If input tensor T1 is also notspecified, then type defaults to ‘float’.
k: (Optional) Index of the diagonal to be populated with ones. Default is 0. If T2 is the output, this op sets T2[i, i+k] = 1. k = 0 populates the main diagonal, k > 0 populates an upper diagonal, and k < 0 populates a lower diagonal. Default value is
0.
Inputs
input (heterogeneous) - T1: 2D input tensor to copy shape, and optionally, type information from.
Outputs
output (heterogeneous) - T2: Output tensor, same shape as input tensor T1.
Type Constraints
T1 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Strings and complex are not supported.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Strings and complex are not supported.
OnnxEyeLike_9#
- class mlprodict.npy.xop_auto_import_.OnnxEyeLike_9(*args, **kwargs)#
Version
name: EyeLike (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Generate a 2D tensor (matrix) with ones on the diagonal and zeros everywhere else. Only 2D tensors are supported, i.e. input T1 must be of rank 2. The shape of the output tensor is the same as the input tensor. The data type can be specified by the ‘dtype’ argument. If ‘dtype’ is not specified, then the type of input tensor is used. By default, the main diagonal is populated with ones, but attribute ‘k’ can be used to populate upper or lower diagonals. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message and be valid as an output type.
Attributes
dtype: (Optional) The data type for the elements of the output tensor. If not specified,the data type of the input tensor T1 is used. If input tensor T1 is also notspecified, then type defaults to ‘float’.
k: (Optional) Index of the diagonal to be populated with ones. Default is 0. If T2 is the output, this op sets T2[i, i+k] = 1. k = 0 populates the main diagonal, k > 0 populates an upper diagonal, and k < 0 populates a lower diagonal. Default value is
0.
Inputs
input (heterogeneous) - T1: 2D input tensor to copy shape, and optionally, type information from.
Outputs
output (heterogeneous) - T2: Output tensor, same shape as input tensor T1.
Type Constraints
T1 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types. Strings and complex are not supported.
T2 in ( tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types. Strings and complex are not supported.
OnnxFlatten#
- class mlprodict.npy.xop_auto_import_.OnnxFlatten(*args, **kwargs)#
Version
name: Flatten (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Flattens the input tensor into a 2D matrix. If input tensor has shape (d_0, d_1, … d_n) then the output will have shape (d_0 X d_1 … d_(axis-1), d_axis X d_(axis+1) … X dn).
Attributes
axis: Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 … d_n), where the shape of the input tensor is (d_0, d_1, … d_n). Default value is
1.
Inputs
input (heterogeneous) - T: A tensor of rank >= axis.
Outputs
output (heterogeneous) - T: A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output to all tensor types.
OnnxFlatten_1#
- class mlprodict.npy.xop_auto_import_.OnnxFlatten_1(*args, **kwargs)#
Version
name: Flatten (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Flattens the input tensor into a 2D matrix. If input tensor has shape (d_0, d_1, … d_n) then the output will have shape (d_0 X d_1 … d_(axis-1), d_axis X d_(axis+1) … X dn).
Attributes
axis: Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [0, R], where R is the rank of the input tensor. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 … d_n), where the shape of the input tensor is (d_0, d_1, … d_n). Default value is
1.
Inputs
input (heterogeneous) - T: A tensor of rank >= axis.
Outputs
output (heterogeneous) - T: A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxFlatten_11#
- class mlprodict.npy.xop_auto_import_.OnnxFlatten_11(*args, **kwargs)#
Version
name: Flatten (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Flattens the input tensor into a 2D matrix. If input tensor has shape (d_0, d_1, … d_n) then the output will have shape (d_0 X d_1 … d_(axis-1), d_axis X d_(axis+1) … X dn).
Attributes
axis: Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 … d_n), where the shape of the input tensor is (d_0, d_1, … d_n). Default value is
1.
Inputs
input (heterogeneous) - T: A tensor of rank >= axis.
Outputs
output (heterogeneous) - T: A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output to all tensor types.
OnnxFlatten_13#
- class mlprodict.npy.xop_auto_import_.OnnxFlatten_13(*args, **kwargs)#
Version
name: Flatten (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Flattens the input tensor into a 2D matrix. If input tensor has shape (d_0, d_1, … d_n) then the output will have shape (d_0 X d_1 … d_(axis-1), d_axis X d_(axis+1) … X dn).
Attributes
axis: Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [-r, r], where r is the rank of the input tensor. Negative value means counting dimensions from the back. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 … d_n), where the shape of the input tensor is (d_0, d_1, … d_n). Default value is
1.
Inputs
input (heterogeneous) - T: A tensor of rank >= axis.
Outputs
output (heterogeneous) - T: A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output to all tensor types.
OnnxFlatten_9#
- class mlprodict.npy.xop_auto_import_.OnnxFlatten_9(*args, **kwargs)#
Version
name: Flatten (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Flattens the input tensor into a 2D matrix. If input tensor has shape (d_0, d_1, … d_n) then the output will have shape (d_0 X d_1 … d_(axis-1), d_axis X d_(axis+1) … X dn).
Attributes
axis: Indicate up to which input dimensions (exclusive) should be flattened to the outer dimension of the output. The value for axis must be in the range [0, R], where R is the rank of the input tensor. When axis = 0, the shape of the output tensor is (1, (d_0 X d_1 … d_n), where the shape of the input tensor is (d_0, d_1, … d_n). Default value is
1.
Inputs
input (heterogeneous) - T: A tensor of rank >= axis.
Outputs
output (heterogeneous) - T: A 2D tensor with the contents of the input tensor, with input dimensions up to axis flattened to the outer dimension of the output and remaining input dimensions flattened into the inner dimension of the output.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output to all tensor types.
OnnxFloor#
- class mlprodict.npy.xop_auto_import_.OnnxFloor(*args, **kwargs)#
Version
name: Floor (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Floor takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the floor is, y = floor(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxFloor_1#
- class mlprodict.npy.xop_auto_import_.OnnxFloor_1(*args, **kwargs)#
Version
name: Floor (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Floor takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the floor is, y = floor(x), is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxFloor_13#
- class mlprodict.npy.xop_auto_import_.OnnxFloor_13(*args, **kwargs)#
Version
name: Floor (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Floor takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the floor is, y = floor(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxFloor_6#
- class mlprodict.npy.xop_auto_import_.OnnxFloor_6(*args, **kwargs)#
Version
name: Floor (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Floor takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the floor is, y = floor(x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGRU#
- class mlprodict.npy.xop_auto_import_.OnnxGRU(*args, **kwargs)#
Version
name: GRU (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Computes an one-layer GRU. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
z - update gate
r - reset gate
h - hidden gate
t - time step (t-1 means previous time step)
W[zrh] - W parameter weight matrix for update, reset, and hidden gates
R[zrh] - R recurrence weight matrix for update, reset, and hidden gates
Wb[zrh] - W bias vectors for update, reset, and hidden gates
Rb[zrh] - R bias vectors for update, reset, and hidden gates
WB[zrh] - W parameter weight matrix for backward update, reset, and hidden gates
RB[zrh] - R recurrence weight matrix for backward update, reset, and hidden gates
WBb[zrh] - W bias vectors for backward update, reset, and hidden gates
RBb[zrh] - R bias vectors for backward update, reset, and hidden gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh):
zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz)
rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr)
ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0
ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0
Ht = (1 - zt) (.) ht + zt (.) Ht-1
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
layout: The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size]. Default value is
0.linear_before_reset: When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[zrh] and WB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[zrh] and RB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for the gates. Concatenation of [Wb[zrh], Rb[zrh]] and [WBb[zrh], RBb[zrh]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 6*hidden_size]. Optional: If not specified - assumed to be 0
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxGRU_1#
- class mlprodict.npy.xop_auto_import_.OnnxGRU_1(*args, **kwargs)#
Version
name: GRU (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Computes an one-layer GRU. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
z - update gate
r - reset gate
h - hidden gate
t - time step (t-1 means previous time step)
W[zrh] - W parameter weight matrix for update, reset, and hidden gates
R[zrh] - R recurrence weight matrix for update, reset, and hidden gates
Wb[zrh] - W bias vectors for update, reset, and hidden gates
Rb[zrh] - R bias vectors for update, reset, and hidden gates
WB[zrh] - W parameter weight matrix for backward update, reset, and hidden gates
RB[zrh] - R recurrence weight matrix for backward update, reset, and hidden gates
WBb[zrh] - W bias vectors for backward update, reset, and hidden gates
RBb[zrh] - R bias vectors for backward update, reset, and hidden gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh):
zt = f(Xt*(Wz^T) + Ht-1*Rz + Wbz + Rbz)
rt = f(Xt*(Wr^T) + Ht-1*Rr + Wbr + Rbr)
ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*Rh + Rbh + Wbh) # default, when linear_before_reset = 0
ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*Rh + Rbh) + Wbh) # when linear_before_reset != 0
Ht = (1 - zt) (.) ht + zt (.) Ht-1
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM.
activations: A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'foward'.hidden_size: Number of neurons in the hidden layer
output_sequence: The sequence output for the hidden is optional if 0. Default 0. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[zrh] and WB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[zrh] and RB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for the gates. Concatenation of [Wb[zrh], Rb[zrh]] and [WBb[zrh], RBb[zrh]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 6*hidden_size]. Optional: If not specified - assumed to be 0
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]. It is optional if output_sequence is 0.
Y_h (heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxGRU_14#
- class mlprodict.npy.xop_auto_import_.OnnxGRU_14(*args, **kwargs)#
Version
name: GRU (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Computes an one-layer GRU. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
z - update gate
r - reset gate
h - hidden gate
t - time step (t-1 means previous time step)
W[zrh] - W parameter weight matrix for update, reset, and hidden gates
R[zrh] - R recurrence weight matrix for update, reset, and hidden gates
Wb[zrh] - W bias vectors for update, reset, and hidden gates
Rb[zrh] - R bias vectors for update, reset, and hidden gates
WB[zrh] - W parameter weight matrix for backward update, reset, and hidden gates
RB[zrh] - R recurrence weight matrix for backward update, reset, and hidden gates
WBb[zrh] - W bias vectors for backward update, reset, and hidden gates
RBb[zrh] - R bias vectors for backward update, reset, and hidden gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh):
zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz)
rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr)
ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0
ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0
Ht = (1 - zt) (.) ht + zt (.) Ht-1
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
layout: The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size]. Default value is
0.linear_before_reset: When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[zrh] and WB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[zrh] and RB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for the gates. Concatenation of [Wb[zrh], Rb[zrh]] and [WBb[zrh], RBb[zrh]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 6*hidden_size]. Optional: If not specified - assumed to be 0
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxGRU_3#
- class mlprodict.npy.xop_auto_import_.OnnxGRU_3(*args, **kwargs)#
Version
name: GRU (GitHub)
domain: main
since_version: 3
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 3.
Summary
Computes an one-layer GRU. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
z - update gate
r - reset gate
h - hidden gate
t - time step (t-1 means previous time step)
W[zrh] - W parameter weight matrix for update, reset, and hidden gates
R[zrh] - R recurrence weight matrix for update, reset, and hidden gates
Wb[zrh] - W bias vectors for update, reset, and hidden gates
Rb[zrh] - R bias vectors for update, reset, and hidden gates
WB[zrh] - W parameter weight matrix for backward update, reset, and hidden gates
RB[zrh] - R recurrence weight matrix for backward update, reset, and hidden gates
WBb[zrh] - W bias vectors for backward update, reset, and hidden gates
RBb[zrh] - R bias vectors for backward update, reset, and hidden gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh):
zt = f(Xt*(Wz^T) + Ht-1*Rz + Wbz + Rbz)
rt = f(Xt*(Wr^T) + Ht-1*Rr + Wbr + Rbr)
ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*Rh + Rbh + Wbh) # default, when linear_before_reset = 0
ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*Rh + Rbh) + Wbh) # when linear_before_reset != 0
Ht = (1 - zt) (.) ht + zt (.) Ht-1
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
linear_before_reset: When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate. Default value is
0.output_sequence: The sequence output for the hidden is optional if 0. Default 0. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[zrh] and WB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[zrh] and RB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for the gates. Concatenation of [Wb[zrh], Rb[zrh]] and [WBb[zrh], RBb[zrh]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 6*hidden_size]. Optional: If not specified - assumed to be 0
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]. It is optional if output_sequence is 0.
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxGRU_7#
- class mlprodict.npy.xop_auto_import_.OnnxGRU_7(*args, **kwargs)#
Version
name: GRU (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Computes an one-layer GRU. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
z - update gate
r - reset gate
h - hidden gate
t - time step (t-1 means previous time step)
W[zrh] - W parameter weight matrix for update, reset, and hidden gates
R[zrh] - R recurrence weight matrix for update, reset, and hidden gates
Wb[zrh] - W bias vectors for update, reset, and hidden gates
Rb[zrh] - R bias vectors for update, reset, and hidden gates
WB[zrh] - W parameter weight matrix for backward update, reset, and hidden gates
RB[zrh] - R recurrence weight matrix for backward update, reset, and hidden gates
WBb[zrh] - W bias vectors for backward update, reset, and hidden gates
RBb[zrh] - R bias vectors for backward update, reset, and hidden gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh):
zt = f(Xt*(Wz^T) + Ht-1*(Rz^T) + Wbz + Rbz)
rt = f(Xt*(Wr^T) + Ht-1*(Rr^T) + Wbr + Rbr)
ht = g(Xt*(Wh^T) + (rt (.) Ht-1)*(Rh^T) + Rbh + Wbh) # default, when linear_before_reset = 0
ht = g(Xt*(Wh^T) + (rt (.) (Ht-1*(Rh^T) + Rbh)) + Wbh) # when linear_before_reset != 0
Ht = (1 - zt) (.) ht + zt (.) Ht-1
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 2 (or 4 if bidirectional) activation functions for update, reset, and hidden gates. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
linear_before_reset: When computing the output of the hidden gate, apply the linear transformation before multiplying by the output of the reset gate. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[zrh] and WB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[zrh] and RB[zrh] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 3*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for the gates. Concatenation of [Wb[zrh], Rb[zrh]] and [WBb[zrh], RBb[zrh]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 6*hidden_size]. Optional: If not specified - assumed to be 0
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxGather#
- class mlprodict.npy.xop_auto_import_.OnnxGather(*args, **kwargs)#
Version
name: Gather (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q, gather entries of the axis dimension of data (by default outer-most one as axis=0) indexed by indices, and concatenates them in an output tensor of rank q + (r - 1).
axis = 0 :
Let k = indices[i_{0}, …, i_{q-1}] Then output[i_{0}, …, i_{q-1}, j_{0}, …, j_{r-2}] = input[k , j_{0}, …, j_{r-2}]
data = [ [1.0, 1.2], [2.3, 3.4], [4.5, 5.7], ] indices = [ [0, 1], [1, 2], ] output = [ [ [1.0, 1.2], [2.3, 3.4], ], [ [2.3, 3.4], [4.5, 5.7], ], ]
axis = 1 :
Let k = indices[i_{0}, …, i_{q-1}] Then output[j_{0}, i_{0}, …, i_{q-1}, j_{1}, …, j_{r-2}] = input[j_{0}, k, j_{1}, …, j_{r-2}]
data = [ [1.0, 1.2, 1.9], [2.3, 3.4, 3.9], [4.5, 5.7, 5.9], ] indices = [ [0, 2], ] axis = 1, output = [ [[1.0, 1.9]], [[2.3, 3.9]], [[4.5, 5.9]], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + (r - 1).
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGatherElements#
- class mlprodict.npy.xop_auto_import_.OnnxGatherElements(*args, **kwargs)#
Version
name: GatherElements (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
GatherElements takes two inputs data and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). It is an indexing operation that produces its output by indexing into the input data tensor at index positions determined by elements of the indices tensor. Its output shape is the same as the shape of indices and consists of one value (gathered from the data) for each element in indices.
For instance, in the 3-D case (r = 3), the output produced is determined by the following equations:
out[i][j][k] = input[index[i][j][k]][j][k] if axis = 0, out[i][j][k] = input[i][index[i][j][k]][k] if axis = 1, out[i][j][k] = input[i][j][index[i][j][k]] if axis = 2,
This operator is also the inverse of ScatterElements. It is similar to Torch’s gather operation.
Example 1:
data = [ [1, 2], [3, 4], ] indices = [ [0, 0], [1, 0], ] axis = 1 output = [ [1, 1], [4, 3], ]
Example 2:
data = [ [1, 2, 3], [4, 5, 6], [7, 8, 9], ] indices = [ [1, 2, 0], [2, 0, 0], ] axis = 0 output = [ [4, 8, 3], [7, 2, 3], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of the same shape as indices.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGatherElements_11#
- class mlprodict.npy.xop_auto_import_.OnnxGatherElements_11(*args, **kwargs)#
Version
name: GatherElements (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
GatherElements takes two inputs data and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). It is an indexing operation that produces its output by indexing into the input data tensor at index positions determined by elements of the indices tensor. Its output shape is the same as the shape of indices and consists of one value (gathered from the data) for each element in indices.
For instance, in the 3-D case (r = 3), the output produced is determined by the following equations:
out[i][j][k] = input[index[i][j][k]][j][k] if axis = 0, out[i][j][k] = input[i][index[i][j][k]][k] if axis = 1, out[i][j][k] = input[i][j][index[i][j][k]] if axis = 2,
This operator is also the inverse of ScatterElements. It is similar to Torch’s gather operation.
Example 1:
data = [ [1, 2], [3, 4], ] indices = [ [0, 0], [1, 0], ] axis = 1 output = [ [ [1, 1], [4, 3], ], ]
Example 2:
data = [ [1, 2, 3], [4, 5, 6], [7, 8, 9], ] indices = [ [1, 2, 0], [2, 0, 0], ] axis = 0 output = [ [ [4, 8, 3], [7, 2, 3], ], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of the same shape as indices.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGatherElements_13#
- class mlprodict.npy.xop_auto_import_.OnnxGatherElements_13(*args, **kwargs)#
Version
name: GatherElements (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
GatherElements takes two inputs data and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). It is an indexing operation that produces its output by indexing into the input data tensor at index positions determined by elements of the indices tensor. Its output shape is the same as the shape of indices and consists of one value (gathered from the data) for each element in indices.
For instance, in the 3-D case (r = 3), the output produced is determined by the following equations:
out[i][j][k] = input[index[i][j][k]][j][k] if axis = 0, out[i][j][k] = input[i][index[i][j][k]][k] if axis = 1, out[i][j][k] = input[i][j][index[i][j][k]] if axis = 2,
This operator is also the inverse of ScatterElements. It is similar to Torch’s gather operation.
Example 1:
data = [ [1, 2], [3, 4], ] indices = [ [0, 0], [1, 0], ] axis = 1 output = [ [1, 1], [4, 3], ]
Example 2:
data = [ [1, 2, 3], [4, 5, 6], [7, 8, 9], ] indices = [ [1, 2, 0], [2, 0, 0], ] axis = 0 output = [ [4, 8, 3], [7, 2, 3], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, with the same rank r as the input. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of the same shape as indices.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGatherND#
- class mlprodict.npy.xop_auto_import_.OnnxGatherND(*args, **kwargs)#
Version
name: GatherND (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Given data tensor of rank r >= 1, indices tensor of rank q >= 1, and batch_dims integer b, this operator gathers slices of data into an output tensor of rank q + r - indices_shape[-1] - 1 - b.
indices is an q-dimensional integer tensor, best thought of as a (q-1)-dimensional tensor of index-tuples into data, where each element defines a slice of data
batch_dims (denoted as b) is an integer indicating the number of batch dimensions, i.e the leading b number of dimensions of data tensor and indices are representing the batches, and the gather starts from the b+1 dimension.
Some salient points about the inputs’ rank and shape:
r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks r and q
The first b dimensions of the shape of indices tensor and data tensor must be equal.
b < min(q, r) is to be honored.
The indices_shape[-1] should have a value between 1 (inclusive) and rank r-b (inclusive)
All values in indices are expected to be within bounds [-s, s-1] along axis of size s (i.e.) -data_shape[i] <= indices[…,i] <= data_shape[i] - 1. It is an error if any of the index values are out of bounds.
The output is computed as follows:
The output tensor is obtained by mapping each index-tuple in the indices tensor to the corresponding slice of the input data.
If indices_shape[-1] > r-b => error condition
If indices_shape[-1] == r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensors containing 1-D tensors of dimension r-b, where N is an integer equals to the product of 1 and all the elements in the batch dimensions of the indices_shape. Let us think of each such r-b ranked tensor as indices_slice. Each scalar value corresponding to data[0:b-1,indices_slice] is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Example 1 below)
If indices_shape[-1] < r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensor containing 1-D tensors of dimension < r-b. Let us think of each such tensors as indices_slice. Each tensor slice corresponding to data[0:b-1, indices_slice , :] is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Examples 2, 3, 4 and 5 below)
This operator is the inverse of ScatterND.
Example 1
batch_dims = 0
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[0,0],[1,1]] # indices_shape = [2, 2]
output = [0,3] # output_shape = [2]
Example 2
batch_dims = 0
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[0,1]] # output_shape = [2, 2]
Example 3
batch_dims = 0
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[0,1],[1,0]] # indices_shape = [2, 2]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Example 4
batch_dims = 0
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2]
output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2]
Example 5
batch_dims = 1
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Attributes
batch_dims: The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:] Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxGatherND_11#
- class mlprodict.npy.xop_auto_import_.OnnxGatherND_11(*args, **kwargs)#
Version
name: GatherND (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q >= 1, this operator gathers slices of data into an output tensor of rank q + r - indices_shape[-1] - 1.
indices is an q-dimensional integer tensor, best thought of as a (q-1)-dimensional tensor of index-tuples into data, where each element defines a slice of data
Some salient points about the inputs’ rank and shape:
r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks r and q
The indices_shape[-1] should have a value between 1 (inclusive) and rank r (inclusive)
All values in indices are expected to be within bounds [-s, s-1] along axis of size s (i.e.) -data_shape[i] <= indices[…,i] <= data_shape[i] - 1. It is an error if any of the index values are out of bounds.
The output is computed as follows:
The output tensor is obtained by mapping each index-tuple in the indices tensor to the corresponding slice of the input data.
If indices_shape[-1] > r => error condition
If indices_shape[-1] == r, since the rank of indices is q, indices can be thought of as a (q-1)-dimensional tensor containing 1-D tensors of dimension r. Let us think of each such r ranked tensor as indices_slice. Each scalar value corresponding to data[indices_slice] is filled into the corresponding location of the (q-1)-dimensional tensor to form the output tensor (Example 1 below)
If indices_shape[-1] < r, since the rank of indices is q, indices can be thought of as a (q-1)-dimensional tensor containing 1-D tensors of dimension < r. Let us think of each such tensors as indices_slice. Each tensor slice corresponding to data[indices_slice , :] is filled into the corresponding location of the (q-1)-dimensional tensor to form the output tensor (Examples 2, 3, and 4 below)
This operator is the inverse of ScatterND.
Example 1
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[0,0],[1,1]] # indices_shape = [2, 2]
output = [0,3] # output_shape = [2]
Example 2
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[0,1]] # output_shape = [2, 2]
Example 3
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[0,1],[1,0]] # indices_shape = [2, 2]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Example 4
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2]
output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2]
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxGatherND_12#
- class mlprodict.npy.xop_auto_import_.OnnxGatherND_12(*args, **kwargs)#
Version
name: GatherND (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Given data tensor of rank r >= 1, indices tensor of rank q >= 1, and batch_dims integer b, this operator gathers slices of data into an output tensor of rank q + r - indices_shape[-1] - 1 - b.
indices is an q-dimensional integer tensor, best thought of as a (q-1)-dimensional tensor of index-tuples into data, where each element defines a slice of data
batch_dims (denoted as b) is an integer indicating the number of batch dimensions, i.e the leading b number of dimensions of data tensor and indices are representing the batches, and the gather starts from the b+1 dimension.
Some salient points about the inputs’ rank and shape:
r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks r and q
The first b dimensions of the shape of indices tensor and data tensor must be equal.
b < min(q, r) is to be honored.
The indices_shape[-1] should have a value between 1 (inclusive) and rank r-b (inclusive)
All values in indices are expected to be within bounds [-s, s-1] along axis of size s (i.e.) -data_shape[i] <= indices[…,i] <= data_shape[i] - 1. It is an error if any of the index values are out of bounds.
The output is computed as follows:
The output tensor is obtained by mapping each index-tuple in the indices tensor to the corresponding slice of the input data.
If indices_shape[-1] > r-b => error condition
If indices_shape[-1] == r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensors containing 1-D tensors of dimension r-b, where N is an integer equals to the product of 1 and all the elements in the batch dimensions of the indices_shape. Let us think of each such r-b ranked tensor as indices_slice. Each scalar value corresponding to data[0:b-1,indices_slice] is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Example 1 below)
If indices_shape[-1] < r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensor containing 1-D tensors of dimension < r-b. Let us think of each such tensors as indices_slice. Each tensor slice corresponding to data[0:b-1, indices_slice , :] is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Examples 2, 3, 4 and 5 below)
This operator is the inverse of ScatterND.
Example 1
batch_dims = 0
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[0,0],[1,1]] # indices_shape = [2, 2]
output = [0,3] # output_shape = [2]
Example 2
batch_dims = 0
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[0,1]] # output_shape = [2, 2]
Example 3
batch_dims = 0
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[0,1],[1,0]] # indices_shape = [2, 2]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Example 4
batch_dims = 0
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2]
output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2]
Example 5
batch_dims = 1
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Attributes
batch_dims: The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:] Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxGatherND_13#
- class mlprodict.npy.xop_auto_import_.OnnxGatherND_13(*args, **kwargs)#
Version
name: GatherND (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Given data tensor of rank r >= 1, indices tensor of rank q >= 1, and batch_dims integer b, this operator gathers slices of data into an output tensor of rank q + r - indices_shape[-1] - 1 - b.
indices is an q-dimensional integer tensor, best thought of as a (q-1)-dimensional tensor of index-tuples into data, where each element defines a slice of data
batch_dims (denoted as b) is an integer indicating the number of batch dimensions, i.e the leading b number of dimensions of data tensor and indices are representing the batches, and the gather starts from the b+1 dimension.
Some salient points about the inputs’ rank and shape:
r >= 1 and q >= 1 are to be honored. There is no dependency condition to be met between ranks r and q
The first b dimensions of the shape of indices tensor and data tensor must be equal.
b < min(q, r) is to be honored.
The indices_shape[-1] should have a value between 1 (inclusive) and rank r-b (inclusive)
All values in indices are expected to be within bounds [-s, s-1] along axis of size s (i.e.) -data_shape[i] <= indices[…,i] <= data_shape[i] - 1. It is an error if any of the index values are out of bounds.
The output is computed as follows:
The output tensor is obtained by mapping each index-tuple in the indices tensor to the corresponding slice of the input data.
If indices_shape[-1] > r-b => error condition
If indices_shape[-1] == r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensors containing 1-D tensors of dimension r-b, where N is an integer equals to the product of 1 and all the elements in the batch dimensions of the indices_shape. Let us think of each such r-b ranked tensor as indices_slice. Each scalar value corresponding to data[0:b-1,indices_slice] is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Example 1 below)
If indices_shape[-1] < r-b, since the rank of indices is q, indices can be thought of as N (q-b-1)-dimensional tensor containing 1-D tensors of dimension < r-b. Let us think of each such tensors as indices_slice. Each tensor slice corresponding to data[0:b-1, indices_slice , :] is filled into the corresponding location of the (q-b-1)-dimensional tensor to form the output tensor (Examples 2, 3, 4 and 5 below)
This operator is the inverse of ScatterND.
Example 1
batch_dims = 0
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[0,0],[1,1]] # indices_shape = [2, 2]
output = [0,3] # output_shape = [2]
Example 2
batch_dims = 0
data = [[0,1],[2,3]] # data_shape = [2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[0,1]] # output_shape = [2, 2]
Example 3
batch_dims = 0
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[0,1],[1,0]] # indices_shape = [2, 2]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Example 4
batch_dims = 0
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[[0,1]],[[1,0]]] # indices_shape = [2, 1, 2]
output = [[[2,3]],[[4,5]]] # output_shape = [2, 1, 2]
Example 5
batch_dims = 1
data = [[[0,1],[2,3]],[[4,5],[6,7]]] # data_shape = [2, 2, 2]
indices = [[1],[0]] # indices_shape = [2, 1]
output = [[2,3],[4,5]] # output_shape = [2, 2]
Attributes
batch_dims: The number of batch dimensions. The gather of indexing starts from dimension of data[batch_dims:] Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxGather_1#
- class mlprodict.npy.xop_auto_import_.OnnxGather_1(*args, **kwargs)#
Version
name: Gather (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q, gather entries of the axis dimension of data (by default outer-most one as axis=0) indexed by indices, and concatenates them in an output tensor of rank q + (r - 1). Example 1:
data = [ [1.0, 1.2], [2.3, 3.4], [4.5, 5.7], ] indices = [ [0, 1], [1, 2], ] output = [ [ [1.0, 1.2], [2.3, 3.4], ], [ [2.3, 3.4], [4.5, 5.7], ], ]
Example 2:
data = [ [1.0, 1.2, 1.9], [2.3, 3.4, 3.9], [4.5, 5.7, 5.9], ] indices = [ [0, 2], ] axis = 1, output = [ [ [1.0, 1.9], [2.3, 3.9], [4.5, 5.9], ], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + (r - 1).
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGather_11#
- class mlprodict.npy.xop_auto_import_.OnnxGather_11(*args, **kwargs)#
Version
name: Gather (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q, gather entries of the axis dimension of data (by default outer-most one as axis=0) indexed by indices, and concatenates them in an output tensor of rank q + (r - 1).
axis = 0 :
Let k = indices[i_{0}, …, i_{q-1}] Then output[i_{0}, …, i_{q-1}, j_{0}, …, j_{r-2}] = input[k , j_{0}, …, j_{r-2}]
data = [ [1.0, 1.2], [2.3, 3.4], [4.5, 5.7], ] indices = [ [0, 1], [1, 2], ] output = [ [ [1.0, 1.2], [2.3, 3.4], ], [ [2.3, 3.4], [4.5, 5.7], ], ]
axis = 1 :
Let k = indices[i_{0}, …, i_{q-1}] Then output[i_{0}, …, i_{q-1}, j_{0}, …, j_{r-2}] = input[j_{0}, k, j_{1}, …, j_{r-2}]
data = [ [1.0, 1.2, 1.9], [2.3, 3.4, 3.9], [4.5, 5.7, 5.9], ] indices = [ [0, 2], ] axis = 1, output = [ [ [1.0, 1.9], [2.3, 3.9], [4.5, 5.9], ], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + (r - 1).
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGather_13#
- class mlprodict.npy.xop_auto_import_.OnnxGather_13(*args, **kwargs)#
Version
name: Gather (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Given data tensor of rank r >= 1, and indices tensor of rank q, gather entries of the axis dimension of data (by default outer-most one as axis=0) indexed by indices, and concatenates them in an output tensor of rank q + (r - 1).
axis = 0 :
Let k = indices[i_{0}, …, i_{q-1}] Then output[i_{0}, …, i_{q-1}, j_{0}, …, j_{r-2}] = input[k , j_{0}, …, j_{r-2}]
data = [ [1.0, 1.2], [2.3, 3.4], [4.5, 5.7], ] indices = [ [0, 1], [1, 2], ] output = [ [ [1.0, 1.2], [2.3, 3.4], ], [ [2.3, 3.4], [4.5, 5.7], ], ]
axis = 1 :
Let k = indices[i_{0}, …, i_{q-1}] Then output[j_{0}, i_{0}, …, i_{q-1}, j_{1}, …, j_{r-2}] = input[j_{0}, k, j_{1}, …, j_{r-2}]
data = [ [1.0, 1.2, 1.9], [2.3, 3.4, 3.9], [4.5, 5.7, 5.9], ] indices = [ [0, 2], ] axis = 1, output = [ [[1.0, 1.9]], [[2.3, 3.9]], [[4.5, 5.9]], ]
Attributes
axis: Which axis to gather on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of any rank q. All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
Outputs
output (heterogeneous) - T: Tensor of rank q + (r - 1).
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxGemm#
- class mlprodict.npy.xop_auto_import_.OnnxGemm(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3
A’ = transpose(A) if transA else A
B’ = transpose(B) if transB else B
Compute Y = alpha * A’ * B’ + beta * C, where input tensor A has shape (M, K) or (K, M), input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), and output tensor Y has shape (M, N). A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB. This operator supports unidirectional broadcasting (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check Broadcasting in ONNX. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
1.0.beta: Scalar multiplier for input tensor C. Default value is
1.0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
Between 2 and 3 inputs.
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (optional, heterogeneous) - T: Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxGemm_1#
- class mlprodict.npy.xop_auto_import_.OnnxGemm_1(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 Compute Y = alpha * A * B + beta * C, where input tensor A has dimension (M X K), input tensor B has dimension (K X N), input tensor C and output tensor Y have dimension (M X N). If attribute broadcast is non-zero, input tensor C will be broadcasted to match the dimension requirement. A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B, the default value is 1.0. Default value is
1.0.beta: Scalar multiplier for input tensor C, the default value is 1.0. Default value is
1.0.broadcast: Whether C should be broadcasted Default value is
0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
A (heterogeneous) - T: Input tensor A
B (heterogeneous) - T: Input tensor B
C (heterogeneous) - T: Input tensor C, can be inplace.
Outputs
Y (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGemm_11#
- class mlprodict.npy.xop_auto_import_.OnnxGemm_11(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3
A’ = transpose(A) if transA else A
B’ = transpose(B) if transB else B
Compute Y = alpha * A’ * B’ + beta * C, where input tensor A has shape (M, K) or (K, M), input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), and output tensor Y has shape (M, N). A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB. This operator supports unidirectional broadcasting (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check Broadcasting in ONNX. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
1.0.beta: Scalar multiplier for input tensor C. Default value is
1.0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
Between 2 and 3 inputs.
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (optional, heterogeneous) - T: Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxGemm_13#
- class mlprodict.npy.xop_auto_import_.OnnxGemm_13(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3
A’ = transpose(A) if transA else A
B’ = transpose(B) if transB else B
Compute Y = alpha * A’ * B’ + beta * C, where input tensor A has shape (M, K) or (K, M), input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), and output tensor Y has shape (M, N). A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB. This operator supports unidirectional broadcasting (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check Broadcasting in ONNX. This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
1.0.beta: Scalar multiplier for input tensor C. Default value is
1.0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
Between 2 and 3 inputs.
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (optional, heterogeneous) - T: Optional input tensor C. If not specified, the computation is done as if C is a scalar 0. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxGemm_6#
- class mlprodict.npy.xop_auto_import_.OnnxGemm_6(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3 Compute Y = alpha * A * B + beta * C, where input tensor A has dimension (M X K), input tensor B has dimension (K X N), input tensor C and output tensor Y have dimension (M X N). If attribute broadcast is non-zero, input tensor C will be broadcasted to match the dimension requirement. A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B, the default value is 1.0. Default value is
1.0.beta: Scalar multiplier for input tensor C, the default value is 1.0. Default value is
1.0.broadcast: Whether C should be broadcasted Default value is
0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
A (heterogeneous) - T: Input tensor A
B (heterogeneous) - T: Input tensor B
C (heterogeneous) - T: Input tensor C
Outputs
Y (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGemm_7#
- class mlprodict.npy.xop_auto_import_.OnnxGemm_7(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3
A’ = transpose(A) if transA else A
B’ = transpose(B) if transB else B
Compute Y = alpha * A’ * B’ + beta * C, where input tensor A has shape (M, K) or (K, M), input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), and output tensor Y has shape (M, N). A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB. This operator supports unidirectional broadcasting (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check Broadcasting in ONNX.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
1.0.beta: Scalar multiplier for input tensor C. Default value is
1.0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (heterogeneous) - T: Input tensor C. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGemm_9#
- class mlprodict.npy.xop_auto_import_.OnnxGemm_9(*args, **kwargs)#
Version
name: Gemm (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
General Matrix multiplication: https://en.wikipedia.org/wiki/Basic_Linear_Algebra_Subprograms#Level_3
A’ = transpose(A) if transA else A
B’ = transpose(B) if transB else B
Compute Y = alpha * A’ * B’ + beta * C, where input tensor A has shape (M, K) or (K, M), input tensor B has shape (K, N) or (N, K), input tensor C is broadcastable to shape (M, N), and output tensor Y has shape (M, N). A will be transposed before doing the computation if attribute transA is non-zero, same for B and transB. This operator supports unidirectional broadcasting (tensor C should be unidirectional broadcastable to tensor A * B); for more details please check Broadcasting in ONNX.
Attributes
alpha: Scalar multiplier for the product of input tensors A * B. Default value is
1.0.beta: Scalar multiplier for input tensor C. Default value is
1.0.transA: Whether A should be transposed Default value is
0.transB: Whether B should be transposed Default value is
0.
Inputs
A (heterogeneous) - T: Input tensor A. The shape of A should be (M, K) if transA is 0, or (K, M) if transA is non-zero.
B (heterogeneous) - T: Input tensor B. The shape of B should be (K, N) if transB is 0, or (N, K) if transB is non-zero.
C (heterogeneous) - T: Input tensor C. The shape of C should be unidirectional broadcastable to (M, N).
Outputs
Y (heterogeneous) - T: Output tensor of shape (M, N).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxGlobalAveragePool#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalAveragePool(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
GlobalAveragePool consumes an input tensor X and applies average pooling across the values in the same channel. This is equivalent to AveragePool with kernel size equal to the spatial dimension of input tensor.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGlobalAveragePool_1#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalAveragePool_1(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
GlobalAveragePool consumes an input tensor X and applies average pooling across the values in the same channel. This is equivalent to AveragePool with kernel size equal to the spatial dimension of input tensor.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGlobalLpPool#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalLpPool(*args, **kwargs)#
Version
name: GlobalLpPool (GitHub)
domain: main
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2.
Summary
GlobalLpPool consumes an input tensor X and applies lp pool pooling across the values in the same channel. This is equivalent to LpPool with kernel size equal to the spatial dimension of input tensor.
Attributes
p: p value of the Lp norm used to pool over the input data. Default value is
2.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGlobalLpPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalLpPool_1(*args, **kwargs)#
Version
name: GlobalLpPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
GlobalLpPool consumes an input tensor X and applies lp pool pooling across the the values in the same channel. This is equivalent to LpPool with kernel size equal to the spatial dimension of input tensor.
Attributes
p: p value of the Lp norm used to pool over the input data, default is 2.0. Default value is
2.0.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimension are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. Dimensions will be N x C x 1 x 1
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGlobalLpPool_2#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalLpPool_2(*args, **kwargs)#
Version
name: GlobalLpPool (GitHub)
domain: main
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2.
Summary
GlobalLpPool consumes an input tensor X and applies lp pool pooling across the values in the same channel. This is equivalent to LpPool with kernel size equal to the spatial dimension of input tensor.
Attributes
p: p value of the Lp norm used to pool over the input data. Default value is
2.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGlobalMaxPool#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalMaxPool(*args, **kwargs)#
Version
name: GlobalMaxPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
GlobalMaxPool consumes an input tensor X and applies max pooling across the values in the same channel. This is equivalent to MaxPool with kernel size equal to the spatial dimension of input tensor.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGlobalMaxPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxGlobalMaxPool_1(*args, **kwargs)#
Version
name: GlobalMaxPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
GlobalMaxPool consumes an input tensor X and applies max pooling across the values in the same channel. This is equivalent to MaxPool with kernel size equal to the spatial dimension of input tensor.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from pooling across the input tensor. The output tensor has the same rank as the input. The first two dimensions of output shape are the same as the input (N x C), while the other dimensions are all 1.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGreater#
- class mlprodict.npy.xop_auto_import_.OnnxGreater(*args, **kwargs)#
Version
name: Greater (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the tensor resulted from performing the greater logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreaterOrEqual#
- class mlprodict.npy.xop_auto_import_.OnnxGreaterOrEqual(*args, **kwargs)#
Version
name: GreaterOrEqual (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 16.
Summary
Returns the tensor resulted from performing the greater_equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreaterOrEqual_12#
- class mlprodict.npy.xop_auto_import_.OnnxGreaterOrEqual_12(*args, **kwargs)#
Version
name: GreaterOrEqual (GitHub)
domain: main
since_version: 12
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 12.
Summary
Returns the tensor resulted from performing the greater_equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreaterOrEqual_16#
- class mlprodict.npy.xop_auto_import_.OnnxGreaterOrEqual_16(*args, **kwargs)#
Version
name: GreaterOrEqual (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 16.
Summary
Returns the tensor resulted from performing the greater_equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreater_1#
- class mlprodict.npy.xop_auto_import_.OnnxGreater_1(*args, **kwargs)#
Version
name: Greater (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the tensor resulted from performing the greater logical operation elementwise on the input tensors A and B.
If broadcasting is enabled, the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. See the doc of Add for a detailed description of the broadcasting rules.
Attributes
axis: If set, defines the broadcast dimensions.
broadcast: Enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: Left input tensor for the logical operator.
B (heterogeneous) - T: Right input tensor for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input to float tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreater_13#
- class mlprodict.npy.xop_auto_import_.OnnxGreater_13(*args, **kwargs)#
Version
name: Greater (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the tensor resulted from performing the greater logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreater_7#
- class mlprodict.npy.xop_auto_import_.OnnxGreater_7(*args, **kwargs)#
Version
name: Greater (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the greater logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input to float tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGreater_9#
- class mlprodict.npy.xop_auto_import_.OnnxGreater_9(*args, **kwargs)#
Version
name: Greater (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Returns the tensor resulted from performing the greater logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxGridSample#
- class mlprodict.npy.xop_auto_import_.OnnxGridSample(*args, **kwargs)#
Version
name: GridSample (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Given an input X and a flow-field grid, computes the output Y using X values and pixel locations from grid. Currently, only spatial (4-D) inputs are supported. For input X with shape (N, C, H, W) and grid with shape (N, H_out, W_out, 2), the output Y will have shape (N, C, H_out, W_out).
The tensor X contains values at centers of square pixels in a H by W 2-dimensional image. The tensor grid describes normalized positions where the output Y is to be computed using a specified interpolation method (the mode) and a padding mode (for grid positions falling outside the 2-dimensional image).
Elements in grid[N, H_out, W_out] are size-2 vectors specifying positions in the 2-dimensional space of X. They are used to interpolate output values of Y[N, C, H_out, W_out].
The GridSample operator is often used in doing grid generator and sampler in the [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/master/generated/torch.nn.functional.grid_sample.html#torch-nn-functional-grid-sample).
Attributes
align_corners: If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input’s corner pixels. If align_corners=0, they are instead considered as referring to the corner points of the input’s corner pixels, making the sampling more resolution agnostic. Default value is
0.mode: Three interpolation modes: bilinear (default), nearest and bicubic. Default value is
'bilinear'.padding_mode: Support padding modes for outside grid values: zeros`(default), `border, reflection. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x’ = 1.5, then reflects by border 1 and becomes x’’ = 0.5. Default value is
'zeros'.
Inputs
X (heterogeneous) - T1: 4-D tensor of shape (N, C, H, W), where N is the batch size, C is the numbers of channels, H and W are the height and width of the input data.
grid (heterogeneous) - T2: Input offset, 4-D tensor of shape (N, H_out, W_out, 2), where H_out and W_out are the height and width of grid and output, Grid specifies the sampling pixel locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode.
Outputs
Y (heterogeneous) - T1: 4-D tensor of shape (N, C, H_out, W_out) of sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input X and output Y types to all tensor types.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain grid types to float tensors.
OnnxGridSample_16#
- class mlprodict.npy.xop_auto_import_.OnnxGridSample_16(*args, **kwargs)#
Version
name: GridSample (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Given an input X and a flow-field grid, computes the output Y using X values and pixel locations from grid. Currently, only spatial (4-D) inputs are supported. For input X with shape (N, C, H, W) and grid with shape (N, H_out, W_out, 2), the output Y will have shape (N, C, H_out, W_out).
The tensor X contains values at centers of square pixels in a H by W 2-dimensional image. The tensor grid describes normalized positions where the output Y is to be computed using a specified interpolation method (the mode) and a padding mode (for grid positions falling outside the 2-dimensional image).
Elements in grid[N, H_out, W_out] are size-2 vectors specifying positions in the 2-dimensional space of X. They are used to interpolate output values of Y[N, C, H_out, W_out].
The GridSample operator is often used in doing grid generator and sampler in the [Spatial Transformer Networks](https://arxiv.org/abs/1506.02025). See also in [torch.nn.functional.grid_sample](https://pytorch.org/docs/master/generated/torch.nn.functional.grid_sample.html#torch-nn-functional-grid-sample).
Attributes
align_corners: If align_corners=1, the extrema (-1 and 1) are considered as referring to the center points of the input’s corner pixels. If align_corners=0, they are instead considered as referring to the corner points of the input’s corner pixels, making the sampling more resolution agnostic. Default value is
0.mode: Three interpolation modes: bilinear (default), nearest and bicubic. Default value is
'bilinear'.padding_mode: Support padding modes for outside grid values: zeros`(default), `border, reflection. zeros: use 0 for out-of-bound grid locations, border: use border values for out-of-bound grid locations, reflection: use values at locations reflected by the border for out-of-bound grid locations. If index 0 represents the margin pixel, the reflected value at index -1 will be the same as the value at index 1. For location far away from the border, it will keep being reflected until becoming in bound. If pixel location x = -3.5 reflects by border -1 and becomes x’ = 1.5, then reflects by border 1 and becomes x’’ = 0.5. Default value is
'zeros'.
Inputs
X (heterogeneous) - T1: 4-D tensor of shape (N, C, H, W), where N is the batch size, C is the numbers of channels, H and W are the height and width of the input data.
grid (heterogeneous) - T2: Input offset, 4-D tensor of shape (N, H_out, W_out, 2), where H_out and W_out are the height and width of grid and output, Grid specifies the sampling pixel locations normalized by the input spatial dimensions. Therefore, it should have most values in the range of [-1, 1]. If grid has values outside the range of [-1, 1], the corresponding outputs will be handled as defined by padding_mode.
Outputs
Y (heterogeneous) - T1: 4-D tensor of shape (N, C, H_out, W_out) of sampled values. For integer input types, intermediate values are computed as floating point and cast to integer at the end.
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input X and output Y types to all tensor types.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain grid types to float tensors.
OnnxGroupNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxGroupNormalization(*args, **kwargs)#
Version
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 18.
Summary
A GroupNormalization function. Carries out group normalization as described in the paper https://arxiv.org/abs/1803.08494
This operator transforms input according to
y = scale * (x - mean) / sqrt(variance + epsilon) + bias,
where the mean and variance are computed per instance per group of channels, and scale and bias should be specified for each group of channels. The number of groups num_groups should be divisible by the number of channels so that there are an equal number of channels per group.
When the number of groups is the same as the number of channels, this operator is equivalent to InstanceNormalization. When there is only one group, this operator is equivalent to LayerNormalization.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.num_groups (required): The number of groups of channels. It should be a divisor of the number of channels C.
Inputs
X (heterogeneous) - T: Input data tensor. Dimensions for image cases are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width of the data. Statistics are computed for every group of channels over C, H, and W. For non-image cases, the dimensions are in the form of (N x C x D1 x D2 … Dn).
scale (heterogeneous) - T: Scale tensor of shape (num_groups).
bias (heterogeneous) - T: Bias tensor of shape (num_groups).
Outputs
Y (heterogeneous) - T: The output tensor of the same shape as X.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxGroupNormalization_18#
- class mlprodict.npy.xop_auto_import_.OnnxGroupNormalization_18(*args, **kwargs)#
Version
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 18.
Summary
A GroupNormalization function. Carries out group normalization as described in the paper https://arxiv.org/abs/1803.08494
This operator transforms input according to
y = scale * (x - mean) / sqrt(variance + epsilon) + bias,
where the mean and variance are computed per instance per group of channels, and scale and bias should be specified for each group of channels. The number of groups num_groups should be divisible by the number of channels so that there are an equal number of channels per group.
When the number of groups is the same as the number of channels, this operator is equivalent to InstanceNormalization. When there is only one group, this operator is equivalent to LayerNormalization.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.num_groups (required): The number of groups of channels. It should be a divisor of the number of channels C.
Inputs
X (heterogeneous) - T: Input data tensor. Dimensions for image cases are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width of the data. Statistics are computed for every group of channels over C, H, and W. For non-image cases, the dimensions are in the form of (N x C x D1 x D2 … Dn).
scale (heterogeneous) - T: Scale tensor of shape (num_groups).
bias (heterogeneous) - T: Bias tensor of shape (num_groups).
Outputs
Y (heterogeneous) - T: The output tensor of the same shape as X.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHammingWindow#
- class mlprodict.npy.xop_auto_import_.OnnxHammingWindow(*args, **kwargs)#
Version
name: HammingWindow (GitHub)
domain: main
since_version: 17
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generates a Hamming window as described in the paper https://ieeexplore.ieee.org/document/1455106.
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT. Default value is
1.periodic: If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1. Default value is
1.
Inputs
size (heterogeneous) - T1: A scalar value indicating the length of the window.
Outputs
output (heterogeneous) - T2: A Hamming window with length: size. The output has the shape: [size].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain the input size to int64_t.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to numeric tensors.
OnnxHammingWindow_17#
- class mlprodict.npy.xop_auto_import_.OnnxHammingWindow_17(*args, **kwargs)#
Version
name: HammingWindow (GitHub)
domain: main
since_version: 17
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generates a Hamming window as described in the paper https://ieeexplore.ieee.org/document/1455106.
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT. Default value is
1.periodic: If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1. Default value is
1.
Inputs
size (heterogeneous) - T1: A scalar value indicating the length of the window.
Outputs
output (heterogeneous) - T2: A Hamming window with length: size. The output has the shape: [size].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain the input size to int64_t.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to numeric tensors.
OnnxHannWindow#
- class mlprodict.npy.xop_auto_import_.OnnxHannWindow(*args, **kwargs)#
Version
name: HannWindow (GitHub)
domain: main
since_version: 17
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generates a Hann window as described in the paper https://ieeexplore.ieee.org/document/1455106.
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT. Default value is
1.periodic: If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1. Default value is
1.
Inputs
size (heterogeneous) - T1: A scalar value indicating the length of the window.
Outputs
output (heterogeneous) - T2: A Hann window with length: size. The output has the shape: [size].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain the input size to int64_t.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to numeric tensors.
OnnxHannWindow_17#
- class mlprodict.npy.xop_auto_import_.OnnxHannWindow_17(*args, **kwargs)#
Version
name: HannWindow (GitHub)
domain: main
since_version: 17
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generates a Hann window as described in the paper https://ieeexplore.ieee.org/document/1455106.
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T2. The default value is 1 = FLOAT. Default value is
1.periodic: If 1, returns a window to be used as periodic function. If 0, return a symmetric window. When ‘periodic’ is specified, hann computes a window of length size + 1 and returns the first size points. The default value is 1. Default value is
1.
Inputs
size (heterogeneous) - T1: A scalar value indicating the length of the window.
Outputs
output (heterogeneous) - T2: A Hann window with length: size. The output has the shape: [size].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain the input size to int64_t.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output types to numeric tensors.
OnnxHardSigmoid#
- class mlprodict.npy.xop_auto_import_.OnnxHardSigmoid(*args, **kwargs)#
Version
name: HardSigmoid (GitHub)
domain: main
since_version: 6
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
HardSigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), is applied to the tensor elementwise.
Attributes
alpha: Value of alpha. Default value is
0.20000000298023224.beta: Value of beta. Default value is
0.5.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardSigmoid_1#
- class mlprodict.npy.xop_auto_import_.OnnxHardSigmoid_1(*args, **kwargs)#
Version
name: HardSigmoid (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
HardSigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), is applied to the tensor elementwise.
Attributes
alpha: Value of alpha default to 0.2 Default value is
0.20000000298023224.beta: Value of beta default to 0.5 Default value is
0.5.consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardSigmoid_6#
- class mlprodict.npy.xop_auto_import_.OnnxHardSigmoid_6(*args, **kwargs)#
Version
name: HardSigmoid (GitHub)
domain: main
since_version: 6
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
HardSigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the HardSigmoid function, y = max(0, min(1, alpha * x + beta)), is applied to the tensor elementwise.
Attributes
alpha: Value of alpha. Default value is
0.20000000298023224.beta: Value of beta. Default value is
0.5.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardSwish#
- class mlprodict.npy.xop_auto_import_.OnnxHardSwish(*args, **kwargs)#
Version
name: HardSwish (GitHub)
domain: main
since_version: 14
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
HardSwish takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the HardSwish function, y = x * max(0, min(1, alpha * x + beta)) = x * HardSigmoid<alpha, beta>(x), where alpha = 1/6 and beta = 0.5, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardSwish_14#
- class mlprodict.npy.xop_auto_import_.OnnxHardSwish_14(*args, **kwargs)#
Version
name: HardSwish (GitHub)
domain: main
since_version: 14
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
HardSwish takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the HardSwish function, y = x * max(0, min(1, alpha * x + beta)) = x * HardSigmoid<alpha, beta>(x), where alpha = 1/6 and beta = 0.5, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardmax#
- class mlprodict.npy.xop_auto_import_.OnnxHardmax(*args, **kwargs)#
Version
name: Hardmax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator computes the hardmax values for the given input:
Hardmax(element in input, axis) = 1 if the element is the first maximum value along the specified axis, 0 otherwise
The “axis” attribute indicates the dimension along which Hardmax will be performed. The output tensor has the same shape and contains the Hardmax values of the corresponding input.
Attributes
- axis:
Describes the dimension Hardmax will be performed on. Negative
value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.
Inputs
input (heterogeneous) - T: The input tensor of rank >= axis.
Outputs
output (heterogeneous) - T: The output values with the same shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardmax_1#
- class mlprodict.npy.xop_auto_import_.OnnxHardmax_1(*args, **kwargs)#
Version
name: Hardmax (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
- The operator computes the hardmax (1 for the first maximum value, and 0 for all others) values for each layer in the batch
of the given input. The input is a 2-D tensor (Tensor<float>) of size
(batch_size x input_feature_dimensions). The output tensor has the same shape and contains the hardmax values of the corresponding input.
Input does not need to explicitly be a 2D vector; rather, it will be coerced into one. For an arbitrary n-dimensional tensor input in [a_0, a_1, …, a_{k-1}, a_k, …, a_{n-1}] and k is the axis provided, then input will be coerced into a 2-dimensional tensor with dimensions [a_0 * … * a_{k-1}, a_k * … * a_{n-1}]. For the default case where axis=1, this means the input tensor will be coerced into a 2D tensor of dimensions [a_0, a_1 * … * a_{n-1}], where a_0 is often the batch size. In this situation, we must have a_0 = N and a_1 * … * a_{n-1} = D. Each of these dimensions must be matched correctly, or else the operator will throw errors.
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
1.
Inputs
input (heterogeneous) - T: The input tensor that’s coerced into a 2D matrix of size (NxD) as described above.
Outputs
output (heterogeneous) - T: The output values with the same shape as input tensor (the original size without coercion).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardmax_11#
- class mlprodict.npy.xop_auto_import_.OnnxHardmax_11(*args, **kwargs)#
Version
name: Hardmax (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- The operator computes the hardmax (1 for the first maximum value, and 0 for all others) values for each layer in the batch
of the given input.
The input does not need to explicitly be a 2D vector; rather, it will be coerced into one. For an arbitrary n-dimensional tensor input in [a_0, a_1, …, a_{k-1}, a_k, …, a_{n-1}] and k is the axis provided, then input will be coerced into a 2-dimensional tensor with dimensions [a_0 * … * a_{k-1}, a_k * … * a_{n-1}]. For the default case where axis=1, this means the input tensor will be coerced into a 2D tensor of dimensions [a_0, a_1 * … * a_{n-1}], where a_0 is often the batch size. In this situation, we must have a_0 = N and a_1 * … * a_{n-1} = D. Each of these dimensions must be matched correctly, or else the operator will throw errors. The output tensor has the same shape and contains the hardmax values of the corresponding input.
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
input (heterogeneous) - T: The input tensor that’s coerced into a 2D matrix of size (NxD) as described above.
Outputs
output (heterogeneous) - T: The output values with the same shape as input tensor (the original size without coercion).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxHardmax_13#
- class mlprodict.npy.xop_auto_import_.OnnxHardmax_13(*args, **kwargs)#
Version
name: Hardmax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator computes the hardmax values for the given input:
Hardmax(element in input, axis) = 1 if the element is the first maximum value along the specified axis, 0 otherwise
The “axis” attribute indicates the dimension along which Hardmax will be performed. The output tensor has the same shape and contains the Hardmax values of the corresponding input.
Attributes
- axis:
Describes the dimension Hardmax will be performed on. Negative
value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.
Inputs
input (heterogeneous) - T: The input tensor of rank >= axis.
Outputs
output (heterogeneous) - T: The output values with the same shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxIdentity#
- class mlprodict.npy.xop_auto_import_.OnnxIdentity(*args, **kwargs)#
Version
name: Identity (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Identity operator
Inputs
input (heterogeneous) - V: Input tensor
Outputs
output (heterogeneous) - V: Tensor to copy input into.
Type Constraints
V in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor, sequence, and optional types.
OnnxIdentity_1#
- class mlprodict.npy.xop_auto_import_.OnnxIdentity_1(*args, **kwargs)#
Version
name: Identity (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Identity operator
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: Tensor to copy input into.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxIdentity_13#
- class mlprodict.npy.xop_auto_import_.OnnxIdentity_13(*args, **kwargs)#
Version
name: Identity (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Identity operator
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: Tensor to copy input into.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxIdentity_14#
- class mlprodict.npy.xop_auto_import_.OnnxIdentity_14(*args, **kwargs)#
Version
name: Identity (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Identity operator
Inputs
input (heterogeneous) - V: Input tensor
Outputs
output (heterogeneous) - V: Tensor to copy input into.
Type Constraints
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor and sequence types.
OnnxIdentity_16#
- class mlprodict.npy.xop_auto_import_.OnnxIdentity_16(*args, **kwargs)#
Version
name: Identity (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Identity operator
Inputs
input (heterogeneous) - V: Input tensor
Outputs
output (heterogeneous) - V: Tensor to copy input into.
Type Constraints
V in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor, sequence, and optional types.
OnnxIf#
- class mlprodict.npy.xop_auto_import_.OnnxIf(*args, **kwargs)#
Version
name: If (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
If conditional
Attributes
else_branch (required): Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
then_branch (required): Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
Inputs
cond (heterogeneous) - B: Condition for the if
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - V: Values that are live-out to the enclosing scope. The return values in the then_branch and else_branch must be of the same data type. The then_branch and else_branch may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if- node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of then_branch is typed float tensor with shape [2] and the first output of else_branch is another float tensor with shape [3], If’s first output should have (a) no shape set, or (b) a shape of rank 1 with neither dim_value nor dim_param set, or (c) a shape of rank 1 with a unique dim_param. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
Type Constraints
V in ( optional(seq(tensor(bfloat16))), optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bfloat16)), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bfloat16)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types
B in ( tensor(bool) ): Only bool
OnnxIf_1#
- class mlprodict.npy.xop_auto_import_.OnnxIf_1(*args, **kwargs)#
Version
name: If (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
If conditional
Attributes
else_branch (required): Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
then_branch (required): Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
Inputs
cond (heterogeneous) - B: Condition for the if
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - V: Values that are live-out to the enclosing scope. The return values in the then_branch and else_branch must be of the same shape and same data type.
Type Constraints
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
B in ( tensor(bool) ): Only bool
OnnxIf_11#
- class mlprodict.npy.xop_auto_import_.OnnxIf_11(*args, **kwargs)#
Version
name: If (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
If conditional
Attributes
else_branch (required): Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
then_branch (required): Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
Inputs
cond (heterogeneous) - B: Condition for the if
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - V: Values that are live-out to the enclosing scope. The return values in the then_branch and else_branch must be of the same data type. The then_branch and else_branch may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if- node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of then_branch is typed float tensor with shape [2] and the first output of else_branch is another float tensor with shape [3], If’s first output should have (a) no shape set, or (b) a shape of rank 1 with neither dim_value nor dim_param set, or (c) a shape of rank 1 with a unique dim_param. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
Type Constraints
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
B in ( tensor(bool) ): Only bool
OnnxIf_13#
- class mlprodict.npy.xop_auto_import_.OnnxIf_13(*args, **kwargs)#
Version
name: If (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
If conditional
Attributes
else_branch (required): Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
then_branch (required): Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
Inputs
cond (heterogeneous) - B: Condition for the if
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - V: Values that are live-out to the enclosing scope. The return values in the then_branch and else_branch must be of the same data type. The then_branch and else_branch may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if- node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of then_branch is typed float tensor with shape [2] and the first output of else_branch is another float tensor with shape [3], If’s first output should have (a) no shape set, or (b) a shape of rank 1 with neither dim_value nor dim_param set, or (c) a shape of rank 1 with a unique dim_param. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
Type Constraints
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor and Sequence types
B in ( tensor(bool) ): Only bool
OnnxIf_16#
- class mlprodict.npy.xop_auto_import_.OnnxIf_16(*args, **kwargs)#
Version
name: If (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
If conditional
Attributes
else_branch (required): Graph to run if condition is false. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the then_branch.
then_branch (required): Graph to run if condition is true. Has N outputs: values you wish to be live-out to the enclosing scope. The number of outputs must match the number of outputs in the else_branch.
Inputs
cond (heterogeneous) - B: Condition for the if
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - V: Values that are live-out to the enclosing scope. The return values in the then_branch and else_branch must be of the same data type. The then_branch and else_branch may produce tensors with the same element type and different shapes. If corresponding outputs from the then-branch and the else-branch have static shapes S1 and S2, then the shape of the corresponding output variable of the if- node (if present) must be compatible with both S1 and S2 as it represents the union of both possible shapes.For example, if in a model file, the first output of then_branch is typed float tensor with shape [2] and the first output of else_branch is another float tensor with shape [3], If’s first output should have (a) no shape set, or (b) a shape of rank 1 with neither dim_value nor dim_param set, or (c) a shape of rank 1 with a unique dim_param. In contrast, the first output cannot have the shape [2] since [2] and [3] are not compatible.
Type Constraints
V in ( optional(seq(tensor(bfloat16))), optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bfloat16)), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bfloat16)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types
B in ( tensor(bool) ): Only bool
OnnxInstanceNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxInstanceNormalization(*args, **kwargs)#
Version
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Carries out instance normalization as described in the paper https://arxiv.org/abs/1607.08022.
y = scale * (x - mean) / sqrt(variance + epsilon) + B, where mean and variance are computed per instance per channel.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.
Inputs
input (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
scale (heterogeneous) - T: The input 1-dimensional scale tensor of size C.
B (heterogeneous) - T: The input 1-dimensional bias tensor of size C.
Outputs
output (heterogeneous) - T: The output tensor of the same shape as input.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxInstanceNormalization_1#
- class mlprodict.npy.xop_auto_import_.OnnxInstanceNormalization_1(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Carries out instance normalization as described in the paper https://arxiv.org/abs/1607.08022.
y = scale * (x - mean) / sqrt(variance + epsilon) + B, where mean and variance are computed per instance per channel.
Attributes
consumed_inputs: legacy optimization attribute.
epsilon: The epsilon value to use to avoid division by zero, default is 1e-5f. Default value is
9.999999747378752e-06.
Inputs
input (heterogeneous) - T: The input 4-dimensional tensor of shape NCHW.
scale (heterogeneous) - T: The input 1-dimensional scale tensor of size C.
B (heterogeneous) - T: The input 1-dimensional bias tensor of size C.
Outputs
output (heterogeneous) - T: The output 4-dimensional tensor of the same shape as input.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxInstanceNormalization_6#
- class mlprodict.npy.xop_auto_import_.OnnxInstanceNormalization_6(*args, **kwargs)#
Version
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Carries out instance normalization as described in the paper https://arxiv.org/abs/1607.08022.
y = scale * (x - mean) / sqrt(variance + epsilon) + B, where mean and variance are computed per instance per channel.
Attributes
epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.
Inputs
input (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
scale (heterogeneous) - T: The input 1-dimensional scale tensor of size C.
B (heterogeneous) - T: The input 1-dimensional bias tensor of size C.
Outputs
output (heterogeneous) - T: The output tensor of the same shape as input.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxIsInf#
- class mlprodict.npy.xop_auto_import_.OnnxIsInf(*args, **kwargs)#
Version
name: IsInf (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Map infinity to true and other values to false.
Attributes
detect_negative: (Optional) Whether map negative infinity to true. Default to 1 so that negative infinity induces true. Set this attribute to 0 if negative infinity should be mapped to false. Default value is
1.detect_positive: (Optional) Whether map positive infinity to true. Default to 1 so that positive infinity induces true. Set this attribute to 0 if positive infinity should be mapped to false. Default value is
1.
Inputs
X (heterogeneous) - T1: input
Outputs
Y (heterogeneous) - T2: output
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float tensors.
T2 in ( tensor(bool) ): Constrain output types to boolean tensors.
OnnxIsInf_10#
- class mlprodict.npy.xop_auto_import_.OnnxIsInf_10(*args, **kwargs)#
Version
name: IsInf (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Map infinity to true and other values to false.
Attributes
detect_negative: (Optional) Whether map negative infinity to true. Default to 1 so that negative infinity induces true. Set this attribute to 0 if negative infinity should be mapped to false. Default value is
1.detect_positive: (Optional) Whether map positive infinity to true. Default to 1 so that positive infinity induces true. Set this attribute to 0 if positive infinity should be mapped to false. Default value is
1.
Inputs
X (heterogeneous) - T1: input
Outputs
Y (heterogeneous) - T2: output
Type Constraints
T1 in ( tensor(double), tensor(float) ): Constrain input types to float tensors.
T2 in ( tensor(bool) ): Constrain output types to boolean tensors.
OnnxIsNaN#
- class mlprodict.npy.xop_auto_import_.OnnxIsNaN(*args, **kwargs)#
Version
name: IsNaN (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns which elements of the input are NaN.
Inputs
X (heterogeneous) - T1: input
Outputs
Y (heterogeneous) - T2: output
Type Constraints
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(bool) ): Constrain output types to boolean tensors.
OnnxIsNaN_13#
- class mlprodict.npy.xop_auto_import_.OnnxIsNaN_13(*args, **kwargs)#
Version
name: IsNaN (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns which elements of the input are NaN.
Inputs
X (heterogeneous) - T1: input
Outputs
Y (heterogeneous) - T2: output
Type Constraints
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(bool) ): Constrain output types to boolean tensors.
OnnxIsNaN_9#
- class mlprodict.npy.xop_auto_import_.OnnxIsNaN_9(*args, **kwargs)#
Version
name: IsNaN (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Returns which elements of the input are NaN.
Inputs
X (heterogeneous) - T1: input
Outputs
Y (heterogeneous) - T2: output
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(bool) ): Constrain output types to boolean tensors.
OnnxLRN#
- class mlprodict.npy.xop_auto_import_.OnnxLRN(*args, **kwargs)#
Version
name: LRN (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Local Response Normalization proposed in the [AlexNet paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf). It normalizes over local input regions. The local region is defined across the channels. For an element X[n, c, d1, …, dk] in a tensor of shape (N x C x D1 x D2, …, Dk), its region is {X[n, i, d1, …, dk] | max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))}.
square_sum[n, c, d1, …, dk] = sum(X[n, i, d1, …, dk] ^ 2), where max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2)).
Y[n, c, d1, …, dk] = X[n, c, d1, …, dk] / (bias + alpha / size * square_sum[n, c, d1, …, dk] ) ^ beta
Attributes
alpha: Scaling parameter. Default value is
9.999999747378752e-05.beta: The exponent. Default value is
0.75.bias:
Default value is
1.0.size (required): The number of channels to sum over
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output tensor, which has the shape and type as input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLRN_1#
- class mlprodict.npy.xop_auto_import_.OnnxLRN_1(*args, **kwargs)#
Version
name: LRN (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Local Response Normalization proposed in the [AlexNet paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf). It normalizes over local input regions. The local region is defined across the channels. For an element X[n, c, d1, …, dk] in a tensor of shape (N x C x D1 x D2, …, Dk), its region is {X[n, i, d1, …, dk] | max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))}.
square_sum[n, c, d1, …, dk] = sum(X[n, i, d1, …, dk] ^ 2), where max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2)).
Y[n, c, d1, …, dk] = X[n, c, d1, …, dk] / (bias + alpha / size * square_sum[n, c, d1, …, dk] ) ^ beta
Attributes
alpha: Scaling parameter. Default value is
9.999999747378752e-05.beta: The exponent. Default value is
0.75.bias:
Default value is
1.0.size (required): The number of channels to sum over
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output tensor, which has the shape and type as input tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLRN_13#
- class mlprodict.npy.xop_auto_import_.OnnxLRN_13(*args, **kwargs)#
Version
name: LRN (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Local Response Normalization proposed in the [AlexNet paper](https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks.pdf). It normalizes over local input regions. The local region is defined across the channels. For an element X[n, c, d1, …, dk] in a tensor of shape (N x C x D1 x D2, …, Dk), its region is {X[n, i, d1, …, dk] | max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2))}.
square_sum[n, c, d1, …, dk] = sum(X[n, i, d1, …, dk] ^ 2), where max(0, c - floor((size - 1) / 2)) <= i <= min(C - 1, c + ceil((size - 1) / 2)).
Y[n, c, d1, …, dk] = X[n, c, d1, …, dk] / (bias + alpha / size * square_sum[n, c, d1, …, dk] ) ^ beta
Attributes
alpha: Scaling parameter. Default value is
9.999999747378752e-05.beta: The exponent. Default value is
0.75.bias:
Default value is
1.0.size (required): The number of channels to sum over
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output tensor, which has the shape and type as input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLSTM#
- class mlprodict.npy.xop_auto_import_.OnnxLSTM(*args, **kwargs)#
Version
name: LSTM (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Computes an one-layer LSTM. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
o - output gate
f - forget gate
c - cell gate
t - time step (t-1 means previous time step)
W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
Wb[iofc] - W bias vectors for input, output, forget, and cell gates
Rb[iofc] - R bias vectors for input, output, forget, and cell gates
P[iof] - P peephole weight vector for input, output, and forget gates
WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
PB[iof] - P peephole weight vector for backward input, output, and forget gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
Ct = ft (.) Ct-1 + it (.) ct
ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo)
Ht = ot (.) h(Ct)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
input_forget: Couple the input and forget gates if 1. Default value is
0.layout: The shape format of inputs X, initial_h, initial_c and outputs Y, Y_h, Y_c. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [batch_size, num_directions, hidden_size]. Default value is
0.
Inputs
Between 3 and 8 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, 4*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 4*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxLSTM_1#
- class mlprodict.npy.xop_auto_import_.OnnxLSTM_1(*args, **kwargs)#
Version
name: LSTM (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes an one-layer LSTM. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
o - output gate
f - forget gate
c - cell gate
t - time step (t-1 means previous time step)
W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
Wb[iofc] - W bias vectors for input, output, forget, and cell gates
Rb[iofc] - R bias vectors for input, output, forget, and cell gates
P[iof] - P peephole weight vector for input, output, and forget gates
WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
PB[iof] - P peephole weight vector for backward input, output, and forget gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
it = f(Xt*(Wi^T) + Ht-1*Ri + Pi (.) Ct-1 + Wbi + Rbi)
ft = f(Xt*(Wf^T) + Ht-1*Rf + Pf (.) Ct-1 + Wbf + Rbf)
ct = g(Xt*(Wc^T) + Ht-1*Rc + Wbc + Rbc)
Ct = ft (.) Ct-1 + it (.) ct
ot = f(Xt*(Wo^T) + Ht-1*Ro + Po (.) Ct + Wbo + Rbo)
Ht = ot (.) h(Ct)
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
input_forget: Couple the input and forget gates if 1, default 0. Default value is
0.output_sequence: The sequence output for the hidden is optional if 0. Default 0. Default value is
0.
Inputs
Between 3 and 8 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, 4*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 4*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]. It is optional if output_sequence is 0.
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxLSTM_14#
- class mlprodict.npy.xop_auto_import_.OnnxLSTM_14(*args, **kwargs)#
Version
name: LSTM (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Computes an one-layer LSTM. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
o - output gate
f - forget gate
c - cell gate
t - time step (t-1 means previous time step)
W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
Wb[iofc] - W bias vectors for input, output, forget, and cell gates
Rb[iofc] - R bias vectors for input, output, forget, and cell gates
P[iof] - P peephole weight vector for input, output, and forget gates
WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
PB[iof] - P peephole weight vector for backward input, output, and forget gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
Ct = ft (.) Ct-1 + it (.) ct
ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo)
Ht = ot (.) h(Ct)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
input_forget: Couple the input and forget gates if 1. Default value is
0.layout: The shape format of inputs X, initial_h, initial_c and outputs Y, Y_h, Y_c. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = initial_c.shape = Y_c.shape = [batch_size, num_directions, hidden_size]. Default value is
0.
Inputs
Between 3 and 8 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, 4*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 4*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxLSTM_7#
- class mlprodict.npy.xop_auto_import_.OnnxLSTM_7(*args, **kwargs)#
Version
name: LSTM (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Computes an one-layer LSTM. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
o - output gate
f - forget gate
c - cell gate
t - time step (t-1 means previous time step)
W[iofc] - W parameter weight matrix for input, output, forget, and cell gates
R[iofc] - R recurrence weight matrix for input, output, forget, and cell gates
Wb[iofc] - W bias vectors for input, output, forget, and cell gates
Rb[iofc] - R bias vectors for input, output, forget, and cell gates
P[iof] - P peephole weight vector for input, output, and forget gates
WB[iofc] - W parameter weight matrix for backward input, output, forget, and cell gates
RB[iofc] - R recurrence weight matrix for backward input, output, forget, and cell gates
WBb[iofc] - W bias vectors for backward input, output, forget, and cell gates
RBb[iofc] - R bias vectors for backward input, output, forget, and cell gates
PB[iof] - P peephole weight vector for backward input, output, and forget gates
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Sigmoid, g=Tanh, h=Tanh):
it = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Pi (.) Ct-1 + Wbi + Rbi)
ft = f(Xt*(Wf^T) + Ht-1*(Rf^T) + Pf (.) Ct-1 + Wbf + Rbf)
ct = g(Xt*(Wc^T) + Ht-1*(Rc^T) + Wbc + Rbc)
Ct = ft (.) Ct-1 + it (.) ct
ot = f(Xt*(Wo^T) + Ht-1*(Ro^T) + Po (.) Ct + Wbo + Rbo)
Ht = ot (.) h(Ct)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: A list of 3 (or 6 if bidirectional) activation functions for input, output, forget, cell, and hidden. The activation functions must be one of the activation functions specified above. Optional: See the equations for default if not specified.
clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
input_forget: Couple the input and forget gates if 1. Default value is
0.
Inputs
Between 3 and 8 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for the gates. Concatenation of W[iofc] and WB[iofc] (if bidirectional) along dimension 0. The tensor has shape [num_directions, 4*hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of R[iofc] and RB[iofc] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 4*hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wb[iofc], Rb[iofc]], and [WBb[iofc], RBb[iofc]] (if bidirectional) along dimension 0. This tensor has shape [num_directions, 8*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
initial_c (optional, heterogeneous) - T: Optional initial value of the cell. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
P (optional, heterogeneous) - T: The weight tensor for peepholes. Concatenation of P[iof] and PB[iof] (if bidirectional) along dimension 0. It has shape [num_directions, 3*hidde_size]. Optional: If not specified - assumed to be 0.
Outputs
Between 0 and 3 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Y_c (optional, heterogeneous) - T: The last output value of the cell. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxLayerNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxLayerNormalization(*args, **kwargs)#
Version
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
This is layer normalization defined in ONNX as function. The overall computation can be split into two stages. The first stage is standardization, which makes the normalized elements have zero mean and unit variances. The computation required by standardization can be described by the following equations.
Mean = ReduceMean<axes=normalized_axes>(X) D = Sub(X, Mean) DD = Mul(D, D) Var = ReduceMean<axes=normalized_axes>(DD) VarEps = Add(Var, epsilon) StdDev = Sqrt(VarEps) InvStdDev = Reciprocal(StdDev) Normalized = Mul(D, InvStdDev)
where normalized_axes is [axis, …, rank of X - 1]. The variables Var and StdDev stand for variance and standard deviation, respectively. The second output is Mean and the last one is InvStdDev. Depending on stash_type attribute, the actual computation must happen in different floating-point precision. For example, if stash_type is 1, this operator casts all input variables to 32-bit float, perform the computation, and finally cast Normalized back to the original type of X. The second stage then scales and shifts the outcome of the first stage using
NormalizedScaled = Mul(Normalized, Scale) Y = Add(NormalizedScaled, B)
The second stage doesn’t depends on stash_type. All equations are in [this syntax](onnx/onnx). The same variable (i.e., input, output, and attribute) uses the same name in the equations above and this operator’s definition. Let d[i] indicate the i-th dimension of X. If X’s shape is [d[0], …, d[axis-1], d[axis], …, d[rank-1]], the shape of Mean and InvStdDev is [d[0], …, d[axis-1], 1, …, 1]. Y and X have the same shape.
Attributes
axis: The first normalization dimension. If rank(X) is r, axis’ allowed range is [-r, r]. Negative value means counting dimensions from the back. Default value is
-1.epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.stash_type: Type of Mean and InvStdDev. This also specifies stage one’s computation precision. Default value is
1.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Tensor to be normalized.
Scale (heterogeneous) - T: Scale tensor.
B (optional, heterogeneous) - T: Bias tensor.
Outputs
Between 1 and 3 outputs.
Y (heterogeneous) - T: Normalized tensor.
Mean (optional, heterogeneous) - U: Saved mean used during training to speed up gradient computation
InvStdDev (optional, heterogeneous) - U: Saved inverse standard deviation used during training to speed up gradient computation.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input types and output Y type to float tensors.
U in ( tensor(bfloat16), tensor(float) ): Type of Mean and InvStdDev tensors.
OnnxLayerNormalization_17#
- class mlprodict.npy.xop_auto_import_.OnnxLayerNormalization_17(*args, **kwargs)#
Version
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
This is layer normalization defined in ONNX as function. The overall computation can be split into two stages. The first stage is standardization, which makes the normalized elements have zero mean and unit variances. The computation required by standardization can be described by the following equations.
Mean = ReduceMean<axes=normalized_axes>(X) D = Sub(X, Mean) DD = Mul(D, D) Var = ReduceMean<axes=normalized_axes>(DD) VarEps = Add(Var, epsilon) StdDev = Sqrt(VarEps) InvStdDev = Reciprocal(StdDev) Normalized = Mul(D, InvStdDev)
where normalized_axes is [axis, …, rank of X - 1]. The variables Var and StdDev stand for variance and standard deviation, respectively. The second output is Mean and the last one is InvStdDev. Depending on stash_type attribute, the actual computation must happen in different floating-point precision. For example, if stash_type is 1, this operator casts all input variables to 32-bit float, perform the computation, and finally cast Normalized back to the original type of X. The second stage then scales and shifts the outcome of the first stage using
NormalizedScaled = Mul(Normalized, Scale) Y = Add(NormalizedScaled, B)
The second stage doesn’t depends on stash_type. All equations are in [this syntax](onnx/onnx). The same variable (i.e., input, output, and attribute) uses the same name in the equations above and this operator’s definition. Let d[i] indicate the i-th dimension of X. If X’s shape is [d[0], …, d[axis-1], d[axis], …, d[rank-1]], the shape of Mean and InvStdDev is [d[0], …, d[axis-1], 1, …, 1]. Y and X have the same shape.
Attributes
axis: The first normalization dimension. If rank(X) is r, axis’ allowed range is [-r, r]. Negative value means counting dimensions from the back. Default value is
-1.epsilon: The epsilon value to use to avoid division by zero. Default value is
9.999999747378752e-06.stash_type: Type of Mean and InvStdDev. This also specifies stage one’s computation precision. Default value is
1.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T: Tensor to be normalized.
Scale (heterogeneous) - T: Scale tensor.
B (optional, heterogeneous) - T: Bias tensor.
Outputs
Between 1 and 3 outputs.
Y (heterogeneous) - T: Normalized tensor.
Mean (optional, heterogeneous) - U: Saved mean used during training to speed up gradient computation
InvStdDev (optional, heterogeneous) - U: Saved inverse standard deviation used during training to speed up gradient computation.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input types and output Y type to float tensors.
U in ( tensor(bfloat16), tensor(float) ): Type of Mean and InvStdDev tensors.
OnnxLeakyRelu#
- class mlprodict.npy.xop_auto_import_.OnnxLeakyRelu(*args, **kwargs)#
Version
name: LeakyRelu (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
LeakyRelu takes input data (Tensor<T>) and an argument alpha, and produces one output data (Tensor<T>) where the function f(x) = alpha * x for x < 0, f(x) = x for x >= 0, is applied to the data tensor elementwise.
History - Version 16 adds bfloat16 to the types allowed.
Attributes
alpha: Coefficient of leakage. Default value is
0.009999999776482582.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLeakyRelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxLeakyRelu_1(*args, **kwargs)#
Version
name: LeakyRelu (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
LeakyRelu takes input data (Tensor<T>) and an argument alpha, and produces one output data (Tensor<T>) where the function f(x) = alpha * x for x < 0, f(x) = x for x >= 0, is applied to the data tensor elementwise.
Attributes
alpha: Coefficient of leakage default to 0.01. Default value is
0.009999999776482582.consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLeakyRelu_16#
- class mlprodict.npy.xop_auto_import_.OnnxLeakyRelu_16(*args, **kwargs)#
Version
name: LeakyRelu (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
LeakyRelu takes input data (Tensor<T>) and an argument alpha, and produces one output data (Tensor<T>) where the function f(x) = alpha * x for x < 0, f(x) = x for x >= 0, is applied to the data tensor elementwise.
History - Version 16 adds bfloat16 to the types allowed.
Attributes
alpha: Coefficient of leakage. Default value is
0.009999999776482582.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLeakyRelu_6#
- class mlprodict.npy.xop_auto_import_.OnnxLeakyRelu_6(*args, **kwargs)#
Version
name: LeakyRelu (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
LeakyRelu takes input data (Tensor<T>) and an argument alpha, and produces one output data (Tensor<T>) where the function f(x) = alpha * x for x < 0, f(x) = x for x >= 0, is applied to the data tensor elementwise.
Attributes
alpha: Coefficient of leakage. Default value is
0.009999999776482582.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLess#
- class mlprodict.npy.xop_auto_import_.OnnxLess(*args, **kwargs)#
Version
name: Less (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the tensor resulted from performing the less logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLessOrEqual#
- class mlprodict.npy.xop_auto_import_.OnnxLessOrEqual(*args, **kwargs)#
Version
name: LessOrEqual (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 16.
Summary
Returns the tensor resulted from performing the less_equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLessOrEqual_12#
- class mlprodict.npy.xop_auto_import_.OnnxLessOrEqual_12(*args, **kwargs)#
Version
name: LessOrEqual (GitHub)
domain: main
since_version: 12
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 12.
Summary
Returns the tensor resulted from performing the less_equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLessOrEqual_16#
- class mlprodict.npy.xop_auto_import_.OnnxLessOrEqual_16(*args, **kwargs)#
Version
name: LessOrEqual (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 16.
Summary
Returns the tensor resulted from performing the less_equal logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLess_1#
- class mlprodict.npy.xop_auto_import_.OnnxLess_1(*args, **kwargs)#
Version
name: Less (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the tensor resulted from performing the less logical operation elementwise on the input tensors A and B.
If broadcasting is enabled, the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. See the doc of Add for a detailed description of the broadcasting rules.
Attributes
axis: If set, defines the broadcast dimensions.
broadcast: Enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: Left input tensor for the logical operator.
B (heterogeneous) - T: Right input tensor for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input to float tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLess_13#
- class mlprodict.npy.xop_auto_import_.OnnxLess_13(*args, **kwargs)#
Version
name: Less (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the tensor resulted from performing the less logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLess_7#
- class mlprodict.npy.xop_auto_import_.OnnxLess_7(*args, **kwargs)#
Version
name: Less (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the less logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input to float tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLess_9#
- class mlprodict.npy.xop_auto_import_.OnnxLess_9(*args, **kwargs)#
Version
name: Less (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Returns the tensor resulted from performing the less logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all numeric tensors.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxLog#
- class mlprodict.npy.xop_auto_import_.OnnxLog(*args, **kwargs)#
Version
name: Log (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculates the natural log of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The natural log of the input tensor computed element-wise
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLogSoftmax#
- class mlprodict.npy.xop_auto_import_.OnnxLogSoftmax(*args, **kwargs)#
Version
name: LogSoftmax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator computes the log of softmax values for the given input:
LogSoftmax(input, axis) = Log(Softmax(input, axis=axis))
The “axis” attribute indicates the dimension along which LogSoftmax will be performed. The output tensor has the same shape and contains the LogSoftmax values of the corresponding input.
Attributes
- axis:
Describes the dimension LogSoftmax will be performed on. Negative
value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.
Inputs
input (heterogeneous) - T: The input tensor of rank >= axis.
Outputs
output (heterogeneous) - T: The output values with the same shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLogSoftmax_1#
- class mlprodict.npy.xop_auto_import_.OnnxLogSoftmax_1(*args, **kwargs)#
Version
name: LogSoftmax (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
- The operator computes the logsoftmax (log of softmax) values for each layer in the batch
of the given input. The input is a 2-D tensor (Tensor<float>) of size
(batch_size x input_feature_dimensions). The output tensor has the same shape and contains the logsoftmax values of the corresponding input.
Input does not need to explicitly be a 2D vector; rather, it will be coerced into one. For an arbitrary n-dimensional tensor input in [a_0, a_1, …, a_{k-1}, a_k, …, a_{n-1}] and k is the axis provided, then input will be coerced into a 2-dimensional tensor with dimensions [a_0 * … * a_{k-1}, a_k * … * a_{n-1}]. For the default case where axis=1, this means the input tensor will be coerced into a 2D tensor of dimensions [a_0, a_1 * … * a_{n-1}], where a_0 is often the batch size. In this situation, we must have a_0 = N and a_1 * … * a_{n-1} = D. Each of these dimensions must be matched correctly, or else the operator will throw errors.
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
1.
Inputs
input (heterogeneous) - T: The input tensor that’s coerced into a 2D matrix of size (NxD) as described above.
Outputs
output (heterogeneous) - T: The output values with the same shape as input tensor (the original size without coercion).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLogSoftmax_11#
- class mlprodict.npy.xop_auto_import_.OnnxLogSoftmax_11(*args, **kwargs)#
Version
name: LogSoftmax (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- The operator computes the logsoftmax (log of softmax) values for each layer in the batch
of the given input.
The input does not need to explicitly be a 2D vector; rather, it will be coerced into one. For an arbitrary n-dimensional tensor input in [a_0, a_1, …, a_{k-1}, a_k, …, a_{n-1}] and k is the axis provided, then input will be coerced into a 2-dimensional tensor with dimensions [a_0 * … * a_{k-1}, a_k * … * a_{n-1}]. For the default case where axis=1, this means the input tensor will be coerced into a 2D tensor of dimensions [a_0, a_1 * … * a_{n-1}], where a_0 is often the batch size. In this situation, we must have a_0 = N and a_1 * … * a_{n-1} = D. Each of these dimensions must be matched correctly, or else the operator will throw errors. The output tensor has the same shape and contains the logsoftmax values of the corresponding input.
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
input (heterogeneous) - T: The input tensor that’s coerced into a 2D matrix of size (NxD) as described above.
Outputs
output (heterogeneous) - T: The output values with the same shape as input tensor (the original size without coercion).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLogSoftmax_13#
- class mlprodict.npy.xop_auto_import_.OnnxLogSoftmax_13(*args, **kwargs)#
Version
name: LogSoftmax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator computes the log of softmax values for the given input:
LogSoftmax(input, axis) = Log(Softmax(input, axis=axis))
The “axis” attribute indicates the dimension along which LogSoftmax will be performed. The output tensor has the same shape and contains the LogSoftmax values of the corresponding input.
Attributes
- axis:
Describes the dimension LogSoftmax will be performed on. Negative
value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.
Inputs
input (heterogeneous) - T: The input tensor of rank >= axis.
Outputs
output (heterogeneous) - T: The output values with the same shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLog_1#
- class mlprodict.npy.xop_auto_import_.OnnxLog_1(*args, **kwargs)#
Version
name: Log (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Calculates the natural log of the given input tensor, element-wise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The natural log of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLog_13#
- class mlprodict.npy.xop_auto_import_.OnnxLog_13(*args, **kwargs)#
Version
name: Log (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculates the natural log of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The natural log of the input tensor computed element-wise
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLog_6#
- class mlprodict.npy.xop_auto_import_.OnnxLog_6(*args, **kwargs)#
Version
name: Log (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Calculates the natural log of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The natural log of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLoop#
- class mlprodict.npy.xop_auto_import_.OnnxLoop(*args, **kwargs)#
Version
name: Loop (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Generic Looping construct. This loop has multiple termination conditions:
Trip count. Iteration count specified at runtime. Set by specifying the input M. Optional. Set to empty string to omit. Note that a static trip count (specified at graph construction time) can be specified by passing in a constant node for input M.
Loop termination condition. This is an input to the op that determines whether to run the first iteration and also a loop-carried dependency for the body graph. The body graph must yield a value for the condition variable, whether this input is provided or not.
This table summarizes the operating modes of this operator with equivalent C-style code:
Operator inputs defined as (max_trip_count, condition_var).
- input (“”, “”):
- for (int i=0; ; ++i) {
cond = … // Note this value is ignored, but is required in the body
}
- input (“”, cond) // Note this is analogous to a while loop
bool cond = …; for (int i=0; cond; ++i) {
cond = …;
}
- input (“”, 1) // Note this is analogous to a do-while loop
bool cond = true for (int i=0; cond; ++i) {
cond = …;
}
- input (trip_count, “”) // Note this is analogous to a for loop
int trip_count = … for (int i=0; i < trip_count; ++i) {
cond = …; // ignored
}
- input (trip_count, cond)
int trip_count = …; bool cond = …; for (int i=0; i < trip_count && cond; ++i) {
cond = …;
}
Sample usage - cond as well as trip count
- graph predict-net {
%a = Constant[value = <Scalar Tensor [3]>]() %b = Constant[value = <Scalar Tensor [6]>]() %keepgoing = Constant[value = <Scalar Tensor [1]>]() %max_trip_count = Constant[value = <Scalar Tensor [10]>]() %keepgoing_out, %b_out, %user_defined_vals = Loop[body = <graph body-net>](%max_trip_count, %keepgoing, %b) return
}
- graph body-net (
%i[INT32, scalar] // iteration number %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used %b_in[INT32, scalar] // incoming value of loop-carried-dependency b
- ) {
%my_local = Add(%a, %b_in) %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated return %keepgoing_out, %b_out, %user_defined_val
}
Sample equivalent C code
- {
/* User-defined code (enclosing scope) / int a = 3, b = 6; bool keepgoing = true; // Analogous to input cond / End user-defined code */
/* Implicitly-defined code / const int max_trip_count = 10; // Analogous to input M int user_defined_vals[]; // Imagine this is resizable / End implicitly-defined code / / initialize loop-carried variables and scan-output variables */ bool keepgoing_out = keepgoing int b_out = b
- for (int i=0; i < max_trip_count && keepgoing_out; ++i) {
- /* Implicitly-defined code: bind actual parameter values
to formal parameter variables of loop-body */
bool keepgoing_in = keepgoing_out; bool b_in = b_out;
/* User-defined code (loop body) / int my_local = a + b_in; // Reading value “a” from the enclosing scope is fine b_out = a - b_in; keepgoing_out = my_local > b_out; user_defined_val = b_in + b_in; // b_in and b_out are different variables / End user-defined code */
/* Implicitly defined-code */ user_defined_vals[i] = user_defined_val // accumulate scan-output values
} // int t = my_local; // Can’t do this. my_local is not accessible here.
// The values below are bound to the output variables of the loop and therefore accessible // b_out; user_defined_vals; keepgoing_out;
}
There are several things of note in this code snippet:
Values from the enclosing scope (i.e. variable “a” here) are in scope and can be referenced in the inputs of the loop.
Any values computed in the loop body that needs to be used in a subsequent iteration or after the loop are modelled using a pair of variables in the loop-body, consisting of an input variable (eg., b_in) and an output variable (eg., b_out). These are referred to as loop-carried dependences. The loop operation node supplies the input value of the input variable for the first iteration, and returns the output value of the output variable produced by the final iteration.
Scan_output variables are used to implicitly concatenate values computed across all the iterations. In the above example, the value of user_defined_val computed over all iterations are concatenated and returned as the value of user_defined_vals after the loop.
Values created in the body cannot be accessed in the enclosing scope, except using the mechanism described above.
Note that the semantics of this op support “diagonal” or “wavefront” execution. (See Step 3 here for an example: https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). Frontends should emit multi-layer RNNs as a series of While operators (with time being the inner looping dimension), with each successive layer consuming the scan_outputs from the previous layer, possibly going through several point-wise operators (e.g. dropout, residual connections, linear layer).
The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order.
Attributes
body (required): The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies…). It has 1+N+K outputs: (condition, loop carried dependencies…, scan_outputs…). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
Inputs
Between 2 and 2147483647 inputs.
M (optional, heterogeneous) - I: A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
cond (optional, heterogeneous) - B: A boolean termination condition. Optional. Pass empty string to skip.
v_initial (variadic) - V: The initial values of any loop-carried dependencies (values that change across loop iterations)
Outputs
Between 1 and 2147483647 outputs.
v_final_and_scan_outputs (variadic) - V: Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
Type Constraints
V in ( optional(seq(tensor(bfloat16))), optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bfloat16)), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bfloat16)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types
I in ( tensor(int64) ): tensor of int64, which should be a scalar.
B in ( tensor(bool) ): tensor of bool, which should be a scalar.
OnnxLoop_1#
- class mlprodict.npy.xop_auto_import_.OnnxLoop_1(*args, **kwargs)#
Version
name: Loop (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generic Looping construct. This loop has multiple termination conditions:
Trip count. Iteration count specified at runtime. Set by specifying the input M. Optional. Set to empty string to omit. Note that a static trip count (specified at graph construction time) can be specified by passing in a constant node for input M.
Loop termination condition. This is an input to the op that determines whether to run the first iteration and also a loop-carried dependency for the body graph. The body graph must yield a value for the condition variable, whether this input is provided or not.
This table summarizes the operating modes of this operator with equivalent C-style code:
Operator inputs defined as (max_trip_count, condition_var).
- input (“”, “”):
- for (int i=0; ; ++i) {
cond = … // Note this value is ignored, but is required in the body
}
- input (“”, cond) // Note this is analogous to a while loop
bool cond = …; for (int i=0; cond; ++i) {
cond = …;
}
- input (“”, 1) // Note this is analogous to a do-while loop
bool cond = true for (int i=0; cond; ++i) {
cond = …;
}
- input (trip_count, “”) // Note this is analogous to a for loop
int trip_count = … for (int i=0; i < trip_count; ++i) {
cond = …; // ignored
}
- input (trip_count, cond)
int trip_count = …; bool cond = …; for (int i=0; i < trip_count && cond; ++i) {
cond = …;
}
Sample usage - cond as well as trip count
- graph predict-net {
%a = Constant[value = <Scalar Tensor [3]>]() %b = Constant[value = <Scalar Tensor [6]>]() %keepgoing = Constant[value = <Scalar Tensor [1]>]() %max_trip_count = Constant[value = <Scalar Tensor [10]>]() %keepgoing_out, %b_out, %user_defined_vals = Loop[body = <graph body-net>](%max_trip_count, %keepgoing, %b) return
}
- graph body-net (
%i[INT32, scalar] %keepgoing[BOOL, scalar] %b[INT32, scalar]
- ) {
%my_local = Add(%a, %b) %b_out = Sub(%a, %b) %keepgoing_out = Greater(%my_local, %b_out) %user_defined_vals = Add(%b, %b) return %keepgoing_out, %b_out, %user_defined_vals
}
Sample equivalent C code
- {
/* User-defined code (enclosing scope) / int a = 3, b = 6; bool keepgoing = true; // Analogous to input cond / End user-defined code */
/* Implicitly-defined code / const int max_trip_count = 10; // Analogous to input M int user_defined_vals[]; // Imagine this is resizable / End implicitly-defined code */ for (int i=0; i < max_trip_count && keepgoing; ++i) {
/* User-defined code (loop body) / int my_local = a + b; // Reading values in the enclosing scope is fine b = a - b; // writes fine if we specify b as a loop-carried dependency keepgoing = my_local > b; // keepgoing is a loop-carried dependency user_defined_vals[i] = b + b; / End user-defined code */
} // my_local = 123; // Can’t do this. my_local was defined in the body
// These below values are live-out from the loop and therefore accessible b_out; user_defined_vals; keepgoing_out;
}
There are several things of note in this code snippet:
Values from the enclosing scope (i.e. variable a here) are in scope and can be referenced in the inputs of the loop.
Any variables which you wish to make available in the enclosing scope (i.e. the variables b and keepgoing) must be declared as either loop-carried dependencies (both at the op inputs and output and at the body net input and output) or scan_outputs.
Values created in the body cannot be accessed in the enclosing scope.
Note that the semantics of this op support “diagonal” or “wavefront” execution. (See Step 3 here for an example: https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). Frontends should emit multi-layer RNNs as a series of While operators (with time being the inner looping dimension), with each successive layer consuming the scan_outputs from the previous layer, possibly going through several point-wise operators (e.g. dropout, residual connections, linear layer).
Attributes
body (required): The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies…). It has 1+N+K outputs: (condition, loop carried dependencies…, scan_outputs…). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
Inputs
Between 3 and 2147483647 inputs.
M (optional, heterogeneous) - I: A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
cond (optional, heterogeneous) - B: A boolean termination condition. Optional. Pass empty string to skip.
v_initial (variadic) - V: The initial values of any loop-carried dependencies (values that change across loop iterations)
Outputs
Between 1 and 2147483647 outputs.
v_final_and_scan_outputs (variadic) - V: Final N loop carried dependency values then K scan_outputs
Type Constraints
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
I in ( tensor(int64) ): tensor of int64, which should be a scalar.
B in ( tensor(bool) ): tensor of bool, which should be a scalar.
OnnxLoop_11#
- class mlprodict.npy.xop_auto_import_.OnnxLoop_11(*args, **kwargs)#
Version
name: Loop (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Generic Looping construct. This loop has multiple termination conditions:
Trip count. Iteration count specified at runtime. Set by specifying the input M. Optional. Set to empty string to omit. Note that a static trip count (specified at graph construction time) can be specified by passing in a constant node for input M.
Loop termination condition. This is an input to the op that determines whether to run the first iteration and also a loop-carried dependency for the body graph. The body graph must yield a value for the condition variable, whether this input is provided or not.
This table summarizes the operating modes of this operator with equivalent C-style code:
Operator inputs defined as (max_trip_count, condition_var).
- input (“”, “”):
- for (int i=0; ; ++i) {
cond = … // Note this value is ignored, but is required in the body
}
- input (“”, cond) // Note this is analogous to a while loop
bool cond = …; for (int i=0; cond; ++i) {
cond = …;
}
- input (“”, 1) // Note this is analogous to a do-while loop
bool cond = true for (int i=0; cond; ++i) {
cond = …;
}
- input (trip_count, “”) // Note this is analogous to a for loop
int trip_count = … for (int i=0; i < trip_count; ++i) {
cond = …; // ignored
}
- input (trip_count, cond)
int trip_count = …; bool cond = …; for (int i=0; i < trip_count && cond; ++i) {
cond = …;
}
Sample usage - cond as well as trip count
- graph predict-net {
%a = Constant[value = <Scalar Tensor [3]>]() %b = Constant[value = <Scalar Tensor [6]>]() %keepgoing = Constant[value = <Scalar Tensor [1]>]() %max_trip_count = Constant[value = <Scalar Tensor [10]>]() %keepgoing_out, %b_out, %user_defined_vals = Loop[body = <graph body-net>](%max_trip_count, %keepgoing, %b) return
}
- graph body-net (
%i[INT32, scalar] // iteration number %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used %b_in[INT32, scalar] // incoming value of loop-carried-dependency b
- ) {
%my_local = Add(%a, %b_in) %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated return %keepgoing_out, %b_out, %user_defined_val
}
Sample equivalent C code
- {
/* User-defined code (enclosing scope) / int a = 3, b = 6; bool keepgoing = true; // Analogous to input cond / End user-defined code */
/* Implicitly-defined code / const int max_trip_count = 10; // Analogous to input M int user_defined_vals[]; // Imagine this is resizable / End implicitly-defined code / / initialize loop-carried variables and scan-output variables */ bool keepgoing_out = keepgoing int b_out = b
- for (int i=0; i < max_trip_count && keepgoing_out; ++i) {
- /* Implicitly-defined code: bind actual parameter values
to formal parameter variables of loop-body */
bool keepgoing_in = keepgoing_out; bool b_in = b_out;
/* User-defined code (loop body) / int my_local = a + b_in; // Reading value “a” from the enclosing scope is fine b_out = a - b_in; keepgoing_out = my_local > b_out; user_defined_val = b_in + b_in; // b_in and b_out are different variables / End user-defined code */
/* Implicitly defined-code */ user_defined_vals[i] = user_defined_val // accumulate scan-output values
} // int t = my_local; // Can’t do this. my_local is not accessible here.
// The values below are bound to the output variables of the loop and therefore accessible // b_out; user_defined_vals; keepgoing_out;
}
There are several things of note in this code snippet:
Values from the enclosing scope (i.e. variable “a” here) are in scope and can be referenced in the inputs of the loop.
Any values computed in the loop body that needs to be used in a subsequent iteration or after the loop are modelled using a pair of variables in the loop-body, consisting of an input variable (eg., b_in) and an output variable (eg., b_out). These are referred to as loop-carried dependences. The loop operation node supplies the input value of the input variable for the first iteration, and returns the output value of the output variable produced by the final iteration.
Scan_output variables are used to implicitly concatenate values computed across all the iterations. In the above example, the value of user_defined_val computed over all iterations are concatenated and returned as the value of user_defined_vals after the loop.
Values created in the body cannot be accessed in the enclosing scope, except using the mechanism described above.
Note that the semantics of this op support “diagonal” or “wavefront” execution. (See Step 3 here for an example: https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). Frontends should emit multi-layer RNNs as a series of While operators (with time being the inner looping dimension), with each successive layer consuming the scan_outputs from the previous layer, possibly going through several point-wise operators (e.g. dropout, residual connections, linear layer).
Attributes
body (required): The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies…). It has 1+N+K outputs: (condition, loop carried dependencies…, scan_outputs…). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
Inputs
Between 2 and 2147483647 inputs.
M (optional, heterogeneous) - I: A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
cond (optional, heterogeneous) - B: A boolean termination condition. Optional. Pass empty string to skip.
v_initial (variadic) - V: The initial values of any loop-carried dependencies (values that change across loop iterations)
Outputs
Between 1 and 2147483647 outputs.
v_final_and_scan_outputs (variadic) - V: Final N loop carried dependency values then K scan_outputs
Type Constraints
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
I in ( tensor(int64) ): tensor of int64, which should be a scalar.
B in ( tensor(bool) ): tensor of bool, which should be a scalar.
OnnxLoop_13#
- class mlprodict.npy.xop_auto_import_.OnnxLoop_13(*args, **kwargs)#
Version
name: Loop (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Generic Looping construct. This loop has multiple termination conditions:
Trip count. Iteration count specified at runtime. Set by specifying the input M. Optional. Set to empty string to omit. Note that a static trip count (specified at graph construction time) can be specified by passing in a constant node for input M.
Loop termination condition. This is an input to the op that determines whether to run the first iteration and also a loop-carried dependency for the body graph. The body graph must yield a value for the condition variable, whether this input is provided or not.
This table summarizes the operating modes of this operator with equivalent C-style code:
Operator inputs defined as (max_trip_count, condition_var).
- input (“”, “”):
- for (int i=0; ; ++i) {
cond = … // Note this value is ignored, but is required in the body
}
- input (“”, cond) // Note this is analogous to a while loop
bool cond = …; for (int i=0; cond; ++i) {
cond = …;
}
- input (“”, 1) // Note this is analogous to a do-while loop
bool cond = true for (int i=0; cond; ++i) {
cond = …;
}
- input (trip_count, “”) // Note this is analogous to a for loop
int trip_count = … for (int i=0; i < trip_count; ++i) {
cond = …; // ignored
}
- input (trip_count, cond)
int trip_count = …; bool cond = …; for (int i=0; i < trip_count && cond; ++i) {
cond = …;
}
Sample usage - cond as well as trip count
- graph predict-net {
%a = Constant[value = <Scalar Tensor [3]>]() %b = Constant[value = <Scalar Tensor [6]>]() %keepgoing = Constant[value = <Scalar Tensor [1]>]() %max_trip_count = Constant[value = <Scalar Tensor [10]>]() %keepgoing_out, %b_out, %user_defined_vals = Loop[body = <graph body-net>](%max_trip_count, %keepgoing, %b) return
}
- graph body-net (
%i[INT32, scalar] // iteration number %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used %b_in[INT32, scalar] // incoming value of loop-carried-dependency b
- ) {
%my_local = Add(%a, %b_in) %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated return %keepgoing_out, %b_out, %user_defined_val
}
Sample equivalent C code
- {
/* User-defined code (enclosing scope) / int a = 3, b = 6; bool keepgoing = true; // Analogous to input cond / End user-defined code */
/* Implicitly-defined code / const int max_trip_count = 10; // Analogous to input M int user_defined_vals[]; // Imagine this is resizable / End implicitly-defined code / / initialize loop-carried variables and scan-output variables */ bool keepgoing_out = keepgoing int b_out = b
- for (int i=0; i < max_trip_count && keepgoing_out; ++i) {
- /* Implicitly-defined code: bind actual parameter values
to formal parameter variables of loop-body */
bool keepgoing_in = keepgoing_out; bool b_in = b_out;
/* User-defined code (loop body) / int my_local = a + b_in; // Reading value “a” from the enclosing scope is fine b_out = a - b_in; keepgoing_out = my_local > b_out; user_defined_val = b_in + b_in; // b_in and b_out are different variables / End user-defined code */
/* Implicitly defined-code */ user_defined_vals[i] = user_defined_val // accumulate scan-output values
} // int t = my_local; // Can’t do this. my_local is not accessible here.
// The values below are bound to the output variables of the loop and therefore accessible // b_out; user_defined_vals; keepgoing_out;
}
There are several things of note in this code snippet:
Values from the enclosing scope (i.e. variable “a” here) are in scope and can be referenced in the inputs of the loop.
Any values computed in the loop body that needs to be used in a subsequent iteration or after the loop are modelled using a pair of variables in the loop-body, consisting of an input variable (eg., b_in) and an output variable (eg., b_out). These are referred to as loop-carried dependences. The loop operation node supplies the input value of the input variable for the first iteration, and returns the output value of the output variable produced by the final iteration.
Scan_output variables are used to implicitly concatenate values computed across all the iterations. In the above example, the value of user_defined_val computed over all iterations are concatenated and returned as the value of user_defined_vals after the loop.
Values created in the body cannot be accessed in the enclosing scope, except using the mechanism described above.
Note that the semantics of this op support “diagonal” or “wavefront” execution. (See Step 3 here for an example: https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). Frontends should emit multi-layer RNNs as a series of While operators (with time being the inner looping dimension), with each successive layer consuming the scan_outputs from the previous layer, possibly going through several point-wise operators (e.g. dropout, residual connections, linear layer).
The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order.
Attributes
body (required): The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies…). It has 1+N+K outputs: (condition, loop carried dependencies…, scan_outputs…). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
Inputs
Between 2 and 2147483647 inputs.
M (optional, heterogeneous) - I: A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
cond (optional, heterogeneous) - B: A boolean termination condition. Optional. Pass empty string to skip.
v_initial (variadic) - V: The initial values of any loop-carried dependencies (values that change across loop iterations)
Outputs
Between 1 and 2147483647 outputs.
v_final_and_scan_outputs (variadic) - V: Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
Type Constraints
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor and Sequence types
I in ( tensor(int64) ): tensor of int64, which should be a scalar.
B in ( tensor(bool) ): tensor of bool, which should be a scalar.
OnnxLoop_16#
- class mlprodict.npy.xop_auto_import_.OnnxLoop_16(*args, **kwargs)#
Version
name: Loop (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Generic Looping construct. This loop has multiple termination conditions:
Trip count. Iteration count specified at runtime. Set by specifying the input M. Optional. Set to empty string to omit. Note that a static trip count (specified at graph construction time) can be specified by passing in a constant node for input M.
Loop termination condition. This is an input to the op that determines whether to run the first iteration and also a loop-carried dependency for the body graph. The body graph must yield a value for the condition variable, whether this input is provided or not.
This table summarizes the operating modes of this operator with equivalent C-style code:
Operator inputs defined as (max_trip_count, condition_var).
- input (“”, “”):
- for (int i=0; ; ++i) {
cond = … // Note this value is ignored, but is required in the body
}
- input (“”, cond) // Note this is analogous to a while loop
bool cond = …; for (int i=0; cond; ++i) {
cond = …;
}
- input (“”, 1) // Note this is analogous to a do-while loop
bool cond = true for (int i=0; cond; ++i) {
cond = …;
}
- input (trip_count, “”) // Note this is analogous to a for loop
int trip_count = … for (int i=0; i < trip_count; ++i) {
cond = …; // ignored
}
- input (trip_count, cond)
int trip_count = …; bool cond = …; for (int i=0; i < trip_count && cond; ++i) {
cond = …;
}
Sample usage - cond as well as trip count
- graph predict-net {
%a = Constant[value = <Scalar Tensor [3]>]() %b = Constant[value = <Scalar Tensor [6]>]() %keepgoing = Constant[value = <Scalar Tensor [1]>]() %max_trip_count = Constant[value = <Scalar Tensor [10]>]() %keepgoing_out, %b_out, %user_defined_vals = Loop[body = <graph body-net>](%max_trip_count, %keepgoing, %b) return
}
- graph body-net (
%i[INT32, scalar] // iteration number %keepgoing_in[BOOL, scalar] // incoming loop-termination-condition; not used %b_in[INT32, scalar] // incoming value of loop-carried-dependency b
- ) {
%my_local = Add(%a, %b_in) %b_out = Sub(%a, %b_in) // outgoing value of loop-carried-dependency b %keepgoing_out = Greater(%my_local, %b_out) // outgoing loop-termination-condition %user_defined_val = Add(%b_in, %b_in) // scan-output value to be accumulated return %keepgoing_out, %b_out, %user_defined_val
}
Sample equivalent C code
- {
/* User-defined code (enclosing scope) / int a = 3, b = 6; bool keepgoing = true; // Analogous to input cond / End user-defined code */
/* Implicitly-defined code / const int max_trip_count = 10; // Analogous to input M int user_defined_vals[]; // Imagine this is resizable / End implicitly-defined code / / initialize loop-carried variables and scan-output variables */ bool keepgoing_out = keepgoing int b_out = b
- for (int i=0; i < max_trip_count && keepgoing_out; ++i) {
- /* Implicitly-defined code: bind actual parameter values
to formal parameter variables of loop-body */
bool keepgoing_in = keepgoing_out; bool b_in = b_out;
/* User-defined code (loop body) / int my_local = a + b_in; // Reading value “a” from the enclosing scope is fine b_out = a - b_in; keepgoing_out = my_local > b_out; user_defined_val = b_in + b_in; // b_in and b_out are different variables / End user-defined code */
/* Implicitly defined-code */ user_defined_vals[i] = user_defined_val // accumulate scan-output values
} // int t = my_local; // Can’t do this. my_local is not accessible here.
// The values below are bound to the output variables of the loop and therefore accessible // b_out; user_defined_vals; keepgoing_out;
}
There are several things of note in this code snippet:
Values from the enclosing scope (i.e. variable “a” here) are in scope and can be referenced in the inputs of the loop.
Any values computed in the loop body that needs to be used in a subsequent iteration or after the loop are modelled using a pair of variables in the loop-body, consisting of an input variable (eg., b_in) and an output variable (eg., b_out). These are referred to as loop-carried dependences. The loop operation node supplies the input value of the input variable for the first iteration, and returns the output value of the output variable produced by the final iteration.
Scan_output variables are used to implicitly concatenate values computed across all the iterations. In the above example, the value of user_defined_val computed over all iterations are concatenated and returned as the value of user_defined_vals after the loop.
Values created in the body cannot be accessed in the enclosing scope, except using the mechanism described above.
Note that the semantics of this op support “diagonal” or “wavefront” execution. (See Step 3 here for an example: https://devblogs.nvidia.com/optimizing-recurrent-neural-networks-cudnn-5/). Frontends should emit multi-layer RNNs as a series of While operators (with time being the inner looping dimension), with each successive layer consuming the scan_outputs from the previous layer, possibly going through several point-wise operators (e.g. dropout, residual connections, linear layer).
The input/output of subgraph (produced by loop node) matching is based on order instead of name. The implementation will figure out the names based on this order.
Attributes
body (required): The graph run each iteration. It has 2+N inputs: (iteration_num, condition, loop carried dependencies…). It has 1+N+K outputs: (condition, loop carried dependencies…, scan_outputs…). Each scan_output is created by concatenating the value of the specified output value at the end of each iteration of the loop. It is an error if the dimensions or data type of these scan_outputs change across loop iterations.
Inputs
Between 2 and 2147483647 inputs.
M (optional, heterogeneous) - I: A maximum trip-count for the loop specified at runtime. Optional. Pass empty string to skip.
cond (optional, heterogeneous) - B: A boolean termination condition. Optional. Pass empty string to skip.
v_initial (variadic) - V: The initial values of any loop-carried dependencies (values that change across loop iterations)
Outputs
Between 1 and 2147483647 outputs.
v_final_and_scan_outputs (variadic) - V: Final N loop carried dependency values then K scan_outputs. Scan outputs must be Tensors.
Type Constraints
V in ( optional(seq(tensor(bfloat16))), optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bfloat16)), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bfloat16)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor, Sequence(Tensor), Optional(Tensor), and Optional(Sequence(Tensor)) types
I in ( tensor(int64) ): tensor of int64, which should be a scalar.
B in ( tensor(bool) ): tensor of bool, which should be a scalar.
OnnxLpNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxLpNormalization(*args, **kwargs)#
Version
name: LpNormalization (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Given a matrix, apply Lp-normalization along the provided axis.
Attributes
axis: The axis on which to apply normalization, -1 mean last axis. Default value is
-1.p: The order of the normalization, only 1 or 2 are supported. Default value is
2.
Inputs
input (heterogeneous) - T: Input matrix
Outputs
output (heterogeneous) - T: Matrix after normalization
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLpNormalization_1#
- class mlprodict.npy.xop_auto_import_.OnnxLpNormalization_1(*args, **kwargs)#
Version
name: LpNormalization (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Given a matrix, apply Lp-normalization along the provided axis.
Attributes
axis: The axis on which to apply normalization, -1 mean last axis. Default value is
-1.p: The order of the normalization, only 1 or 2 are supported. Default value is
2.
Inputs
input (heterogeneous) - T: Input matrix
Outputs
output (heterogeneous) - T: Matrix after normalization
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLpPool#
- class mlprodict.npy.xop_auto_import_.OnnxLpPool(*args, **kwargs)#
Version
name: LpPool (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
LpPool consumes an input tensor X and applies Lp pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. Lp pooling consisting of computing the Lp norm on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - {kernelSpatialShape} + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + {kernelSpatialShape} - input_spatial_shape[i]
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
kernel_shape (required): The size of the kernel along each axis.
p: p value of the Lp norm used to pool over the input data. Default value is
2.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLpPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxLpPool_1(*args, **kwargs)#
Version
name: LpPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
LpPool consumes an input tensor X and applies Lp pooling across the the tensor according to kernel sizes, stride sizes, and pad lengths. Lp pooling consisting of computing the Lp norm on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. DEPRECATION NOTE: auto_pad is only intended to support legacy uses, and for framework authors, one is explicitly encouraged to use explicit padding specified in the pads attribute. Default value is
'NOTSET'.kernel_shape: The size of the kernel along each axis.
p: p value of the Lp norm used to pool over the input data, default is 2.0. Default value is
2.0.pads: Padding for the beginning and ending along each axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute.
strides: Stride along each axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimension are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLpPool_11#
- class mlprodict.npy.xop_auto_import_.OnnxLpPool_11(*args, **kwargs)#
Version
name: LpPool (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
LpPool consumes an input tensor X and applies Lp pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. Lp pooling consisting of computing the Lp norm on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.kernel_shape (required): The size of the kernel along each axis.
p: p value of the Lp norm used to pool over the input data. Default value is
2.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLpPool_18#
- class mlprodict.npy.xop_auto_import_.OnnxLpPool_18(*args, **kwargs)#
Version
name: LpPool (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
LpPool consumes an input tensor X and applies Lp pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. Lp pooling consisting of computing the Lp norm on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - {kernelSpatialShape}) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - {kernelSpatialShape} + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + {kernelSpatialShape} - input_spatial_shape[i]
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults is 1 along each spatial axis.
kernel_shape (required): The size of the kernel along each axis.
p: p value of the Lp norm used to pool over the input data. Default value is
2.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxLpPool_2#
- class mlprodict.npy.xop_auto_import_.OnnxLpPool_2(*args, **kwargs)#
Version
name: LpPool (GitHub)
domain: main
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2.
Summary
LpPool consumes an input tensor X and applies Lp pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. Lp pooling consisting of computing the Lp norm on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.kernel_shape (required): The size of the kernel along each axis.
p: p value of the Lp norm used to pool over the input data. Default value is
2.pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size.
Outputs
Y (heterogeneous) - T: Output data tensor from Lp pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMatMul#
- class mlprodict.npy.xop_auto_import_.OnnxMatMul(*args, **kwargs)#
Version
name: MatMul (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results from A * B
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxMatMulInteger#
- class mlprodict.npy.xop_auto_import_.OnnxMatMulInteger(*args, **kwargs)#
Version
name: MatMulInteger (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html. The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
Inputs
Between 2 and 4 inputs.
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
a_zero_point (optional, heterogeneous) - T1: Zero point tensor for input ‘A’. It’s optional and default value is 0. It could be a scalar or N-D tensor. Scalar refers to per tensor quantization whereas N-D refers to per row quantization. If the input is 2D of shape [M, K] then zero point tensor may be an M element vector [zp_1, zp_2, …, zp_M]. If the input is N-D tensor with shape [D1, D2, M, K] then zero point tensor may have shape [D1, D2, M, 1].
b_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a N-D tensor, Scalar refers to per tensor quantization whereas N-D refers to per col quantization. If the input is 2D of shape [K, N] then zero point tensor may be an N element vector [zp_1, zp_2, …, zp_N]. If the input is N-D tensor with shape [D1, D2, K, N] then zero point tensor may have shape [D1, D2, 1, N].
Outputs
Y (heterogeneous) - T3: Matrix multiply results from A * B
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input A data type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain input B data type to 8-bit integer tensor.
T3 in ( tensor(int32) ): Constrain output Y data type as 32-bit integer tensor.
OnnxMatMulInteger_10#
- class mlprodict.npy.xop_auto_import_.OnnxMatMulInteger_10(*args, **kwargs)#
Version
name: MatMulInteger (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html. The production MUST never overflow. The accumulation may overflow if and only if in 32 bits.
Inputs
Between 2 and 4 inputs.
A (heterogeneous) - T1: N-dimensional matrix A
B (heterogeneous) - T2: N-dimensional matrix B
a_zero_point (optional, heterogeneous) - T1: Zero point tensor for input ‘A’. It’s optional and default value is 0. It could be a scalar or N-D tensor. Scalar refers to per tensor quantization whereas N-D refers to per row quantization. If the input is 2D of shape [M, K] then zero point tensor may be an M element vector [zp_1, zp_2, …, zp_M]. If the input is N-D tensor with shape [D1, D2, M, K] then zero point tensor may have shape [D1, D2, M, 1].
b_zero_point (optional, heterogeneous) - T2: Zero point tensor for input ‘B’. It’s optional and default value is 0. It could be a scalar or a N-D tensor, Scalar refers to per tensor quantization whereas N-D refers to per col quantization. If the input is 2D of shape [K, N] then zero point tensor may be an N element vector [zp_1, zp_2, …, zp_N]. If the input is N-D tensor with shape [D1, D2, K, N] then zero point tensor may have shape [D1, D2, 1, N].
Outputs
Y (heterogeneous) - T3: Matrix multiply results from A * B
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input A data type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain input B data type to 8-bit integer tensor.
T3 in ( tensor(int32) ): Constrain output Y data type as 32-bit integer tensor.
OnnxMatMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxMatMul_1(*args, **kwargs)#
Version
name: MatMul (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results from A * B
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMatMul_13#
- class mlprodict.npy.xop_auto_import_.OnnxMatMul_13(*args, **kwargs)#
Version
name: MatMul (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results from A * B
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxMatMul_9#
- class mlprodict.npy.xop_auto_import_.OnnxMatMul_9(*args, **kwargs)#
Version
name: MatMul (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html
Inputs
A (heterogeneous) - T: N-dimensional matrix A
B (heterogeneous) - T: N-dimensional matrix B
Outputs
Y (heterogeneous) - T: Matrix multiply results from A * B
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxMax#
- class mlprodict.npy.xop_auto_import_.OnnxMax(*args, **kwargs)#
Version
name: Max (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise max of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for max.
Outputs
max (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
OnnxMaxPool#
- class mlprodict.npy.xop_auto_import_.OnnxMaxPool(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.dilations: Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. This attribute is used only to convert an n-tuple index value into a single integer value for producing the second output. Default value is
0.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int8), tensor(uint8) ): Constrain input and output types to float and 8 bit tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxMaxPool_1(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) * pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMaxPool_10#
- class mlprodict.npy.xop_auto_import_.OnnxMaxPool_10(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.dilations: Dilation value along each spatial axis of filter.
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. Default value is
0.strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxPool_11#
- class mlprodict.npy.xop_auto_import_.OnnxMaxPool_11(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.dilations: Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. Default value is
0.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxPool_12#
- class mlprodict.npy.xop_auto_import_.OnnxMaxPool_12(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
or#
output_spatial_shape[i] = ceil((input_spatial_shape[i] + pad_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1)) / strides_spatial_shape[i] + 1)
if ceil_mode is enabled
* pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + ((kernel_spatial_shape[i] - 1) * dilations[i] + 1) - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.ceil_mode: Whether to use ceil or floor (default) to compute the output shape. Default value is
0.dilations: Dilation value along each spatial axis of filter. If not present, the dilation defaults to 1 along each spatial axis.
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. This attribute is used only to convert an n-tuple index value into a single integer value for producing the second output. Default value is
0.strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int8), tensor(uint8) ): Constrain input and output types to float and 8 bit tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxPool_8#
- class mlprodict.npy.xop_auto_import_.OnnxMaxPool_8(*args, **kwargs)#
Version
name: MaxPool (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
MaxPool consumes an input tensor X and applies max pooling across the tensor according to kernel sizes, stride sizes, and pad lengths. max pooling consisting of computing the max on all values of a subset of the input tensor according to the kernel size and downsampling the data into the output tensor Y for further processing. The output spatial shape will be following:
output_spatial_shape[i] = floor((input_spatial_shape[i] + pad_shape[i] - kernel_spatial_shape[i]) / strides_spatial_shape[i] + 1) * pad_shape[i] is sum of pads along axis i
auto_pad is a DEPRECATED attribute. If you are using them currently, the output spatial shape will be following:
VALID: output_spatial_shape[i] = ceil((input_spatial_shape[i] - kernel_spatial_shape[i] + 1) / strides_spatial_shape[i]) SAME_UPPER or SAME_LOWER: output_spatial_shape[i] = ceil(input_spatial_shape[i] / strides_spatial_shape[i])
And pad shape will be following if SAME_UPPER or SAME_LOWER:
pad_shape[i] = (output_spatial_shape[i] - 1) * strides_spatial_shape[i] + kernel_spatial_shape[i] - input_spatial_shape[i]
The output of each pooling window is maximum number of elements exclude pad.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that the output spatial size match the input.In case of odd number add the extra padding at the end for SAME_UPPER and at the beginning for SAME_LOWER. VALID mean no padding. Default value is
'NOTSET'.kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
storage_order: The storage order of the tensor. 0 is row major, and 1 is column major. Default value is
0.strides: Stride along each spatial axis.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
Outputs
Between 1 and 2 outputs.
Y (heterogeneous) - T: Output data tensor from average or max pooling across the input tensor. Dimensions will vary based on various kernel, stride, and pad sizes. Floor value of the dimension is used
Indices (optional, heterogeneous) - I: Indices tensor from max pooling across the input tensor. The dimensions of indices are the same as output tensor. The values in indices of are the indices of the selected values during pooling. The indices are computed as flatten 1-D tensor, and the indices do not consider padding. So the values in indices are in [0, N x C x D1 x … x Dn).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxRoiPool#
- class mlprodict.npy.xop_auto_import_.OnnxMaxRoiPool(*args, **kwargs)#
Version
name: MaxRoiPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
ROI max pool consumes an input tensor X and region of interests (RoIs) to apply max pooling across each RoI, to produce output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
Attributes
pooled_shape (required): ROI pool output shape (height, width).
spatial_scale: Multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling. Default value is
1.0.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T: RoIs (Regions of Interest) to pool over. Should be a 2-D tensor of shape (num_rois, 5) given as [[batch_id, x1, y1, x2, y2], …].
Outputs
Y (heterogeneous) - T: RoI pooled output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMaxRoiPool_1#
- class mlprodict.npy.xop_auto_import_.OnnxMaxRoiPool_1(*args, **kwargs)#
Version
name: MaxRoiPool (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
ROI max pool consumes an input tensor X and region of interests (RoIs) to apply max pooling across each RoI, to produce output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
Attributes
pooled_shape (required): ROI pool output shape (height, width).
spatial_scale: Multiplicative spatial scale factor to translate ROI coordinates from their input scale to the scale used when pooling. Default value is
1.0.
Inputs
X (heterogeneous) - T: Input data tensor from the previous operator; dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T: RoIs (Regions of Interest) to pool over. Should be a 2-D tensor of shape (num_rois, 5) given as [[batch_id, x1, y1, x2, y2], …].
Outputs
Y (heterogeneous) - T: RoI pooled output 4-D tensor of shape (num_rois, channels, pooled_shape[0], pooled_shape[1]).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMaxUnpool#
- class mlprodict.npy.xop_auto_import_.OnnxMaxUnpool(*args, **kwargs)#
Version
name: MaxUnpool (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- MaxUnpool essentially computes the partial inverse of the MaxPool op.
The input information to this op is typically the output information from a MaxPool op. The first input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corrsponding to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. The third (optional) input is a tensor that specifies the output size of the unpooling operation.
- MaxUnpool is intended to do ‘partial’ inverse of the MaxPool op. ‘Partial’ because all the non-maximal
values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling the result of an unpooling operation should give back the original input to the unpooling op.
- MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous.
The third input argument, output_size, is meant to disambiguate the op and produce output tensor of known/predictable size.
- In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads,
which define the exact unpooling op. The attributes typically have the same values as the corrsponding pooling op that the unpooling op is trying to invert.
Attributes
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T1: Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
I (heterogeneous) - T2: Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x … x Dn).
output_shape (optional, heterogeneous) - T2: The shape of the output can be explicitly set which will cause pads values to be auto generated. If ‘output_shape’ is specified, ‘pads’ values are ignored.
Outputs
output (heterogeneous) - T1: Output data tensor that contains the result of the unpooling.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T2 in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxUnpool_11#
- class mlprodict.npy.xop_auto_import_.OnnxMaxUnpool_11(*args, **kwargs)#
Version
name: MaxUnpool (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- MaxUnpool essentially computes the partial inverse of the MaxPool op.
The input information to this op is typically the output information from a MaxPool op. The first input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corrsponding to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. The third (optional) input is a tensor that specifies the output size of the unpooling operation.
- MaxUnpool is intended to do ‘partial’ inverse of the MaxPool op. ‘Partial’ because all the non-maximal
values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling the result of an unpooling operation should give back the original input to the unpooling op.
- MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous.
The third input argument, output_size, is meant to disambiguate the op and produce output tensor of known/predictable size.
- In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads,
which define the exact unpooling op. The attributes typically have the same values as the corrsponding pooling op that the unpooling op is trying to invert.
Attributes
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T1: Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
I (heterogeneous) - T2: Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x … x Dn).
output_shape (optional, heterogeneous) - T2: The shape of the output can be explicitly set which will cause pads values to be auto generated. If ‘output_shape’ is specified, ‘pads’ values are ignored.
Outputs
output (heterogeneous) - T1: Output data tensor that contains the result of the unpooling.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T2 in ( tensor(int64) ): Constrain index tensor to int64
OnnxMaxUnpool_9#
- class mlprodict.npy.xop_auto_import_.OnnxMaxUnpool_9(*args, **kwargs)#
Version
name: MaxUnpool (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
- MaxUnpool essentially computes the partial inverse of the MaxPool op.
The input information to this op is typically the output information from a MaxPool op. The first input tensor X is the tensor that needs to be unpooled, which is typically the pooled tensor (first output) from MaxPool. The second input tensor, I, contains the indices to the (locally maximal) elements corrsponding to the elements in the first input tensor X. Input tensor I is typically the second output of the MaxPool op. The third (optional) input is a tensor that specifies the output size of the unpooling operation.
- MaxUnpool is intended to do ‘partial’ inverse of the MaxPool op. ‘Partial’ because all the non-maximal
values from the original input to MaxPool are set to zero in the output of the MaxUnpool op. Pooling the result of an unpooling operation should give back the original input to the unpooling op.
- MaxUnpool can produce the same output size for several input sizes, which makes unpooling op ambiguous.
The third input argument, output_size, is meant to disambiguate the op and produce output tensor of known/predictable size.
- In addition to the inputs, MaxUnpool takes three attributes, namely kernel_shape, strides, and pads,
which define the exact unpooling op. The attributes typically have the same values as the corrsponding pooling op that the unpooling op is trying to invert.
Attributes
kernel_shape (required): The size of the kernel along each axis.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0. The value represent the number of pixels added to the beginning and end part of the corresponding axis. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i. This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaults to 0 along start and end of each spatial axis.
strides: Stride along each spatial axis.
Inputs
Between 2 and 3 inputs.
X (heterogeneous) - T1: Input data tensor that has to be unpooled. This tensor is typically the first output of the MaxPool op.Dimensions for image case are (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data. For non-image case, the dimensions are in the form of (N x C x D1 x D2 … Dn), where N is the batch size. Optionally, if dimension denotation is in effect, the operation expects the input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
I (heterogeneous) - T2: Input data tensor containing the indices corresponding to elements in the first input tensor X.This tensor is typically the second output of the MaxPool op.Dimensions must be the same as input tensor X. The indices are linear, i.e. computed considering the tensor as flattened 1-D tensor, assuming row-major storage. Also, the linear indices should not consider padding. So the values in indices are in the range [0, N x C x D1 x … x Dn).
output_shape (optional, heterogeneous) - T2: The shape of the output can be explicitly set which will cause pads values to be auto generated. If ‘output_shape’ is specified, ‘pads’ values are ignored.
Outputs
output (heterogeneous) - T1: Output data tensor that contains the result of the unpooling.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T2 in ( tensor(int64) ): Constrain index tensor to int64
OnnxMax_1#
- class mlprodict.npy.xop_auto_import_.OnnxMax_1(*args, **kwargs)#
Version
name: Max (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Element-wise max of each of the input tensors. All inputs and outputs must have the same shape and data type.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Max.
Outputs
max (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMax_12#
- class mlprodict.npy.xop_auto_import_.OnnxMax_12(*args, **kwargs)#
Version
name: Max (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Element-wise max of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for max.
Outputs
max (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
OnnxMax_13#
- class mlprodict.npy.xop_auto_import_.OnnxMax_13(*args, **kwargs)#
Version
name: Max (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise max of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for max.
Outputs
max (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
OnnxMax_6#
- class mlprodict.npy.xop_auto_import_.OnnxMax_6(*args, **kwargs)#
Version
name: Max (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Element-wise max of each of the input tensors. All inputs and outputs must have the same shape and data type.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Max.
Outputs
max (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMax_8#
- class mlprodict.npy.xop_auto_import_.OnnxMax_8(*args, **kwargs)#
Version
name: Max (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
Element-wise max of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for max.
Outputs
max (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMean#
- class mlprodict.npy.xop_auto_import_.OnnxMean(*args, **kwargs)#
Version
name: Mean (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise mean of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for mean.
Outputs
mean (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMeanVarianceNormalization#
- class mlprodict.npy.xop_auto_import_.OnnxMeanVarianceNormalization(*args, **kwargs)#
Version
domain: main
since_version: 13
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 13.
Summary
A MeanVarianceNormalization Function: Perform mean variance normalization on the input tensor X using formula: <br/>
` (X-EX)/sqrt(E(X-EX)^2) `Attributes
axes: A list of integers, along which to reduce. The default is to caculate along axes [0,2,3] for calculating mean and variance along each channel. Two variables with the same C-coordinate are associated with the same mean and variance. Default value is
[0 2 3].
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to all numeric tensors.
OnnxMeanVarianceNormalization_13#
- class mlprodict.npy.xop_auto_import_.OnnxMeanVarianceNormalization_13(*args, **kwargs)#
Version
domain: main
since_version: 13
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 13.
Summary
A MeanVarianceNormalization Function: Perform mean variance normalization on the input tensor X using formula: <br/>
` (X-EX)/sqrt(E(X-EX)^2) `Attributes
axes: A list of integers, along which to reduce. The default is to caculate along axes [0,2,3] for calculating mean and variance along each channel. Two variables with the same C-coordinate are associated with the same mean and variance. Default value is
[0 2 3].
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to all numeric tensors.
OnnxMeanVarianceNormalization_9#
- class mlprodict.npy.xop_auto_import_.OnnxMeanVarianceNormalization_9(*args, **kwargs)#
Version
domain: main
since_version: 9
function: True
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 9.
Summary
A MeanVarianceNormalization Function: Perform mean variance normalization on the input tensor X using formula: <br/>
` (X-EX)/sqrt(E(X-EX)^2) `Attributes
axes: A list of integers, along which to reduce. The default is to caculate along axes [0,2,3] for calculating mean and variance along each channel. Two variables with the same C-coordinate are associated with the same mean and variance. Default value is
[0 2 3].
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to all numeric tensors.
OnnxMean_1#
- class mlprodict.npy.xop_auto_import_.OnnxMean_1(*args, **kwargs)#
Version
name: Mean (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Element-wise mean of each of the input tensors. All inputs and outputs must have the same shape and data type.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Mean.
Outputs
mean (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMean_13#
- class mlprodict.npy.xop_auto_import_.OnnxMean_13(*args, **kwargs)#
Version
name: Mean (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise mean of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for mean.
Outputs
mean (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMean_6#
- class mlprodict.npy.xop_auto_import_.OnnxMean_6(*args, **kwargs)#
Version
name: Mean (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Element-wise mean of each of the input tensors. All inputs and outputs must have the same shape and data type.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Mean.
Outputs
mean (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMean_8#
- class mlprodict.npy.xop_auto_import_.OnnxMean_8(*args, **kwargs)#
Version
name: Mean (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
Element-wise mean of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for mean.
Outputs
mean (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMelWeightMatrix#
- class mlprodict.npy.xop_auto_import_.OnnxMelWeightMatrix(*args, **kwargs)#
Version
name: MelWeightMatrix (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. This function defines the mel scale in terms of a frequency in hertz according to the following formula:
mel(f) = 2595 * log10(1 + f/700)
In the returned matrix, all the triangles (filterbanks) have a peak value of 1.0.
The returned MelWeightMatrix can be used to right-multiply a spectrogram S of shape [frames, num_spectrogram_bins] of linear scale spectrum values (e.g. STFT magnitudes) to generate a “mel spectrogram” M of shape [frames, num_mel_bins].
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T3. The default value is 1 = FLOAT. Default value is
1.
Inputs
num_mel_bins (heterogeneous) - T1: The number of bands in the mel spectrum.
dft_length (heterogeneous) - T1: The size of the original DFT. The size of the original DFT is used to infer the size of the onesided DFT, which is understood to be floor(dft_length/2) + 1, i.e. the spectrogram only contains the nonredundant DFT bins.
sample_rate (heterogeneous) - T1: Samples per second of the input signal used to create the spectrogram. Used to figure out the frequencies corresponding to each spectrogram bin, which dictates how they are mapped into the mel scale.
lower_edge_hertz (heterogeneous) - T2: Lower bound on the frequencies to be included in the mel spectrum. This corresponds to the lower edge of the lowest triangular band.
upper_edge_hertz (heterogeneous) - T2: The desired top edge of the highest frequency band.
Outputs
output (heterogeneous) - T3: The Mel Weight Matrix. The output has the shape: [floor(dft_length/2) + 1][num_mel_bins].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain to integer tensors.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain to float tensors
T3 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any numerical types.
OnnxMelWeightMatrix_17#
- class mlprodict.npy.xop_auto_import_.OnnxMelWeightMatrix_17(*args, **kwargs)#
Version
name: MelWeightMatrix (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. This function defines the mel scale in terms of a frequency in hertz according to the following formula:
mel(f) = 2595 * log10(1 + f/700)
In the returned matrix, all the triangles (filterbanks) have a peak value of 1.0.
The returned MelWeightMatrix can be used to right-multiply a spectrogram S of shape [frames, num_spectrogram_bins] of linear scale spectrum values (e.g. STFT magnitudes) to generate a “mel spectrogram” M of shape [frames, num_mel_bins].
Attributes
output_datatype: The data type of the output tensor. Strictly must be one of the values from DataType enum in TensorProto whose values correspond to T3. The default value is 1 = FLOAT. Default value is
1.
Inputs
num_mel_bins (heterogeneous) - T1: The number of bands in the mel spectrum.
dft_length (heterogeneous) - T1: The size of the original DFT. The size of the original DFT is used to infer the size of the onesided DFT, which is understood to be floor(dft_length/2) + 1, i.e. the spectrogram only contains the nonredundant DFT bins.
sample_rate (heterogeneous) - T1: Samples per second of the input signal used to create the spectrogram. Used to figure out the frequencies corresponding to each spectrogram bin, which dictates how they are mapped into the mel scale.
lower_edge_hertz (heterogeneous) - T2: Lower bound on the frequencies to be included in the mel spectrum. This corresponds to the lower edge of the lowest triangular band.
upper_edge_hertz (heterogeneous) - T2: The desired top edge of the highest frequency band.
Outputs
output (heterogeneous) - T3: The Mel Weight Matrix. The output has the shape: [floor(dft_length/2) + 1][num_mel_bins].
Type Constraints
T1 in ( tensor(int32), tensor(int64) ): Constrain to integer tensors.
T2 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain to float tensors
T3 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any numerical types.
OnnxMin#
- class mlprodict.npy.xop_auto_import_.OnnxMin(*args, **kwargs)#
Version
name: Min (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise min of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for min.
Outputs
min (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
OnnxMin_1#
- class mlprodict.npy.xop_auto_import_.OnnxMin_1(*args, **kwargs)#
Version
name: Min (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Element-wise min of each of the input tensors. All inputs and outputs must have the same shape and data type.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Min
Outputs
min (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMin_12#
- class mlprodict.npy.xop_auto_import_.OnnxMin_12(*args, **kwargs)#
Version
name: Min (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Element-wise min of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for min.
Outputs
min (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
OnnxMin_13#
- class mlprodict.npy.xop_auto_import_.OnnxMin_13(*args, **kwargs)#
Version
name: Min (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise min of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for min.
Outputs
min (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
OnnxMin_6#
- class mlprodict.npy.xop_auto_import_.OnnxMin_6(*args, **kwargs)#
Version
name: Min (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Element-wise min of each of the input tensors. All inputs and outputs must have the same shape and data type.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Min
Outputs
min (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMin_8#
- class mlprodict.npy.xop_auto_import_.OnnxMin_8(*args, **kwargs)#
Version
name: Min (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
Element-wise min of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for min.
Outputs
min (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMish#
- class mlprodict.npy.xop_auto_import_.OnnxMish(*args, **kwargs)#
Version
name: Mish (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Mish: A Self Regularized Non-Monotonic Neural Activation Function.
Perform the linear unit element-wise on the input tensor X using formula:
mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + e^{x}))
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input X and output types to float tensors.
OnnxMish_18#
- class mlprodict.npy.xop_auto_import_.OnnxMish_18(*args, **kwargs)#
Version
name: Mish (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Mish: A Self Regularized Non-Monotonic Neural Activation Function.
Perform the linear unit element-wise on the input tensor X using formula:
mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + e^{x}))
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input X and output types to float tensors.
OnnxMod#
- class mlprodict.npy.xop_auto_import_.OnnxMod(*args, **kwargs)#
Version
name: Mod (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
- Performs element-wise binary modulus (with Numpy-style broadcasting support).
The sign of the remainder is the same as that of the Divisor.
Mod operator can also behave like C fmod() or numpy.fmod. In this case, the sign of the remainder however, will be the same as the Dividend (in contrast to integer mod). To force a behavior like numpy.fmod() an ‘fmod’ Attribute is provided. This attribute is set to 0 by default causing the behavior to be like integer mod. Setting this attribute to 1 causes the remainder to be calculated similar to that of numpy.fmod().
If the input type is floating point, then fmod attribute must be set to 1.
In case of dividend being zero, the results will be platform dependent.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Attributes
fmod: Whether the operator should behave like fmod (default=0 meaning it will do integer mods); Set this to 1 to force fmod treatment Default value is
0.
Inputs
A (heterogeneous) - T: Dividend tensor
B (heterogeneous) - T: Divisor tensor
Outputs
C (heterogeneous) - T: Remainder tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision numeric tensors.
OnnxMod_10#
- class mlprodict.npy.xop_auto_import_.OnnxMod_10(*args, **kwargs)#
Version
name: Mod (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
- Performs element-wise binary modulus (with Numpy-style broadcasting support).
The sign of the remainder is the same as that of the Divisor.
Mod operator can also behave like C fmod() or numpy.fmod. In this case, the sign of the remainder however, will be the same as the Dividend (in contrast to integer mod). To force a behavior like numpy.fmod() an ‘fmod’ Attribute is provided. This attribute is set to 0 by default causing the behavior to be like integer mod. Setting this attribute to 1 causes the remainder to be calculated similar to that of numpy.fmod().
If the input type is floating point, then fmod attribute must be set to 1.
In case of dividend being zero, the results will be platform dependent.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Attributes
fmod: Whether the operator should behave like fmod (default=0 meaning it will do integer mods); Set this to 1 to force fmod treatment Default value is
0.
Inputs
A (heterogeneous) - T: Dividend tensor
B (heterogeneous) - T: Divisor tensor
Outputs
C (heterogeneous) - T: Remainder tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision numeric tensors.
OnnxMod_13#
- class mlprodict.npy.xop_auto_import_.OnnxMod_13(*args, **kwargs)#
Version
name: Mod (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
- Performs element-wise binary modulus (with Numpy-style broadcasting support).
The sign of the remainder is the same as that of the Divisor.
Mod operator can also behave like C fmod() or numpy.fmod. In this case, the sign of the remainder however, will be the same as the Dividend (in contrast to integer mod). To force a behavior like numpy.fmod() an ‘fmod’ Attribute is provided. This attribute is set to 0 by default causing the behavior to be like integer mod. Setting this attribute to 1 causes the remainder to be calculated similar to that of numpy.fmod().
If the input type is floating point, then fmod attribute must be set to 1.
In case of dividend being zero, the results will be platform dependent.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Attributes
fmod: Whether the operator should behave like fmod (default=0 meaning it will do integer mods); Set this to 1 to force fmod treatment Default value is
0.
Inputs
A (heterogeneous) - T: Dividend tensor
B (heterogeneous) - T: Divisor tensor
Outputs
C (heterogeneous) - T: Remainder tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision numeric tensors.
OnnxMul#
- class mlprodict.npy.xop_auto_import_.OnnxMul(*args, **kwargs)#
Version
name: Mul (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary multiplication (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxMul_1#
- class mlprodict.npy.xop_auto_import_.OnnxMul_1(*args, **kwargs)#
Version
name: Mul (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Performs element-wise binary multiplication (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.consumed_inputs: legacy optimization attribute.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxMul_13#
- class mlprodict.npy.xop_auto_import_.OnnxMul_13(*args, **kwargs)#
Version
name: Mul (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Performs element-wise binary multiplication (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxMul_14#
- class mlprodict.npy.xop_auto_import_.OnnxMul_14(*args, **kwargs)#
Version
name: Mul (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary multiplication (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxMul_6#
- class mlprodict.npy.xop_auto_import_.OnnxMul_6(*args, **kwargs)#
Version
name: Mul (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Performs element-wise binary multiplication (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxMul_7#
- class mlprodict.npy.xop_auto_import_.OnnxMul_7(*args, **kwargs)#
Version
name: Mul (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Performs element-wise binary multiplication (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxMultinomial#
- class mlprodict.npy.xop_auto_import_.OnnxMultinomial(*args, **kwargs)#
Version
name: Multinomial (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Generate a tensor of samples from a multinomial distribution according to the probabilities of each of the possible outcomes.
Attributes
dtype: (Optional) The data type for the elements of the output tensor, if not specified, we will use int32. Default value is
6.sample_size: Number of times to sample. Default value is
1.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: Input tensor with shape [batch_size, class_size], where class_size is the number of all possible outcomes. Each value along the axis zero represents the unnormalized log-probability of each corresponding outcome in a batch.
Outputs
output (heterogeneous) - T2: Output tensor with shape [batch_size, sample_size], where sample_size is the number of times to sample. Each value along the axis zero represents the outcome of the corresponding sample in a batch.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(int32), tensor(int64) ): Constrain output types to integral tensors.
OnnxMultinomial_7#
- class mlprodict.npy.xop_auto_import_.OnnxMultinomial_7(*args, **kwargs)#
Version
name: Multinomial (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Generate a tensor of samples from a multinomial distribution according to the probabilities of each of the possible outcomes.
Attributes
dtype: (Optional) The data type for the elements of the output tensor, if not specified, we will use int32. Default value is
6.sample_size: Number of times to sample. Default value is
1.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: Input tensor with shape [batch_size, class_size], where class_size is the number of all possible outcomes. Each value along the axis zero represents the unnormalized log-probability of each corresponding outcome in a batch.
Outputs
output (heterogeneous) - T2: Output tensor with shape [batch_size, sample_size], where sample_size is the number of times to sample. Each value along the axis zero represents the outcome of the corresponding sample in a batch.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T2 in ( tensor(int32), tensor(int64) ): Constrain output types to integral tensors.
OnnxNeg#
- class mlprodict.npy.xop_auto_import_.OnnxNeg(*args, **kwargs)#
Version
name: Neg (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Neg takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where each element flipped sign, y = -x, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8) ): Constrain input and output types to signed numeric tensors.
OnnxNeg_1#
- class mlprodict.npy.xop_auto_import_.OnnxNeg_1(*args, **kwargs)#
Version
name: Neg (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Neg takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where each element flipped sign, y = -x, is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxNeg_13#
- class mlprodict.npy.xop_auto_import_.OnnxNeg_13(*args, **kwargs)#
Version
name: Neg (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Neg takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where each element flipped sign, y = -x, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8) ): Constrain input and output types to signed numeric tensors.
OnnxNeg_6#
- class mlprodict.npy.xop_auto_import_.OnnxNeg_6(*args, **kwargs)#
Version
name: Neg (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Neg takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where each element flipped sign, y = -x, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8) ): Constrain input and output types to signed numeric tensors.
OnnxNegativeLogLikelihoodLoss#
- class mlprodict.npy.xop_auto_import_.OnnxNegativeLogLikelihoodLoss(*args, **kwargs)#
Version
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. Its “input” tensor has the shape of (N, C, d1, d2, …, dk) where k >= 0. The “input” tensor contains log-probabilities for input[n, :, d_1, d_2,…, d_k] being in a class of [0, C). The operator’s “target” input tensor has the shape of (N, d1, d2, …, dk). It encodes class labels (one of C classes) or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x … x dk samples. The loss value for input[n, :, d_1, d_2,…d_k] being classified as class c = target[n][d_1][d_2]…[d_k] is computed as:
loss[n][d_1][d_2]…[d_k] = -input[n][c][d_1][d_2]…[d_k].
When an optional “weight” is provided, the sample loss is calculated as:
loss[n][d_1][d_2]…[d_k] = -input[n][c][d_1][d_2]…[d_k] * weight[c].
loss is zero for the case when target-value equals ignore_index.
loss[n][d_1][d_2]…[d_k] = 0, when target[n][d_1][d_2]…[d_k] = ignore_index
If “reduction” attribute is set to “none”, the operator’s output will be the above loss with shape (N, d1, d2, …, dk). If “reduction” attribute is set to “mean” (the default attribute value), the output loss is (weight) averaged:
mean(loss), if “weight” is not provided,
or if weight is provided,
sum(loss) / sum(weight[target[n][d_1][d_2]…[d_k]]]), for all samples.
- If “reduction” attribute is set to “sum”, the output is a scalar:
sum(loss).
See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss.
Example 1:
// negative log likelihood loss, “none” reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]]
loss = np.zeros((N, d1)) for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1]
// print(loss) // [[-3. -2.] // [-0. -2.]]
Example 2:
// weighted negative log likelihood loss, sum reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] weight = [0.2, 0.3, 0.1] loss = np.zeros((N, d1)) for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1] * weight[c]
loss = np.sum(loss) // print(loss) // -1.1
Example 3:
// weighted negative log likelihood loss, mean reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] weight = [0.2, 0.3, 0.1] loss = np.zeros((N, d1)) weight_total = 0 for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1] * weight[c] weight_total = weight_total + weight[c]
loss = np.sum(loss) / weight_total // print(loss) // -1.57
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. It’s an optional value.
reduction: Type of reduction to apply to loss: none, sum, mean (default). ‘none’: the output is the loss for each sample. ‘sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the sum of applied weights. Default value is
'mean'.
Inputs
Between 2 and 3 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input, weight, and output types to floating-point tensors.
Tind in ( tensor(int32), tensor(int64) ): Constrain target to integer types
OnnxNegativeLogLikelihoodLoss_12#
- class mlprodict.npy.xop_auto_import_.OnnxNegativeLogLikelihoodLoss_12(*args, **kwargs)#
Version
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. Its “input” tensor has the shape of (N, C, d1, d2, …, dk) where k >= 0. The “input” tensor contains log-probabilities for input[n, :, d_1, d_2,…, d_k] being in a class of [0, C). The operator’s “target” input tensor has the shape of (N, d1, d2, …, dk). It encodes class labels (one of C classes) or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x … x dk samples. The loss value for input[n, :, d_1, d_2,…d_k] being classified as class c = target[n][d_1][d_2]…[d_k] is computed as:
loss[n][d_1][d_2]…[d_k] = -input[n][c][d_1][d_2]…[d_k].
- When an optional “weight” is provided, the sample loss is calculated as:
loss[n][d_1][d_2]…[d_k] = -input[n][c][d_1][d_2]…[d_k] * weight[c].
loss is zero for the case when target-value equals ignore_index.
loss[n][d_1][d_2]…[d_k] = 0, when target[n][d_1][d_2]…[d_k] = ignore_index
If “reduction” attribute is set to “none”, the operator’s output will be the above loss with shape (N, d1, d2, …, dk). If “reduction” attribute is set to “mean” (the default attribute value), the output loss is (weight) averaged:
mean(loss), if “weight” is not provided,
- or if weight is provided,
sum(loss) / sum(weight[target[n][d_1][d_2]…[d_k]]]), for all samples.
- If “reduction” attribute is set to “sum”, the output is a scalar:
sum(loss).
See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss. Example 1:
// negative log likelihood loss, “none” reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] loss = np.zeros((N, d1)) for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1]
// print(loss) // [[-3. -2.] // [-0. -2.]]
- Example 2:
// weighted negative log likelihood loss, sum reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] weight = [0.2, 0.3, 0.1] loss = np.zeros((N, d1)) for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1] * weight[c]
loss = np.sum(loss) // print(loss) // -1.1
- Example 3:
// weighted negative log likelihood loss, mean reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] weight = [0.2, 0.3, 0.1] loss = np.zeros((N, d1)) weight_total = 0 for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1] * weight[c] weight_total = weight_total + weight[c]
loss = np.sum(loss) / weight_total // print(loss) // -1.57
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. It’s an optional value.
reduction: Type of reduction to apply to loss: none, sum, mean (default). ‘none’: the output is the loss for each sample. ‘sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the sum of applied weights. Default value is
'mean'.
Inputs
Between 2 and 3 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input, weight, and output types to floating-point tensors.
Tind in ( tensor(int32), tensor(int64) ): Constrain target to integer types
OnnxNegativeLogLikelihoodLoss_13#
- class mlprodict.npy.xop_auto_import_.OnnxNegativeLogLikelihoodLoss_13(*args, **kwargs)#
Version
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
A NegativeLogLikelihoodLoss operator computes (weighted) negative log likelihood loss. Its “input” tensor has the shape of (N, C, d1, d2, …, dk) where k >= 0. The “input” tensor contains log-probabilities for input[n, :, d_1, d_2,…, d_k] being in a class of [0, C). The operator’s “target” input tensor has the shape of (N, d1, d2, …, dk). It encodes class labels (one of C classes) or it may contain a special value (indicated by an attribute ignore_index) for N x d1 x d2 x … x dk samples. The loss value for input[n, :, d_1, d_2,…d_k] being classified as class c = target[n][d_1][d_2]…[d_k] is computed as:
loss[n][d_1][d_2]…[d_k] = -input[n][c][d_1][d_2]…[d_k].
When an optional “weight” is provided, the sample loss is calculated as:
loss[n][d_1][d_2]…[d_k] = -input[n][c][d_1][d_2]…[d_k] * weight[c].
loss is zero for the case when target-value equals ignore_index.
loss[n][d_1][d_2]…[d_k] = 0, when target[n][d_1][d_2]…[d_k] = ignore_index
If “reduction” attribute is set to “none”, the operator’s output will be the above loss with shape (N, d1, d2, …, dk). If “reduction” attribute is set to “mean” (the default attribute value), the output loss is (weight) averaged:
mean(loss), if “weight” is not provided,
or if weight is provided,
sum(loss) / sum(weight[target[n][d_1][d_2]…[d_k]]]), for all samples.
- If “reduction” attribute is set to “sum”, the output is a scalar:
sum(loss).
See also https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss.
Example 1:
// negative log likelihood loss, “none” reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]]
loss = np.zeros((N, d1)) for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1]
// print(loss) // [[-3. -2.] // [-0. -2.]]
Example 2:
// weighted negative log likelihood loss, sum reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] weight = [0.2, 0.3, 0.1] loss = np.zeros((N, d1)) for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1] * weight[c]
loss = np.sum(loss) // print(loss) // -1.1
Example 3:
// weighted negative log likelihood loss, mean reduction N, C, d1 = 2, 3, 2 input = [[[1.0, 2.0], [2.0, 2.0], [3.0, 2.0]],
[[0.0, 1.0], [2.0, 2.0], [1.0, 2]]]
target = [[2, 1], [0, 2]] weight = [0.2, 0.3, 0.1] loss = np.zeros((N, d1)) weight_total = 0 for n in range(N):
- for d_1 in range(d1):
c = target[n][d_1] loss[n][d_1] = -input[n][c][d_1] * weight[c] weight_total = weight_total + weight[c]
loss = np.sum(loss) / weight_total // print(loss) // -1.57
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. It’s an optional value.
reduction: Type of reduction to apply to loss: none, sum, mean (default). ‘none’: the output is the loss for each sample. ‘sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the sum of applied weights. Default value is
'mean'.
Inputs
Between 2 and 3 inputs.
input (heterogeneous) - T: Input tensor of shape (N, C) or (N, C, d1, d2, …, dk).
target (heterogeneous) - Tind: Target tensor of shape (N) or (N, d1, d2, …, dk). Target element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the target values should either be in the range [0, C) or have the value ignore_index.
weight (optional, heterogeneous) - T: Optional rescaling weight tensor. If given, it has to be a tensor of size C. Otherwise, it is treated as if having all ones.
Outputs
loss (heterogeneous) - T: The negative log likelihood loss
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input, weight, and output types to floating-point tensors.
Tind in ( tensor(int32), tensor(int64) ): Constrain target to integer types
OnnxNonMaxSuppression#
- class mlprodict.npy.xop_auto_import_.OnnxNonMaxSuppression(*args, **kwargs)#
Version
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. Bounding boxes with score less than score_threshold are removed. Bounding box format is indicated by attribute center_point_box. Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system result in the same boxes being selected by the algorithm. The selected_indices output is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. The bounding box coordinates corresponding to the selected indices can then be obtained using the Gather or GatherND operation.
Attributes
center_point_box: Integer indicate the format of the box data. The default is 0. 0 - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (i.e., lying in the interval [0, 1]) or absolute. Mostly used for TF models. 1 - the box data is supplied as [x_center, y_center, width, height]. Mostly used for Pytorch models. Default value is
0.
Inputs
Between 2 and 5 inputs.
boxes (heterogeneous) - tensor(float): An input tensor with shape [num_batches, spatial_dimension, 4]. The single box data format is indicated by center_point_box.
scores (heterogeneous) - tensor(float): An input tensor with shape [num_batches, num_classes, spatial_dimension]
max_output_boxes_per_class (optional, heterogeneous) - tensor(int64): Integer representing the maximum number of boxes to be selected per batch per class. It is a scalar. Default to 0, which means no output.
iou_threshold (optional, heterogeneous) - tensor(float): Float representing the threshold for deciding whether boxes overlap too much with respect to IOU. It is scalar. Value range [0, 1]. Default to 0.
score_threshold (optional, heterogeneous) - tensor(float): Float representing the threshold for deciding when to remove boxes based on score. It is a scalar.
Outputs
selected_indices (heterogeneous) - tensor(int64): selected indices from the boxes tensor. [num_selected_indices, 3], the selected index format is [batch_index, class_index, box_index].
OnnxNonMaxSuppression_10#
- class mlprodict.npy.xop_auto_import_.OnnxNonMaxSuppression_10(*args, **kwargs)#
Version
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. Bounding boxes with score less than score_threshold are removed. Bounding box format is indicated by attribute center_point_box. Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system result in the same boxes being selected by the algorithm. The selected_indices output is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. The bounding box coordinates corresponding to the selected indices can then be obtained using the Gather or GatherND operation.
Attributes
center_point_box: Integer indicate the format of the box data. The default is 0. 0 - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (i.e., lying in the interval [0, 1]) or absolute. Mostly used for TF models. 1 - the box data is supplied as [x_center, y_center, width, height]. Mostly used for Pytorch models. Default value is
0.
Inputs
Between 2 and 5 inputs.
boxes (heterogeneous) - tensor(float): An input tensor with shape [num_batches, spatial_dimension, 4]. The single box data format is indicated by center_point_box.
scores (heterogeneous) - tensor(float): An input tensor with shape [num_batches, num_classes, spatial_dimension]
max_output_boxes_per_class (optional, heterogeneous) - tensor(int64): Integer representing the maximum number of boxes to be selected per batch per class. It is a scalar. Default to 0, which means no output.
iou_threshold (optional, heterogeneous) - tensor(float): Float representing the threshold for deciding whether boxes overlap too much with respect to IOU. It is scalar. Value range [0, 1]. Default to 0.
score_threshold (optional, heterogeneous) - tensor(float): Float representing the threshold for deciding when to remove boxes based on score. It is a scalar.
Outputs
selected_indices (heterogeneous) - tensor(int64): selected indices from the boxes tensor. [num_selected_indices, 3], the selected index format is [batch_index, class_index, box_index].
OnnxNonMaxSuppression_11#
- class mlprodict.npy.xop_auto_import_.OnnxNonMaxSuppression_11(*args, **kwargs)#
Version
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. Bounding boxes with score less than score_threshold are removed. Bounding box format is indicated by attribute center_point_box. Note that this algorithm is agnostic to where the origin is in the coordinate system and more generally is invariant to orthogonal transformations and translations of the coordinate system; thus translating or reflections of the coordinate system result in the same boxes being selected by the algorithm. The selected_indices output is a set of integers indexing into the input collection of bounding boxes representing the selected boxes. The bounding box coordinates corresponding to the selected indices can then be obtained using the Gather or GatherND operation.
Attributes
center_point_box: Integer indicate the format of the box data. The default is 0. 0 - the box data is supplied as [y1, x1, y2, x2] where (y1, x1) and (y2, x2) are the coordinates of any diagonal pair of box corners and the coordinates can be provided as normalized (i.e., lying in the interval [0, 1]) or absolute. Mostly used for TF models. 1 - the box data is supplied as [x_center, y_center, width, height]. Mostly used for Pytorch models. Default value is
0.
Inputs
Between 2 and 5 inputs.
boxes (heterogeneous) - tensor(float): An input tensor with shape [num_batches, spatial_dimension, 4]. The single box data format is indicated by center_point_box.
scores (heterogeneous) - tensor(float): An input tensor with shape [num_batches, num_classes, spatial_dimension]
max_output_boxes_per_class (optional, heterogeneous) - tensor(int64): Integer representing the maximum number of boxes to be selected per batch per class. It is a scalar. Default to 0, which means no output.
iou_threshold (optional, heterogeneous) - tensor(float): Float representing the threshold for deciding whether boxes overlap too much with respect to IOU. It is scalar. Value range [0, 1]. Default to 0.
score_threshold (optional, heterogeneous) - tensor(float): Float representing the threshold for deciding when to remove boxes based on score. It is a scalar.
Outputs
selected_indices (heterogeneous) - tensor(int64): selected indices from the boxes tensor. [num_selected_indices, 3], the selected index format is [batch_index, class_index, box_index].
OnnxNonZero#
- class mlprodict.npy.xop_auto_import_.OnnxNonZero(*args, **kwargs)#
Version
name: NonZero (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the indices of the elements that are non-zero (in row-major order - by dimension). NonZero behaves similar to numpy.nonzero: https://docs.scipy.org/doc/numpy/reference/generated/numpy.nonzero.html, but for scalar input, NonZero produces output shape (0, N) instead of (1, N), which is different from Numpy’s behavior.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - tensor(int64): output
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to all tensor types.
OnnxNonZero_13#
- class mlprodict.npy.xop_auto_import_.OnnxNonZero_13(*args, **kwargs)#
Version
name: NonZero (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Returns the indices of the elements that are non-zero (in row-major order - by dimension). NonZero behaves similar to numpy.nonzero: https://docs.scipy.org/doc/numpy/reference/generated/numpy.nonzero.html, but for scalar input, NonZero produces output shape (0, N) instead of (1, N), which is different from Numpy’s behavior.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - tensor(int64): output
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to all tensor types.
OnnxNonZero_9#
- class mlprodict.npy.xop_auto_import_.OnnxNonZero_9(*args, **kwargs)#
Version
name: NonZero (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Returns the indices of the elements that are non-zero (in row-major order - by dimension). NonZero behaves similar to numpy.nonzero: https://docs.scipy.org/doc/numpy/reference/generated/numpy.nonzero.html, but for scalar input, NonZero produces output shape (0, N) instead of (1, N), which is different from Numpy’s behavior.
Inputs
X (heterogeneous) - T: input
Outputs
Y (heterogeneous) - tensor(int64): output
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to all tensor types.
OnnxNot#
- class mlprodict.npy.xop_auto_import_.OnnxNot(*args, **kwargs)#
Version
name: Not (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the negation of the input tensor element-wise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bool) ): Constrain input/output to boolean tensors.
OnnxNot_1#
- class mlprodict.npy.xop_auto_import_.OnnxNot_1(*args, **kwargs)#
Version
name: Not (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the negation of the input tensor element-wise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bool) ): Constrain input/output to boolean tensors.
OnnxOneHot#
- class mlprodict.npy.xop_auto_import_.OnnxOneHot(*args, **kwargs)#
Version
name: OneHot (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Produces a one-hot tensor based on inputs. The locations represented by the index values in the ‘indices’ input tensor will have ‘on_value’ and the other locations will have ‘off_value’ in the output tensor, where ‘on_value’ and ‘off_value’ are specified as part of required input argument ‘values’, which is a two-element tensor of format [off_value, on_value]. The rank of the output tensor will be one greater than the rank of the input tensor. The additional dimension is for one-hot representation. The additional dimension will be inserted at the position specified by ‘axis’. If ‘axis’ is not specified then then additional dimension will be inserted as the innermost dimension, i.e. axis=-1. The size of the additional dimension is specified by required scalar input ‘depth’. The type of the output tensor is the same as the type of the ‘values’ input. Any entries in the ‘indices’ input tensor with values outside the range [-depth, depth-1] will result in one-hot representation with all ‘off_value’ values in the output tensor.
when axis = 0: output[input[i, j, k], i, j, k] = 1 for all i, j, k and 0 otherwise.
when axis = -1: output[i, j, k, input[i, j, k]] = 1 for all i, j, k and 0 otherwise.
Attributes
axis: (Optional) Axis along which one-hot representation in added. Default: axis=-1. axis=-1 means that the additional dimension will be inserted as the innermost/last dimension in the output tensor. Negative value means counting dimensions from the back. Accepted range is [-r-1, r] where r = rank(indices). Default value is
-1.
Inputs
indices (heterogeneous) - T1: Input tensor containing indices. Any entries in the ‘indices’ input tensor with values outside the range [-depth, depth-1] will result in one-hot representation with all ‘off_value’ values in the output tensor.In case ‘indices’ is of non-integer type, the values will be casted to int64 before use.
depth (heterogeneous) - T2: Scalar specifying the number of classes in one-hot tensor. This is also the size of the one-hot dimension (specified by ‘axis’ attribute) added on in the output tensor. The values in the ‘indices’ input tensor are expected to be in the range [-depth, depth-1]. In case ‘depth’ is of non-integer type, it will be casted to int64 before use.
values (heterogeneous) - T3: Rank 1 tensor containing exactly two elements, in the format [off_value, on_value], where ‘on_value’ is the value used for filling locations specified in ‘indices’ input tensor, and ‘off_value’ is the value used for filling locations other than those specified in ‘indices’ input tensor.
Outputs
output (heterogeneous) - T3: Tensor of rank one greater than input tensor ‘indices’, i.e. rank(output) = rank(indices) + 1. The data type for the elements of the output tensor is the same as the type of input ‘values’ is used.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
T2 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
T3 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
OnnxOneHot_11#
- class mlprodict.npy.xop_auto_import_.OnnxOneHot_11(*args, **kwargs)#
Version
name: OneHot (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Produces a one-hot tensor based on inputs. The locations represented by the index values in the ‘indices’ input tensor will have ‘on_value’ and the other locations will have ‘off_value’ in the output tensor, where ‘on_value’ and ‘off_value’ are specified as part of required input argument ‘values’, which is a two-element tensor of format [off_value, on_value]. The rank of the output tensor will be one greater than the rank of the input tensor. The additional dimension is for one-hot representation. The additional dimension will be inserted at the position specified by ‘axis’. If ‘axis’ is not specified then then additional dimension will be inserted as the innermost dimension, i.e. axis=-1. The size of the additional dimension is specified by required scalar input ‘depth’. The type of the output tensor is the same as the type of the ‘values’ input. Any entries in the ‘indices’ input tensor with values outside the range [-depth, depth-1] will result in one-hot representation with all ‘off_value’ values in the output tensor.
when axis = 0: output[input[i, j, k], i, j, k] = 1 for all i, j, k and 0 otherwise.
when axis = -1: output[i, j, k, input[i, j, k]] = 1 for all i, j, k and 0 otherwise.
Attributes
axis: (Optional) Axis along which one-hot representation in added. Default: axis=-1. axis=-1 means that the additional dimension will be inserted as the innermost/last dimension in the output tensor. Negative value means counting dimensions from the back. Accepted range is [-r-1, r] where r = rank(indices). Default value is
-1.
Inputs
indices (heterogeneous) - T1: Input tensor containing indices. Any entries in the ‘indices’ input tensor with values outside the range [-depth, depth-1] will result in one-hot representation with all ‘off_value’ values in the output tensor.In case ‘indices’ is of non-integer type, the values will be casted to int64 before use.
depth (heterogeneous) - T2: Scalar specifying the number of classes in one-hot tensor. This is also the size of the one-hot dimension (specified by ‘axis’ attribute) added on in the output tensor. The values in the ‘indices’ input tensor are expected to be in the range [-depth, depth-1]. In case ‘depth’ is of non-integer type, it will be casted to int64 before use.
values (heterogeneous) - T3: Rank 1 tensor containing exactly two elements, in the format [off_value, on_value], where ‘on_value’ is the value used for filling locations specified in ‘indices’ input tensor, and ‘off_value’ is the value used for filling locations other than those specified in ‘indices’ input tensor.
Outputs
output (heterogeneous) - T3: Tensor of rank one greater than input tensor ‘indices’, i.e. rank(output) = rank(indices) + 1. The data type for the elements of the output tensor is the same as the type of input ‘values’ is used.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
T2 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
T3 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
OnnxOneHot_9#
- class mlprodict.npy.xop_auto_import_.OnnxOneHot_9(*args, **kwargs)#
Version
name: OneHot (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Produces a one-hot tensor based on inputs. The locations represented by the index values in the ‘indices’ input tensor will have ‘on_value’ and the other locations will have ‘off_value’ in the output tensor, where ‘on_value’ and ‘off_value’ are specified as part of required input argument ‘values’, which is a two-element tensor of format [off_value, on_value]. The rank of the output tensor will be one greater than the rank of the input tensor. The additional dimension is for one-hot representation. The additional dimension will be inserted at the position specified by ‘axis’. If ‘axis’ is not specified then then additional dimension will be inserted as the innermost dimension, i.e. axis=-1. The size of the additional dimension is specified by required scalar input ‘depth’. The type of the output tensor is the same as the type of the ‘values’ input. Any entries in the ‘indices’ input tensor with values outside the range [0, depth) will result in one-hot representation with all ‘off_value’ values in the output tensor.
Attributes
axis: (Optional) Axis along which one-hot representation in added. Default: axis=-1. axis=-1 means that the additional dimension will be inserted as the innermost/last dimension in the output tensor. Default value is
-1.
Inputs
indices (heterogeneous) - T1: Input tensor containing indices. The values must be non-negative integers. Any entries in the ‘indices’ input tensor with values outside the range [0, depth) will result in one-hot representation with all ‘off_value’ values in the output tensor.In case ‘indices’ is of non-integer type, the values will be casted to int64 before use.
depth (heterogeneous) - T2: Scalar specifying the number of classes in one-hot tensor. This is also the size of the one-hot dimension (specified by ‘axis’ attribute) added on in the output tensor. The values in the ‘indices’ input tensor are expected to be in the range [0, depth). In case ‘depth’ is of non-integer type, it will be casted to int64 before use.
values (heterogeneous) - T3: Rank 1 tensor containing exactly two elements, in the format [off_value, on_value], where ‘on_value’ is the value used for filling locations specified in ‘indices’ input tensor, and ‘off_value’ is the value used for filling locations other than those specified in ‘indices’ input tensor.
Outputs
output (heterogeneous) - T3: Tensor of rank one greater than input tensor ‘indices’, i.e. rank(output) = rank(indices) + 1. The data type for the elements of the output tensor is the same as the type of input ‘values’ is used.
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
T2 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
T3 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
OnnxOptional#
- class mlprodict.npy.xop_auto_import_.OnnxOptional(*args, **kwargs)#
Version
name: Optional (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Constructs an optional-type value containing either an empty optional of a certain type specified by the attribute, or a non-empty value containing the input element.
Attributes
type: Type of the element in the optional output
Inputs
Between 0 and 1 inputs.
input (optional, heterogeneous) - V: The input element.
Outputs
output (heterogeneous) - O: The optional output enclosing the input element.
Type Constraints
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input type to all tensor and sequence types.
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)) ): Constrain output type to all optional tensor or optional sequence types.
OnnxOptionalGetElement#
- class mlprodict.npy.xop_auto_import_.OnnxOptionalGetElement(*args, **kwargs)#
Version
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
If the input is a tensor or sequence type, it returns the input. If the input is an optional type, it outputs the element in the input. It is an error if the input is an empty optional-type (i.e. does not have an element) and the behavior is undefined in this case.
Inputs
input (heterogeneous) - O: The optional input.
Outputs
output (heterogeneous) - V: Output element in the optional input.
Type Constraints
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input type to optional tensor and optional sequence types.
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output type to all tensor or sequence types.
OnnxOptionalGetElement_15#
- class mlprodict.npy.xop_auto_import_.OnnxOptionalGetElement_15(*args, **kwargs)#
Version
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Outputs the element in the optional-type input. It is an error if the input value does not have an element and the behavior is undefined in this case.
Inputs
input (heterogeneous) - O: The optional input.
Outputs
output (heterogeneous) - V: Output element in the optional input.
Type Constraints
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)) ): Constrain input type to optional tensor and optional sequence types.
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output type to all tensor or sequence types.
OnnxOptionalGetElement_18#
- class mlprodict.npy.xop_auto_import_.OnnxOptionalGetElement_18(*args, **kwargs)#
Version
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
If the input is a tensor or sequence type, it returns the input. If the input is an optional type, it outputs the element in the input. It is an error if the input is an empty optional-type (i.e. does not have an element) and the behavior is undefined in this case.
Inputs
input (heterogeneous) - O: The optional input.
Outputs
output (heterogeneous) - V: Output element in the optional input.
Type Constraints
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input type to optional tensor and optional sequence types.
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain output type to all tensor or sequence types.
OnnxOptionalHasElement#
- class mlprodict.npy.xop_auto_import_.OnnxOptionalHasElement(*args, **kwargs)#
Version
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns true if (1) the input is an optional-type and contains an element, or, (2) the input is a tensor or sequence type. If the input is not provided or is an empty optional-type, this op returns false.
Inputs
Between 0 and 1 inputs.
input (optional, heterogeneous) - O: The optional input.
Outputs
output (heterogeneous) - B: A scalar boolean tensor. If true, it indicates that optional-type input contains an element. Otherwise, it is empty.
Type Constraints
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input type to optional tensor and optional sequence types.
B in ( tensor(bool) ): Constrain output to a boolean tensor.
OnnxOptionalHasElement_15#
- class mlprodict.npy.xop_auto_import_.OnnxOptionalHasElement_15(*args, **kwargs)#
Version
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Returns true if the optional-type input contains an element. If it is an empty optional-type, this op returns false.
Inputs
input (heterogeneous) - O: The optional input.
Outputs
output (heterogeneous) - B: A scalar boolean tensor. If true, it indicates that optional-type input contains an element. Otherwise, it is empty.
Type Constraints
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)) ): Constrain input type to optional tensor and optional sequence types.
B in ( tensor(bool) ): Constrain output to a boolean tensor.
OnnxOptionalHasElement_18#
- class mlprodict.npy.xop_auto_import_.OnnxOptionalHasElement_18(*args, **kwargs)#
Version
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Returns true if (1) the input is an optional-type and contains an element, or, (2) the input is a tensor or sequence type. If the input is not provided or is an empty optional-type, this op returns false.
Inputs
Between 0 and 1 inputs.
input (optional, heterogeneous) - O: The optional input.
Outputs
output (heterogeneous) - B: A scalar boolean tensor. If true, it indicates that optional-type input contains an element. Otherwise, it is empty.
Type Constraints
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)), seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input type to optional tensor and optional sequence types.
B in ( tensor(bool) ): Constrain output to a boolean tensor.
OnnxOptional_15#
- class mlprodict.npy.xop_auto_import_.OnnxOptional_15(*args, **kwargs)#
Version
name: Optional (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Constructs an optional-type value containing either an empty optional of a certain type specified by the attribute, or a non-empty value containing the input element.
Attributes
type: Type of the element in the optional output
Inputs
Between 0 and 1 inputs.
input (optional, heterogeneous) - V: The input element.
Outputs
output (heterogeneous) - O: The optional output enclosing the input element.
Type Constraints
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input type to all tensor and sequence types.
O in ( optional(seq(tensor(bool))), optional(seq(tensor(complex128))), optional(seq(tensor(complex64))), optional(seq(tensor(double))), optional(seq(tensor(float))), optional(seq(tensor(float16))), optional(seq(tensor(int16))), optional(seq(tensor(int32))), optional(seq(tensor(int64))), optional(seq(tensor(int8))), optional(seq(tensor(string))), optional(seq(tensor(uint16))), optional(seq(tensor(uint32))), optional(seq(tensor(uint64))), optional(seq(tensor(uint8))), optional(tensor(bool)), optional(tensor(complex128)), optional(tensor(complex64)), optional(tensor(double)), optional(tensor(float)), optional(tensor(float16)), optional(tensor(int16)), optional(tensor(int32)), optional(tensor(int64)), optional(tensor(int8)), optional(tensor(string)), optional(tensor(uint16)), optional(tensor(uint32)), optional(tensor(uint64)), optional(tensor(uint8)) ): Constrain output type to all optional tensor or optional sequence types.
OnnxOr#
- class mlprodict.npy.xop_auto_import_.OnnxOr(*args, **kwargs)#
Version
name: Or (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the or logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxOr_1#
- class mlprodict.npy.xop_auto_import_.OnnxOr_1(*args, **kwargs)#
Version
name: Or (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the tensor resulted from performing the or logical operation elementwise on the input tensors A and B.
If broadcasting is enabled, the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. See the doc of Add for a detailed description of the broadcasting rules.
Attributes
axis: If set, defines the broadcast dimensions.
broadcast: Enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: Left input tensor for the logical operator.
B (heterogeneous) - T: Right input tensor for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxOr_7#
- class mlprodict.npy.xop_auto_import_.OnnxOr_7(*args, **kwargs)#
Version
name: Or (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the or logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxOrgPytorchAtenATen#
- class mlprodict.npy.xop_auto_import_.OnnxOrgPytorchAtenATen(*args, **kwargs)#
Version
name: ATen (GitHub)
domain: org.pytorch.aten
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain org.pytorch.aten.
Summary
ATen
Attributes
operator (required): Name of ATen operator. Default value is
?.overload_name: Overload name of ATen operator. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic) - T: ATen Op inputs.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T: ATen Op outputs.
OnnxOrgPytorchAtenATen_1#
- class mlprodict.npy.xop_auto_import_.OnnxOrgPytorchAtenATen_1(*args, **kwargs)#
Version
name: ATen (GitHub)
domain: org.pytorch.aten
since_version: 1
function:
support_level:
shape inference:
This version of the operator has been available since version 1 of domain org.pytorch.aten.
Summary
ATen
Attributes
operator (required): Name of ATen operator. Default value is
?.overload_name: Overload name of ATen operator. Default value is
?.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic) - T: ATen Op inputs.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic) - T: ATen Op outputs.
OnnxPRelu#
- class mlprodict.npy.xop_auto_import_.OnnxPRelu(*args, **kwargs)#
Version
name: PRelu (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.
History - Version 16 adds bfloat16 to the types allowed. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: Input tensor
slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X
Outputs
Y (heterogeneous) - T: Output tensor (same size as X)
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxPRelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxPRelu_1(*args, **kwargs)#
Version
name: PRelu (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
slope (heterogeneous) - T: Slope tensor. If Slope is of size 1, the value is sharedacross different channels
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxPRelu_16#
- class mlprodict.npy.xop_auto_import_.OnnxPRelu_16(*args, **kwargs)#
Version
name: PRelu (GitHub)
domain: main
since_version: 16
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.
History - Version 16 adds bfloat16 to the types allowed. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: Input tensor
slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X
Outputs
Y (heterogeneous) - T: Output tensor (same size as X)
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxPRelu_6#
- class mlprodict.npy.xop_auto_import_.OnnxPRelu_6(*args, **kwargs)#
Version
name: PRelu (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
slope (heterogeneous) - T: Slope tensor. If Slope is of size 1, the value is sharedacross different channels
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxPRelu_7#
- class mlprodict.npy.xop_auto_import_.OnnxPRelu_7(*args, **kwargs)#
Version
name: PRelu (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: Input tensor
slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X
Outputs
Y (heterogeneous) - T: Output tensor (same size as X)
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxPRelu_9#
- class mlprodict.npy.xop_auto_import_.OnnxPRelu_9(*args, **kwargs)#
Version
name: PRelu (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
PRelu takes input data (Tensor<T>) and slope tensor as input, and produces one output data (Tensor<T>) where the function f(x) = slope * x for x < 0, f(x) = x for x >= 0., is applied to the data tensor elementwise. This operator supports unidirectional broadcasting (tensor slope should be unidirectional broadcastable to input tensor X); for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: Input tensor
slope (heterogeneous) - T: Slope tensor. The shape of slope can be smaller then first input X; if so, its shape must be unidirectional broadcastable to X
Outputs
Y (heterogeneous) - T: Output tensor (same size as X)
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to float/int tensors.
OnnxPad#
- class mlprodict.npy.xop_auto_import_.OnnxPad(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Given a tensor containing the data to be padded (data), a tensor containing the number of start and end pad values for axis (pads), (optionally) a mode, and (optionally) constant_value, a padded tensor (output) is generated.
The three supported modes are (similar to corresponding modes supported by numpy.pad):
constant`(default) - pads with a given constant value as specified by `constant_value (which defaults to 0, empty string, or False)
reflect - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis
edge - pads with the edge values of array
- Example 1 (constant mode):
Insert 0 pads to the beginning of the second dimension.
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘constant’
constant_value = 0.0
output = [
[0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7],
]
- Example 2 (reflect mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘reflect’
output = [
[1.0, 1.2, 1.0, 1.2], [2.3, 3.4, 2.3, 3.4], [4.5, 5.7, 4.5, 5.7],
]
- Example 3 (edge mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘edge’
output = [
[1.0, 1.0, 1.0, 1.2], [2.3, 2.3, 2.3, 3.4], [4.5, 4.5, 4.5, 5.7],
]
Attributes
mode: Supported modes: constant`(default), `reflect, edge Default value is
'constant'.
Inputs
Between 2 and 4 inputs.
data (heterogeneous) - T: Input tensor.
pads (heterogeneous) - tensor(int64): Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. pads should be a 1D tensor of shape [2 * num_axes] where num_axes refers to the number of elements in the axes input or the input rank if axes are not provided explicitly. pads format should be: [x1_begin, x2_begin, …, x1_end, x2_end,…], where xi_begin is the number of pad values added at the beginning of axis axes[i] and xi_end, the number of pad values added at the end of axis axes[i].
constant_value (optional, heterogeneous) - T: (Optional) A scalar value to be used if the mode chosen is constant (by default it is 0, empty string or False).
axes (optional, heterogeneous) - Tind: 1-D tensor of axes that pads apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed ([0, 1, …, input_rank-1]).
Outputs
output (heterogeneous) - T: Tensor after padding.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxPad_1#
- class mlprodict.npy.xop_auto_import_.OnnxPad_1(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Given data tensor, paddings, mode, and value. Example:
Insert 0 paddings to the beginning of the second dimension. data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
] paddings = [0, 0, 2, 0] output = [
- [
[0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7],
],
]
Attributes
mode: Three modes: constant(default), reflect, edge Default value is
'constant'.paddings (required): List of integers indicate the padding element count at the beginning and end of each axis, for 2D it is the number of pixel. paddings rank should be double of the input’s rank. paddings format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.
value: One float, indicates the value to be filled, default is 0 Default value is
0.0.
Inputs
data (heterogeneous) - T: Input tensor.
Outputs
output (heterogeneous) - T: Tensor after padding.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxPad_11#
- class mlprodict.npy.xop_auto_import_.OnnxPad_11(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Given a tensor containing the data to be padded (data), a tensor containing the number of start and end pad values for axis (pads), (optionally) a mode, and (optionally) constant_value, a padded tensor (output) is generated.
The three supported modes are (similar to corresponding modes supported by numpy.pad):
constant`(default) - pads with a given constant value as specified by `constant_value (which defaults to 0)
reflect - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis
edge - pads with the edge values of array
- Example 1 (constant mode):
Insert 0 pads to the beginning of the second dimension.
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘constant’
constant_value = 0.0
output = [
[0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7],
]
- Example 2 (reflect mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘reflect’
output = [
[1.0, 1.2, 1.0, 1.2], [2.3, 3.4, 2.3, 3.4], [4.5, 5.7, 4.5, 5.7],
]
- Example 3 (edge mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘edge’
output = [
[1.0, 1.0, 1.0, 1.2], [2.3, 2.3, 2.3, 3.4], [4.5, 4.5, 4.5, 5.7],
]
Attributes
mode: Supported modes: constant`(default), `reflect, edge Default value is
'constant'.
Inputs
Between 2 and 3 inputs.
data (heterogeneous) - T: Input tensor.
pads (heterogeneous) - tensor(int64): Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. pads should be a 1D tensor of shape [2 * input_rank]. pads format should be: [x1_begin, x2_begin,…,x1_end, x2_end,…], where xi_begin is the number of pad values added at the beginning of axis i and xi_end, the number of pad values added at the end of axis i.
constant_value (optional, heterogeneous) - T: (Optional) A scalar value to be used if the mode chosen is constant (by default it is 0).
Outputs
output (heterogeneous) - T: Tensor after padding.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output to only numeric types.
OnnxPad_13#
- class mlprodict.npy.xop_auto_import_.OnnxPad_13(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Given a tensor containing the data to be padded (data), a tensor containing the number of start and end pad values for axis (pads), (optionally) a mode, and (optionally) constant_value, a padded tensor (output) is generated.
The three supported modes are (similar to corresponding modes supported by numpy.pad):
constant`(default) - pads with a given constant value as specified by `constant_value (which defaults to 0, empty string, or False)
reflect - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis
edge - pads with the edge values of array
- Example 1 (constant mode):
Insert 0 pads to the beginning of the second dimension.
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘constant’
constant_value = 0.0
output = [
[0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7],
]
- Example 2 (reflect mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘reflect’
output = [
[1.0, 1.2, 1.0, 1.2], [2.3, 3.4, 2.3, 3.4], [4.5, 5.7, 4.5, 5.7],
]
- Example 3 (edge mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘edge’
output = [
[1.0, 1.0, 1.0, 1.2], [2.3, 2.3, 2.3, 3.4], [4.5, 4.5, 4.5, 5.7],
]
Attributes
mode: Supported modes: constant`(default), `reflect, edge Default value is
'constant'.
Inputs
Between 2 and 3 inputs.
data (heterogeneous) - T: Input tensor.
pads (heterogeneous) - tensor(int64): Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. pads should be a 1D tensor of shape [2 * input_rank]. pads format should be: [x1_begin, x2_begin,…,x1_end, x2_end,…], where xi_begin is the number of pad values added at the beginning of axis i and xi_end, the number of pad values added at the end of axis i.
constant_value (optional, heterogeneous) - T: (Optional) A scalar value to be used if the mode chosen is constant (by default it is 0, empty string or False).
Outputs
output (heterogeneous) - T: Tensor after padding.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxPad_18#
- class mlprodict.npy.xop_auto_import_.OnnxPad_18(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Given a tensor containing the data to be padded (data), a tensor containing the number of start and end pad values for axis (pads), (optionally) a mode, and (optionally) constant_value, a padded tensor (output) is generated.
The three supported modes are (similar to corresponding modes supported by numpy.pad):
constant`(default) - pads with a given constant value as specified by `constant_value (which defaults to 0, empty string, or False)
reflect - pads with the reflection of the vector mirrored on the first and last values of the vector along each axis
edge - pads with the edge values of array
- Example 1 (constant mode):
Insert 0 pads to the beginning of the second dimension.
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘constant’
constant_value = 0.0
output = [
[0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7],
]
- Example 2 (reflect mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘reflect’
output = [
[1.0, 1.2, 1.0, 1.2], [2.3, 3.4, 2.3, 3.4], [4.5, 5.7, 4.5, 5.7],
]
- Example 3 (edge mode):
data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
]
pads = [0, 2, 0, 0]
mode = ‘edge’
output = [
[1.0, 1.0, 1.0, 1.2], [2.3, 2.3, 2.3, 3.4], [4.5, 4.5, 4.5, 5.7],
]
Attributes
mode: Supported modes: constant`(default), `reflect, edge Default value is
'constant'.
Inputs
Between 2 and 4 inputs.
data (heterogeneous) - T: Input tensor.
pads (heterogeneous) - tensor(int64): Tensor of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D input tensor, it is the number of pixels. pads should be a 1D tensor of shape [2 * num_axes] where num_axes refers to the number of elements in the axes input or the input rank if axes are not provided explicitly. pads format should be: [x1_begin, x2_begin, …, x1_end, x2_end,…], where xi_begin is the number of pad values added at the beginning of axis axes[i] and xi_end, the number of pad values added at the end of axis axes[i].
constant_value (optional, heterogeneous) - T: (Optional) A scalar value to be used if the mode chosen is constant (by default it is 0, empty string or False).
axes (optional, heterogeneous) - Tind: 1-D tensor of axes that pads apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated. If not provided, all axes are assumed ([0, 1, …, input_rank-1]).
Outputs
output (heterogeneous) - T: Tensor after padding.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxPad_2#
- class mlprodict.npy.xop_auto_import_.OnnxPad_2(*args, **kwargs)#
Version
name: Pad (GitHub)
domain: main
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2.
Summary
Given data tensor, pads, mode, and value. Example:
Insert 0 pads to the beginning of the second dimension. data = [
[1.0, 1.2], [2.3, 3.4], [4.5, 5.7],
] pads = [0, 2, 0, 0] output = [
- [
[0.0, 0.0, 1.0, 1.2], [0.0, 0.0, 2.3, 3.4], [0.0, 0.0, 4.5, 5.7],
],
]
Attributes
mode: Three modes: constant(default), reflect, edge Default value is
'constant'.pads (required): List of integers indicating the number of padding elements to add or remove (if negative) at the beginning and end of each axis. For 2D it is the number of pixels. pads rank should be double of the input’s rank. pads format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number of pixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.
value: One float, indicates the value to be filled. Default value is
0.0.
Inputs
data (heterogeneous) - T: Input tensor.
Outputs
output (heterogeneous) - T: Tensor after padding.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxPow#
- class mlprodict.npy.xop_auto_import_.OnnxPow(*args, **kwargs)#
Version
name: Pow (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Pow takes input data (Tensor<T>) and exponent Tensor, and produces one output data (Tensor<T>) where the function f(x) = x^exponent, is applied to the data tensor elementwise. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: First operand, base of the exponent.
Y (heterogeneous) - T1: Second operand, power of the exponent.
Outputs
Z (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64) ): Constrain input X and output types to float/int tensors.
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input Y types to float/int tensors.
OnnxPow_1#
- class mlprodict.npy.xop_auto_import_.OnnxPow_1(*args, **kwargs)#
Version
name: Pow (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Pow takes input data (Tensor<T>) and exponent Tensor, and produces one output data (Tensor<T>) where the function f(x) = x^exponent, is applied to the data tensor elementwise.
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.
Inputs
X (heterogeneous) - T: Input tensor of any shape, base of the exponent.
Y (heterogeneous) - T: Input tensor of any shape broadcastable to X shape, the exponent component.
Outputs
Z (heterogeneous) - T: Output tensor (same size as X)
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxPow_12#
- class mlprodict.npy.xop_auto_import_.OnnxPow_12(*args, **kwargs)#
Version
name: Pow (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Pow takes input data (Tensor<T>) and exponent Tensor, and produces one output data (Tensor<T>) where the function f(x) = x^exponent, is applied to the data tensor elementwise. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: First operand, base of the exponent.
Y (heterogeneous) - T1: Second operand, power of the exponent.
Outputs
Z (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64) ): Constrain input X and output types to float/int tensors.
T1 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input Y types to float/int tensors.
OnnxPow_13#
- class mlprodict.npy.xop_auto_import_.OnnxPow_13(*args, **kwargs)#
Version
name: Pow (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Pow takes input data (Tensor<T>) and exponent Tensor, and produces one output data (Tensor<T>) where the function f(x) = x^exponent, is applied to the data tensor elementwise. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: First operand, base of the exponent.
Y (heterogeneous) - T1: Second operand, power of the exponent.
Outputs
Z (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64) ): Constrain input X and output types to float/int tensors.
T1 in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input Y types to float/int tensors.
OnnxPow_15#
- class mlprodict.npy.xop_auto_import_.OnnxPow_15(*args, **kwargs)#
Version
name: Pow (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Pow takes input data (Tensor<T>) and exponent Tensor, and produces one output data (Tensor<T>) where the function f(x) = x^exponent, is applied to the data tensor elementwise. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: First operand, base of the exponent.
Y (heterogeneous) - T1: Second operand, power of the exponent.
Outputs
Z (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64) ): Constrain input X and output types to float/int tensors.
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input Y types to float/int tensors.
OnnxPow_7#
- class mlprodict.npy.xop_auto_import_.OnnxPow_7(*args, **kwargs)#
Version
name: Pow (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Pow takes input data (Tensor<T>) and exponent Tensor, and produces one output data (Tensor<T>) where the function f(x) = x^exponent, is applied to the data tensor elementwise. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
X (heterogeneous) - T: First operand, base of the exponent.
Y (heterogeneous) - T: Second operand, power of the exponent.
Outputs
Z (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxQLinearConv#
- class mlprodict.npy.xop_auto_import_.OnnxQLinearConv(*args, **kwargs)#
Version
name: QLinearConv (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
The convolution operator consumes a quantized input tensor, its scale and zero point, a quantized filter, its scale and zero point, and output’s scale and zero point, and computes the quantized output. Each scale and zero-point pair must have same shape. It means they must be either scalars (per tensor) or 1-D tensors (per output channel). Each input or output and its related zero point must have same type. When bias is present it must be quantized using scale = input scale * weight scale and zero point as 0.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
group: number of groups input channels and output channels are divided into. default is 1. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input ‘w’.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number ofpixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
Between 8 and 9 inputs.
x (heterogeneous) - T1: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
x_scale (heterogeneous) - tensor(float): Scale tensor for input ‘x’. It’s a scalar, which means a per- tensor/layer quantization.
x_zero_point (heterogeneous) - T1: Zero point tensor for input ‘x’. It’s a scalar, which means a per- tensor/layer quantization.
w (heterogeneous) - T2: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
w_scale (heterogeneous) - tensor(float): Scale tensor for input ‘w’. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M).
w_zero_point (heterogeneous) - T2: Zero point tensor for input ‘w’. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M).
y_scale (heterogeneous) - tensor(float): Scale tensor for output ‘y’. It’s a scalar, which means a per- tensor/layer quantization.
y_zero_point (heterogeneous) - T3: Zero point tensor for output ‘y’. It’s a scalar, which means a per- tensor/layer quantization.
B (optional, heterogeneous) - T4: Optional 1D bias to be added to the convolution, has size of M. Bias must be quantized using scale = x_scale * w_scale and zero_point = 0
Outputs
y (heterogeneous) - T3: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain filter type to 8-bit integer tensor.
T3 in ( tensor(int8), tensor(uint8) ): Constrain output type to 8-bit integer tensor.
T4 in ( tensor(int32) ): Constrain bias type to 32-bit integer tensor.
OnnxQLinearConv_10#
- class mlprodict.npy.xop_auto_import_.OnnxQLinearConv_10(*args, **kwargs)#
Version
name: QLinearConv (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
The convolution operator consumes a quantized input tensor, its scale and zero point, a quantized filter, its scale and zero point, and output’s scale and zero point, and computes the quantized output. Each scale and zero-point pair must have same shape. It means they must be either scalars (per tensor) or 1-D tensors (per output channel). Each input or output and its related zero point must have same type. When bias is present it must be quantized using scale = input scale * weight scale and zero point as 0.
Attributes
auto_pad: auto_pad must be either NOTSET, SAME_UPPER, SAME_LOWER or VALID. Where default value is NOTSET, which means explicit padding is used. SAME_UPPER or SAME_LOWER mean pad the input so that output_shape[i] = ceil(input_shape[i] / strides[i]) for each axis i. The padding is split between the two sides equally or almost equally (depending on whether it is even or odd). In case the padding is an odd number, the extra padding is added at the end for SAME_UPPER and at the beginning for SAME_LOWER. Default value is
'NOTSET'.dilations: dilation value along each spatial axis of the filter. If not present, the dilation defaults to 1 along each spatial axis.
group: number of groups input channels and output channels are divided into. default is 1. Default value is
1.kernel_shape: The shape of the convolution kernel. If not present, should be inferred from input ‘w’.
pads: Padding for the beginning and ending along each spatial axis, it can take any value greater than or equal to 0.The value represent the number of pixels added to the beginning and end part of the corresponding axis.`pads` format should be as follow [x1_begin, x2_begin…x1_end, x2_end,…], where xi_begin the number ofpixels added at the beginning of axis i and xi_end, the number of pixels added at the end of axis i.This attribute cannot be used simultaneously with auto_pad attribute. If not present, the padding defaultsto 0 along start and end of each spatial axis.
strides: Stride along each spatial axis. If not present, the stride defaults to 1 along each spatial axis.
Inputs
Between 8 and 9 inputs.
x (heterogeneous) - T1: Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 … x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE …].
x_scale (heterogeneous) - tensor(float): Scale tensor for input ‘x’. It’s a scalar, which means a per- tensor/layer quantization.
x_zero_point (heterogeneous) - T1: Zero point tensor for input ‘x’. It’s a scalar, which means a per- tensor/layer quantization.
w (heterogeneous) - T2: The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x … x kn), where (k1 x k2 x … kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL …]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL.
w_scale (heterogeneous) - tensor(float): Scale tensor for input ‘w’. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M).
w_zero_point (heterogeneous) - T2: Zero point tensor for input ‘w’. It could be a scalar or a 1-D tensor, which means a per-tensor/layer or per output channel quantization. If it’s a 1-D tensor, its number of elements should be equal to the number of output channels (M).
y_scale (heterogeneous) - tensor(float): Scale tensor for output ‘y’. It’s a scalar, which means a per- tensor/layer quantization.
y_zero_point (heterogeneous) - T3: Zero point tensor for output ‘y’. It’s a scalar, which means a per- tensor/layer quantization.
B (optional, heterogeneous) - T4: Optional 1D bias to be added to the convolution, has size of M. Bias must be quantized using scale = x_scale * w_scale and zero_point = 0
Outputs
y (heterogeneous) - T3: Output data tensor that contains the result of the convolution. The output dimensions are functions of the kernel size, stride size, and pad lengths.
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain filter type to 8-bit integer tensor.
T3 in ( tensor(int8), tensor(uint8) ): Constrain output type to 8-bit integer tensor.
T4 in ( tensor(int32) ): Constrain bias type to 32-bit integer tensor.
OnnxQLinearMatMul#
- class mlprodict.npy.xop_auto_import_.OnnxQLinearMatMul(*args, **kwargs)#
Version
name: QLinearMatMul (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html. It consumes two quantized input tensors, their scales and zero points, scale and zero point of output, and computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point). For (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. Scale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor (per row for ‘a’ and per column for ‘b’). Scalar refers to per tensor quantization whereas N-D refers to per row or per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be an M element vector [v_1, v_2, …, v_M] for per row quantization and K element vector of shape [v_1, v_2, …, v_K] for per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may have shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization. Production must never overflow, and accumulation may overflow if and only if in 32 bits.
Inputs
a (heterogeneous) - T1: N-dimensional quantized matrix a
a_scale (heterogeneous) - tensor(float): scale of quantized input a
a_zero_point (heterogeneous) - T1: zero point of quantized input a
b (heterogeneous) - T2: N-dimensional quantized matrix b
b_scale (heterogeneous) - tensor(float): scale of quantized input b
b_zero_point (heterogeneous) - T2: zero point of quantized input b
y_scale (heterogeneous) - tensor(float): scale of quantized output y
y_zero_point (heterogeneous) - T3: zero point of quantized output y
Outputs
y (heterogeneous) - T3: Quantized matrix multiply results from a * b
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input a and its zero point data type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain input b and its zero point data type to 8-bit integer tensor.
T3 in ( tensor(int8), tensor(uint8) ): Constrain output y and its zero point data type to 8-bit integer tensor.
OnnxQLinearMatMul_10#
- class mlprodict.npy.xop_auto_import_.OnnxQLinearMatMul_10(*args, **kwargs)#
Version
name: QLinearMatMul (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Matrix product that behaves like numpy.matmul: https://docs.scipy.org/doc/numpy-1.13.0/reference/generated/numpy.matmul.html. It consumes two quantized input tensors, their scales and zero points, scale and zero point of output, and computes the quantized output. The quantization formula is y = saturate((x / y_scale) + y_zero_point). For (x / y_scale), it is rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. Scale and zero point must have same shape. They must be either scalar (per tensor) or N-D tensor (per row for ‘a’ and per column for ‘b’). Scalar refers to per tensor quantization whereas N-D refers to per row or per column quantization. If the input is 2D of shape [M, K] then zero point and scale tensor may be an M element vector [v_1, v_2, …, v_M] for per row quantization and K element vector of shape [v_1, v_2, …, v_K] for per column quantization. If the input is N-D tensor with shape [D1, D2, M, K] then zero point and scale tensor may have shape [D1, D2, M, 1] for per row quantization and shape [D1, D2, 1, K] for per column quantization. Production must never overflow, and accumulation may overflow if and only if in 32 bits.
Inputs
a (heterogeneous) - T1: N-dimensional quantized matrix a
a_scale (heterogeneous) - tensor(float): scale of quantized input a
a_zero_point (heterogeneous) - T1: zero point of quantized input a
b (heterogeneous) - T2: N-dimensional quantized matrix b
b_scale (heterogeneous) - tensor(float): scale of quantized input b
b_zero_point (heterogeneous) - T2: zero point of quantized input b
y_scale (heterogeneous) - tensor(float): scale of quantized output y
y_zero_point (heterogeneous) - T3: zero point of quantized output y
Outputs
y (heterogeneous) - T3: Quantized matrix multiply results from a * b
Type Constraints
T1 in ( tensor(int8), tensor(uint8) ): Constrain input a and its zero point data type to 8-bit integer tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain input b and its zero point data type to 8-bit integer tensor.
T3 in ( tensor(int8), tensor(uint8) ): Constrain output y and its zero point data type to 8-bit integer tensor.
OnnxQuantizeLinear#
- class mlprodict.npy.xop_auto_import_.OnnxQuantizeLinear(*args, **kwargs)#
Version
name: QuantizeLinear (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The linear quantization operator. It consumes a high precision tensor, a scale, and a zero point to compute the low precision / quantized tensor. The scale factor and zero point must have same shape, and can be either a scalar for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. The quantization formula is y = saturate ((x / y_scale) + y_zero_point). For saturation, it saturates to [0, 255] if it’s uint8, or [-128, 127] if it’s int8. For (x / y_scale), it’s rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. ‘y_zero_point’ and ‘y’ must have same type.
Attributes
axis: (Optional) The axis of the quantization dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
Between 2 and 3 inputs.
x (heterogeneous) - T1: N-D full precision Input tensor to be quantized.
y_scale (heterogeneous) - tensor(float): Scale for doing quantization to get ‘y’. It can be a scalar, which means per-tensor/layer quantization, or a 1-D Tensor for per-axis quantization.
y_zero_point (optional, heterogeneous) - T2: Zero point for doing quantization to get ‘y’. Shape must match y_scale. Default is uint8 with zero point of 0 if it’s not specified.
Outputs
y (heterogeneous) - T2: N-D quantized output tensor. It has same shape as input ‘x’.
Type Constraints
T1 in ( tensor(float), tensor(int32) ): Constrain ‘x’ to float or int32 tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain ‘y_zero_point’ and ‘y’ to 8-bit integer tensor.
OnnxQuantizeLinear_10#
- class mlprodict.npy.xop_auto_import_.OnnxQuantizeLinear_10(*args, **kwargs)#
Version
name: QuantizeLinear (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
The linear per-tensor/layer quantization operator. It consumes a high precision tensor, a scale, a zero point to compute the low precision / quantized tensor. The quantization formula is y = saturate ((x / y_scale) + y_zero_point). For saturation, it saturates to [0, 255] if it’s uint8, or [-128, 127] if it’s int8. For (x / y_scale), it’s rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. ‘y_zero_point’ and ‘y’ must have same type.
Inputs
Between 2 and 3 inputs.
x (heterogeneous) - T1: N-D full precision Input tensor to be quantized.
y_scale (heterogeneous) - tensor(float): Scale for doing quantization to get ‘y’. It’s a scalar, which means a per-tensor/layer quantization.
y_zero_point (optional, heterogeneous) - T2: Zero point for doing quantization to get ‘y’. It’s a scalar, which means a per-tensor/layer quantization. Default value is uint8 typed 0 if it’s not specified.
Outputs
y (heterogeneous) - T2: N-D quantized output tensor. It has same shape as input ‘x’.
Type Constraints
T1 in ( tensor(float), tensor(int32) ): Constrain ‘x’ to float or int32 tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain ‘y_zero_point’ and ‘y’ to 8-bit integer tensor.
OnnxQuantizeLinear_13#
- class mlprodict.npy.xop_auto_import_.OnnxQuantizeLinear_13(*args, **kwargs)#
Version
name: QuantizeLinear (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The linear quantization operator. It consumes a high precision tensor, a scale, and a zero point to compute the low precision / quantized tensor. The scale factor and zero point must have same shape, and can be either a scalar for per-tensor / per layer quantization, or a 1-D tensor for per-axis quantization. The quantization formula is y = saturate ((x / y_scale) + y_zero_point). For saturation, it saturates to [0, 255] if it’s uint8, or [-128, 127] if it’s int8. For (x / y_scale), it’s rounding to nearest ties to even. Refer to https://en.wikipedia.org/wiki/Rounding for details. ‘y_zero_point’ and ‘y’ must have same type.
Attributes
axis: (Optional) The axis of the quantization dimension of the input tensor. Ignored for per-tensor quantization. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
Between 2 and 3 inputs.
x (heterogeneous) - T1: N-D full precision Input tensor to be quantized.
y_scale (heterogeneous) - tensor(float): Scale for doing quantization to get ‘y’. It can be a scalar, which means per-tensor/layer quantization, or a 1-D Tensor for per-axis quantization.
y_zero_point (optional, heterogeneous) - T2: Zero point for doing quantization to get ‘y’. Shape must match y_scale. Default is uint8 with zero point of 0 if it’s not specified.
Outputs
y (heterogeneous) - T2: N-D quantized output tensor. It has same shape as input ‘x’.
Type Constraints
T1 in ( tensor(float), tensor(int32) ): Constrain ‘x’ to float or int32 tensor.
T2 in ( tensor(int8), tensor(uint8) ): Constrain ‘y_zero_point’ and ‘y’ to 8-bit integer tensor.
OnnxRNN#
- class mlprodict.npy.xop_auto_import_.OnnxRNN(*args, **kwargs)#
Version
name: RNN (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Computes an one-layer simple RNN. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
t - time step (t-1 means previous time step)
Wi - W parameter weight matrix for input gate
Ri - R recurrence weight matrix for input gate
Wbi - W parameter bias vector for input gate
Rbi - R parameter bias vector for input gate
WBi - W parameter weight matrix for backward input gate
RBi - R recurrence weight matrix for backward input gate
WBbi - WR bias vectors for backward input gate
RBbi - RR bias vectors for backward input gate
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Tanh):
Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default Tanh if not specified. Default value is
[b'Tanh' b'Tanh'].clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
layout: The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size]. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for input gate. Concatenation of Wi and WBi (if bidirectional). The tensor has shape [num_directions, hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of Ri and RBi (if bidirectional). The tensor has shape [num_directions, hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wbi, Rbi] and [WBbi, RBbi] (if bidirectional). The tensor has shape [num_directions, 2*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxRNN_1#
- class mlprodict.npy.xop_auto_import_.OnnxRNN_1(*args, **kwargs)#
Version
name: RNN (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes an one-layer simple RNN. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
t - time step (t-1 means previous time step)
Wi - W parameter weight matrix for input gate
Ri - R recurrence weight matrix for input gate
Wbi - W parameter bias vector for input gate
Rbi - R parameter bias vector for input gate
WBi - W parameter weight matrix for backward input gate
RBi - R recurrence weight matrix for backward input gate
WBbi - WR bias vectors for backward input gate
RBbi - RR bias vectors for backward input gate
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Tanh):
Ht = f(Xt*(Wi^T) + Ht-1*Ri + Wbi + Rbi)
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default Tanh if not specified. Default value is
[b'Tanh' b'Tanh'].clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
output_sequence: The sequence output for the hidden is optional if 0. Default 0. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for input gate. Concatenation of Wi and WBi (if bidirectional). The tensor has shape [num_directions, hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of Ri and RBi (if bidirectional). The tensor has shape [num_directions, hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wbi, Rbi] and [WBbi, RBbi] (if bidirectional). The tensor has shape [num_directions, 2*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size]. It is optional if output_sequence is 0.
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxRNN_14#
- class mlprodict.npy.xop_auto_import_.OnnxRNN_14(*args, **kwargs)#
Version
name: RNN (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Computes an one-layer simple RNN. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
t - time step (t-1 means previous time step)
Wi - W parameter weight matrix for input gate
Ri - R recurrence weight matrix for input gate
Wbi - W parameter bias vector for input gate
Rbi - R parameter bias vector for input gate
WBi - W parameter weight matrix for backward input gate
RBi - R recurrence weight matrix for backward input gate
WBbi - WR bias vectors for backward input gate
RBbi - RR bias vectors for backward input gate
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Tanh):
Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default Tanh if not specified. Default value is
[b'Tanh' b'Tanh'].clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
layout: The shape format of inputs X, initial_h and outputs Y, Y_h. If 0, the following shapes are expected: X.shape = [seq_length, batch_size, input_size], Y.shape = [seq_length, num_directions, batch_size, hidden_size], initial_h.shape = Y_h.shape = [num_directions, batch_size, hidden_size]. If 1, the following shapes are expected: X.shape = [batch_size, seq_length, input_size], Y.shape = [batch_size, seq_length, num_directions, hidden_size], initial_h.shape = Y_h.shape = [batch_size, num_directions, hidden_size]. Default value is
0.
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for input gate. Concatenation of Wi and WBi (if bidirectional). The tensor has shape [num_directions, hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of Ri and RBi (if bidirectional). The tensor has shape [num_directions, hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wbi, Rbi] and [WBbi, RBbi] (if bidirectional). The tensor has shape [num_directions, 2*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxRNN_7#
- class mlprodict.npy.xop_auto_import_.OnnxRNN_7(*args, **kwargs)#
Version
name: RNN (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Computes an one-layer simple RNN. This operator is usually supported via some custom implementation such as CuDNN.
Notations:
X - input tensor
i - input gate
t - time step (t-1 means previous time step)
Wi - W parameter weight matrix for input gate
Ri - R recurrence weight matrix for input gate
Wbi - W parameter bias vector for input gate
Rbi - R parameter bias vector for input gate
WBi - W parameter weight matrix for backward input gate
RBi - R recurrence weight matrix for backward input gate
WBbi - WR bias vectors for backward input gate
RBbi - RR bias vectors for backward input gate
H - Hidden state
num_directions - 2 if direction == bidirectional else 1
Activation functions:
Relu(x) - max(0, x)
Tanh(x) - (1 - e^{-2x})/(1 + e^{-2x})
Sigmoid(x) - 1/(1 + e^{-x})
(NOTE: Below are optional)
Affine(x) - alpha*x + beta
LeakyRelu(x) - x if x >= 0 else alpha * x
ThresholdedRelu(x) - x if x >= alpha else 0
ScaledTanh(x) - alpha*Tanh(beta*x)
HardSigmoid(x) - min(max(alpha*x + beta, 0), 1)
Elu(x) - x if x >= 0 else alpha*(e^x - 1)
Softsign(x) - x/(1 + |x|)
Softplus(x) - log(1 + e^x)
Equations (Default: f=Tanh):
Ht = f(Xt*(Wi^T) + Ht-1*(Ri^T) + Wbi + Rbi)
This operator has optional inputs/outputs. See ONNX for more details about the representation of optional arguments. An empty string may be used in the place of an actual argument’s name to indicate a missing argument. Trailing optional arguments (those not followed by an argument that is present) may also be simply omitted.
Attributes
activation_alpha: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.For example with LeakyRelu, the default alpha is 0.01.
activation_beta: Optional scaling values used by some activation functions. The values are consumed in the order of activation functions, for example (f, g, h) in LSTM. Default values are the same as of corresponding ONNX operators.
activations: One (or two if bidirectional) activation function for input gate. The activation function must be one of the activation functions specified above. Optional: Default Tanh if not specified. Default value is
[b'Tanh' b'Tanh'].clip: Cell clip threshold. Clipping bounds the elements of a tensor in the range of [-threshold, +threshold] and is applied to the input of activations. No clip if not specified.
direction: Specify if the RNN is forward, reverse, or bidirectional. Must be one of forward (default), reverse, or bidirectional. Default value is
'forward'.hidden_size: Number of neurons in the hidden layer
Inputs
Between 3 and 6 inputs.
X (heterogeneous) - T: The input sequences packed (and potentially padded) into one 3-D tensor with the shape of [seq_length, batch_size, input_size].
W (heterogeneous) - T: The weight tensor for input gate. Concatenation of Wi and WBi (if bidirectional). The tensor has shape [num_directions, hidden_size, input_size].
R (heterogeneous) - T: The recurrence weight tensor. Concatenation of Ri and RBi (if bidirectional). The tensor has shape [num_directions, hidden_size, hidden_size].
B (optional, heterogeneous) - T: The bias tensor for input gate. Concatenation of [Wbi, Rbi] and [WBbi, RBbi] (if bidirectional). The tensor has shape [num_directions, 2*hidden_size]. Optional: If not specified - assumed to be 0.
sequence_lens (optional, heterogeneous) - T1: Optional tensor specifying lengths of the sequences in a batch. If not specified - assumed all sequences in the batch to have length seq_length. It has shape [batch_size].
initial_h (optional, heterogeneous) - T: Optional initial value of the hidden. If not specified - assumed to be 0. It has shape [num_directions, batch_size, hidden_size].
Outputs
Between 0 and 2 outputs.
Y (optional, heterogeneous) - T: A tensor that concats all the intermediate output values of the hidden. It has shape [seq_length, num_directions, batch_size, hidden_size].
Y_h (optional, heterogeneous) - T: The last output value of the hidden. It has shape [num_directions, batch_size, hidden_size].
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
T1 in ( tensor(int32) ): Constrain seq_lens to integer tensor.
OnnxRandomNormal#
- class mlprodict.npy.xop_auto_import_.OnnxRandomNormal(*args, **kwargs)#
Version
name: RandomNormal (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a normal distribution. The shape of the tensor is specified by the shape argument and the parameter of the normal distribution specified by mean and scale.
The data type is specified by the ‘dtype’ argument. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Attributes
dtype: The data type for the elements of the output tensor. Default is TensorProto::FLOAT. Default value is
1.mean: The mean of the normal distribution. Default value is
0.0.scale: The standard deviation of the normal distribution. Default value is
1.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
shape (required): The shape of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor of random values drawn from normal distribution
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomNormalLike#
- class mlprodict.npy.xop_auto_import_.OnnxRandomNormalLike(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a normal distribution. The shape of the output tensor is copied from the shape of the input tensor, and the parameters of the normal distribution are specified by mean and scale.
The data type is specified by the ‘dtype’ argument, or copied from the input tensor if not provided. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message, and be valid as an output type.
Attributes
dtype: (Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
mean: The mean of the normal distribution. Default value is
0.0.scale: The standard deviation of the normal distribution. Default value is
1.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: Input tensor to copy shape and optionally type information from.
Outputs
output (heterogeneous) - T2: Output tensor of random values drawn from normal distribution
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomNormalLike_1#
- class mlprodict.npy.xop_auto_import_.OnnxRandomNormalLike_1(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a normal distribution. The shape of the output tensor is copied from the shape of the input tensor, and the parameters of the normal distribution are specified by mean and scale.
The data type is specified by the ‘dtype’ argument, or copied from the input tensor if not provided. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message, and be valid as an output type.
Attributes
dtype: (Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
mean: The mean of the normal distribution. Default value is
0.0.scale: The standard deviation of the normal distribution. Default value is
1.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: Input tensor to copy shape and optionally type information from.
Outputs
output (heterogeneous) - T2: Output tensor of random values drawn from normal distribution
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomNormal_1#
- class mlprodict.npy.xop_auto_import_.OnnxRandomNormal_1(*args, **kwargs)#
Version
name: RandomNormal (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a normal distribution. The shape of the tensor is specified by the shape argument and the parameter of the normal distribution specified by mean and scale.
The data type is specified by the ‘dtype’ argument. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Attributes
dtype: The data type for the elements of the output tensor. Default is TensorProto::FLOAT. Default value is
1.mean: The mean of the normal distribution. Default value is
0.0.scale: The standard deviation of the normal distribution. Default value is
1.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
shape (required): The shape of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor of random values drawn from normal distribution
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomUniform#
- class mlprodict.npy.xop_auto_import_.OnnxRandomUniform(*args, **kwargs)#
Version
name: RandomUniform (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a uniform distribution. The shape of the tensor is specified by the shape argument and the range by low and high.
The data type is specified by the ‘dtype’ argument. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Attributes
dtype: The data type for the elements of the output tensor. If not specified, default is TensorProto::FLOAT. Default value is
1.high: Upper boundary of the output values. Default value is
1.0.low: Lower boundary of the output values. Default value is
0.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
shape (required): The shape of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor of random values drawn from uniform distribution
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomUniformLike#
- class mlprodict.npy.xop_auto_import_.OnnxRandomUniformLike(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a uniform distribution. The shape of the output tensor is copied from the shape of the input tensor, and the parameters of the uniform distribution are specified by low and high.
The data type is specified by the ‘dtype’ argument, or copied from the input tensor if not provided. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message and be valid as an output type.
Attributes
dtype: (Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
high: Upper boundary of the output values. Default value is
1.0.low: Lower boundary of the output values. Default value is
0.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: Input tensor to copy shape and optionally type information from.
Outputs
output (heterogeneous) - T2: Output tensor of random values drawn from uniform distribution
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomUniformLike_1#
- class mlprodict.npy.xop_auto_import_.OnnxRandomUniformLike_1(*args, **kwargs)#
Version
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a uniform distribution. The shape of the output tensor is copied from the shape of the input tensor, and the parameters of the uniform distribution are specified by low and high.
The data type is specified by the ‘dtype’ argument, or copied from the input tensor if not provided. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message and be valid as an output type.
Attributes
dtype: (Optional) The data type for the elements of the output tensor, if not specified, we will use the data type of the input tensor.
high: Upper boundary of the output values. Default value is
1.0.low: Lower boundary of the output values. Default value is
0.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
Inputs
input (heterogeneous) - T1: Input tensor to copy shape and optionally type information from.
Outputs
output (heterogeneous) - T2: Output tensor of random values drawn from uniform distribution
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type. If the dtype attribute is not provided this must be a valid output type.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRandomUniform_1#
- class mlprodict.npy.xop_auto_import_.OnnxRandomUniform_1(*args, **kwargs)#
Version
name: RandomUniform (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Generate a tensor with random values drawn from a uniform distribution. The shape of the tensor is specified by the shape argument and the range by low and high.
The data type is specified by the ‘dtype’ argument. The ‘dtype’ argument must be one of the data types specified in the ‘DataType’ enum field in the TensorProto message.
Attributes
dtype: The data type for the elements of the output tensor. If not specified, default is TensorProto::FLOAT. Default value is
1.high: Upper boundary of the output values. Default value is
1.0.low: Lower boundary of the output values. Default value is
0.0.seed: (Optional) Seed to the random generator, if not specified we will auto generate one.
shape (required): The shape of the output tensor.
Outputs
output (heterogeneous) - T: Output tensor of random values drawn from uniform distribution
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain output types to float tensors.
OnnxRange#
- class mlprodict.npy.xop_auto_import_.OnnxRange(*args, **kwargs)#
Version
name: Range (GitHub)
domain: main
since_version: 11
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Generate a tensor containing a sequence of numbers that begin at start and extends by increments of delta up to limit (exclusive).
The number of elements in the output of range is computed as below-
number_of_elements = max( ceil( (limit - start) / delta ) , 0 )
The pseudocode determining the contents of the output is shown below-
for(int i=0; i<number_of_elements; ++i)
{
` output[i] = start + (i * delta); `
}
Example 1 Inputs: start = 3, limit = 9, delta = 3 Output: [3, 6]
Example 2 Inputs: start = 10, limit = 4, delta = -2 Output: [10, 8, 6]
Inputs
start (heterogeneous) - T: Scalar. First entry for the range of output values.
limit (heterogeneous) - T: Scalar. Exclusive upper limit for the range of output values.
delta (heterogeneous) - T: Scalar. Value to step by.
Outputs
output (heterogeneous) - T: A 1-D tensor with same type as the inputs containing generated range of values.
Type Constraints
T in ( tensor(double), tensor(float), tensor(int16), tensor(int32), tensor(int64) ): Constrain input types to common numeric type tensors.
OnnxRange_11#
- class mlprodict.npy.xop_auto_import_.OnnxRange_11(*args, **kwargs)#
Version
name: Range (GitHub)
domain: main
since_version: 11
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Generate a tensor containing a sequence of numbers that begin at start and extends by increments of delta up to limit (exclusive).
The number of elements in the output of range is computed as below-
number_of_elements = max( ceil( (limit - start) / delta ) , 0 )
The pseudocode determining the contents of the output is shown below-
for(int i=0; i<number_of_elements; ++i)
{
` output[i] = start + (i * delta); `
}
Example 1 Inputs: start = 3, limit = 9, delta = 3 Output: [3, 6]
Example 2 Inputs: start = 10, limit = 4, delta = -2 Output: [10, 8, 6]
Inputs
start (heterogeneous) - T: Scalar. First entry for the range of output values.
limit (heterogeneous) - T: Scalar. Exclusive upper limit for the range of output values.
delta (heterogeneous) - T: Scalar. Value to step by.
Outputs
output (heterogeneous) - T: A 1-D tensor with same type as the inputs containing generated range of values.
Type Constraints
T in ( tensor(double), tensor(float), tensor(int16), tensor(int32), tensor(int64) ): Constrain input types to common numeric type tensors.
OnnxReciprocal#
- class mlprodict.npy.xop_auto_import_.OnnxReciprocal(*args, **kwargs)#
Version
name: Reciprocal (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Reciprocal takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the reciprocal is, y = 1/x, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxReciprocal_1#
- class mlprodict.npy.xop_auto_import_.OnnxReciprocal_1(*args, **kwargs)#
Version
name: Reciprocal (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Reciprocal takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the reciprocal is, y = 1/x, is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxReciprocal_13#
- class mlprodict.npy.xop_auto_import_.OnnxReciprocal_13(*args, **kwargs)#
Version
name: Reciprocal (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Reciprocal takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the reciprocal is, y = 1/x, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxReciprocal_6#
- class mlprodict.npy.xop_auto_import_.OnnxReciprocal_6(*args, **kwargs)#
Version
name: Reciprocal (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Reciprocal takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the reciprocal is, y = 1/x, is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxReduceL1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL1(*args, **kwargs)#
Version
name: ReduceL1 (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the L1 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL1_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL1_1(*args, **kwargs)#
Version
name: ReduceL1 (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the L1 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL1_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL1_11(*args, **kwargs)#
Version
name: ReduceL1 (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the L1 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL1_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL1_13(*args, **kwargs)#
Version
name: ReduceL1 (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the L1 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL1_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL1_18(*args, **kwargs)#
Version
name: ReduceL1 (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the L1 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL2#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL2(*args, **kwargs)#
Version
name: ReduceL2 (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the L2 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL2_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL2_1(*args, **kwargs)#
Version
name: ReduceL2 (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the L2 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL2_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL2_11(*args, **kwargs)#
Version
name: ReduceL2 (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the L2 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL2_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL2_13(*args, **kwargs)#
Version
name: ReduceL2 (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the L2 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceL2_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceL2_18(*args, **kwargs)#
Version
name: ReduceL2 (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the L2 norm of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSum#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSum(*args, **kwargs)#
Version
name: ReduceLogSum (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the log sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSumExp#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSumExp(*args, **kwargs)#
Version
name: ReduceLogSumExp (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the log sum exponent of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSumExp_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSumExp_1(*args, **kwargs)#
Version
name: ReduceLogSumExp (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the log sum exponent of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSumExp_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSumExp_11(*args, **kwargs)#
Version
name: ReduceLogSumExp (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the log sum exponent of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSumExp_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSumExp_13(*args, **kwargs)#
Version
name: ReduceLogSumExp (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the log sum exponent of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSumExp_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSumExp_18(*args, **kwargs)#
Version
name: ReduceLogSumExp (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the log sum exponent of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSum_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSum_1(*args, **kwargs)#
Version
name: ReduceLogSum (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the log sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSum_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSum_11(*args, **kwargs)#
Version
name: ReduceLogSum (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the log sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSum_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSum_13(*args, **kwargs)#
Version
name: ReduceLogSum (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the log sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceLogSum_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceLogSum_18(*args, **kwargs)#
Version
name: ReduceLogSum (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the log sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMax#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMax(*args, **kwargs)#
Version
name: ReduceMax (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the max of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMax_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMax_1(*args, **kwargs)#
Version
name: ReduceMax (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the max of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMax_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMax_11(*args, **kwargs)#
Version
name: ReduceMax (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the max of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMax_12#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMax_12(*args, **kwargs)#
Version
name: ReduceMax (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Computes the max of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMax_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMax_13(*args, **kwargs)#
Version
name: ReduceMax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the max of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMax_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMax_18(*args, **kwargs)#
Version
name: ReduceMax (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the max of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMean#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMean(*args, **kwargs)#
Version
name: ReduceMean (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the mean of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMean_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMean_1(*args, **kwargs)#
Version
name: ReduceMean (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the mean of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMean_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMean_11(*args, **kwargs)#
Version
name: ReduceMean (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the mean of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMean_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMean_13(*args, **kwargs)#
Version
name: ReduceMean (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the mean of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMean_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMean_18(*args, **kwargs)#
Version
name: ReduceMean (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the mean of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMin#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMin(*args, **kwargs)#
Version
name: ReduceMin (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the min of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMin_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMin_1(*args, **kwargs)#
Version
name: ReduceMin (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the min of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMin_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMin_11(*args, **kwargs)#
Version
name: ReduceMin (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the min of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceMin_12#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMin_12(*args, **kwargs)#
Version
name: ReduceMin (GitHub)
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Computes the min of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMin_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMin_13(*args, **kwargs)#
Version
name: ReduceMin (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the min of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceMin_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceMin_18(*args, **kwargs)#
Version
name: ReduceMin (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the min of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(int8), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to high-precision and 8 bit numeric tensors.
OnnxReduceProd#
- class mlprodict.npy.xop_auto_import_.OnnxReduceProd(*args, **kwargs)#
Version
name: ReduceProd (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the product of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceProd_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceProd_1(*args, **kwargs)#
Version
name: ReduceProd (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the product of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceProd_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceProd_11(*args, **kwargs)#
Version
name: ReduceProd (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the product of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceProd_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceProd_13(*args, **kwargs)#
Version
name: ReduceProd (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the product of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceProd_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceProd_18(*args, **kwargs)#
Version
name: ReduceProd (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the product of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSum#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSum(*args, **kwargs)#
Version
name: ReduceSum (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSumSquare#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSumSquare(*args, **kwargs)#
Version
name: ReduceSumSquare (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the sum square of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSumSquare_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSumSquare_1(*args, **kwargs)#
Version
name: ReduceSumSquare (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the sum square of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSumSquare_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSumSquare_11(*args, **kwargs)#
Version
name: ReduceSumSquare (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the sum square of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSumSquare_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSumSquare_13(*args, **kwargs)#
Version
name: ReduceSumSquare (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the sum square of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSumSquare_18#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSumSquare_18(*args, **kwargs)#
Version
name: ReduceSumSquare (GitHub)
domain: main
since_version: 18
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Computes the sum square of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSum_1#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSum_1(*args, **kwargs)#
Version
name: ReduceSum (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Computes the sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor.
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSum_11#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSum_11(*args, **kwargs)#
Version
name: ReduceSum (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Computes the sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equal 0, then the resulted tensor have the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
axes: A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxReduceSum_13#
- class mlprodict.npy.xop_auto_import_.OnnxReduceSum_13(*args, **kwargs)#
Version
name: ReduceSum (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Computes the sum of the input tensor’s element along the provided axes. The resulting tensor has the same rank as the input if keepdims equals 1. If keepdims equals 0, then the resulting tensor has the reduced dimension pruned.
The above behavior is similar to numpy, with the exception that numpy defaults keepdims to False instead of True.
Attributes
keepdims: Keep the reduced dimension or not, default 1 means keep reduced dimension. Default value is
1.noop_with_empty_axes: Defines behavior if ‘axes’ is empty. Default behavior with ‘false’ is to reduce all axes. When axes is empty and this attribute is set to true, input tensor will not be reduced,and the output tensor would be equivalent to input tensor. Default value is
0.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: An input tensor.
axes (optional, heterogeneous) - tensor(int64): Optional input list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor if ‘noop_with_empty_axes’ is false, else act as an Identity op when ‘noop_with_empty_axes’ is true. Accepted range is [-r, r-1] where r = rank(data).
Outputs
reduced (heterogeneous) - T: Reduced output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxRelu#
- class mlprodict.npy.xop_auto_import_.OnnxRelu(*args, **kwargs)#
Version
name: Relu (GitHub)
domain: main
since_version: 14
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Relu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = max(0, x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8) ): Constrain input and output types to signed numeric tensors.
OnnxRelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxRelu_1(*args, **kwargs)#
Version
name: Relu (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Relu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = max(0, x), is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxRelu_13#
- class mlprodict.npy.xop_auto_import_.OnnxRelu_13(*args, **kwargs)#
Version
name: Relu (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Relu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = max(0, x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxRelu_14#
- class mlprodict.npy.xop_auto_import_.OnnxRelu_14(*args, **kwargs)#
Version
name: Relu (GitHub)
domain: main
since_version: 14
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Relu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = max(0, x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8) ): Constrain input and output types to signed numeric tensors.
OnnxRelu_6#
- class mlprodict.npy.xop_auto_import_.OnnxRelu_6(*args, **kwargs)#
Version
name: Relu (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Relu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = max(0, x), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxReshape#
- class mlprodict.npy.xop_auto_import_.OnnxReshape(*args, **kwargs)#
Version
name: Reshape (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Reshape the input tensor similar to numpy.reshape. First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. At most one dimension of the new shape can be -1. In this case, the value is inferred from the size of the tensor and the remaining dimensions. A dimension could also be 0, in which case the actual dimension value is unchanged (i.e. taken from the input tensor). If ‘allowzero’ is set, and the new shape includes 0, the dimension will be set explicitly to zero (i.e. not taken from input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. The input tensor’s shape and the output tensor’s shape are required to have the same number of elements.
If the attribute ‘allowzero’ is set, it is invalid for the specified shape to contain both a zero value and -1, as the value of the dimension corresponding to -1 cannot be determined uniquely.
Attributes
allowzero: (Optional) By default, when any value in the ‘shape’ input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the ‘shape’ input is set to zero, the zero value is honored, similar to NumPy. Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
shape (heterogeneous) - tensor(int64): Specified shape for output.
Outputs
reshaped (heterogeneous) - T: Reshaped data.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxReshape_1#
- class mlprodict.npy.xop_auto_import_.OnnxReshape_1(*args, **kwargs)#
Version
name: Reshape (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Reshape the input tensor similar to numpy.reshape. It takes a tensor as input and an argument shape. It outputs the reshaped tensor. At most one dimension of the new shape can be -1. In this case, the value is inferred from the size of the tensor and the remaining dimensions. A dimension could also be 0, in which case the actual dimension value is unchanged (i.e. taken from the input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. The input tensor’s shape and the output tensor’s shape are required to have the same number of elements.
Attributes
consumed_inputs: legacy optimization attribute.
shape: New shape
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
reshaped (heterogeneous) - T: Reshaped data.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxReshape_13#
- class mlprodict.npy.xop_auto_import_.OnnxReshape_13(*args, **kwargs)#
Version
name: Reshape (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Reshape the input tensor similar to numpy.reshape. First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. At most one dimension of the new shape can be -1. In this case, the value is inferred from the size of the tensor and the remaining dimensions. A dimension could also be 0, in which case the actual dimension value is unchanged (i.e. taken from the input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. The input tensor’s shape and the output tensor’s shape are required to have the same number of elements.
Inputs
data (heterogeneous) - T: An input tensor.
shape (heterogeneous) - tensor(int64): Specified shape for output.
Outputs
reshaped (heterogeneous) - T: Reshaped data.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxReshape_14#
- class mlprodict.npy.xop_auto_import_.OnnxReshape_14(*args, **kwargs)#
Version
name: Reshape (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Reshape the input tensor similar to numpy.reshape. First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. At most one dimension of the new shape can be -1. In this case, the value is inferred from the size of the tensor and the remaining dimensions. A dimension could also be 0, in which case the actual dimension value is unchanged (i.e. taken from the input tensor). If ‘allowzero’ is set, and the new shape includes 0, the dimension will be set explicitly to zero (i.e. not taken from input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. The input tensor’s shape and the output tensor’s shape are required to have the same number of elements.
If the attribute ‘allowzero’ is set, it is invalid for the specified shape to contain both a zero value and -1, as the value of the dimension corresponding to -1 cannot be determined uniquely.
Attributes
allowzero: (Optional) By default, when any value in the ‘shape’ input is equal to zero the corresponding dimension value is copied from the input tensor dynamically. allowzero=1 indicates that if any value in the ‘shape’ input is set to zero, the zero value is honored, similar to NumPy. Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
shape (heterogeneous) - tensor(int64): Specified shape for output.
Outputs
reshaped (heterogeneous) - T: Reshaped data.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxReshape_5#
- class mlprodict.npy.xop_auto_import_.OnnxReshape_5(*args, **kwargs)#
Version
name: Reshape (GitHub)
domain: main
since_version: 5
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 5.
Summary
Reshape the input tensor similar to numpy.reshape. First input is the data tensor, second input is a shape tensor which specifies the output shape. It outputs the reshaped tensor. At most one dimension of the new shape can be -1. In this case, the value is inferred from the size of the tensor and the remaining dimensions. A dimension could also be 0, in which case the actual dimension value is unchanged (i.e. taken from the input tensor). Shape (second input) could be an empty shape, which means converting to a scalar. The input tensor’s shape and the output tensor’s shape are required to have the same number of elements.
Inputs
data (heterogeneous) - T: An input tensor.
shape (heterogeneous) - tensor(int64): Specified shape for output.
Outputs
reshaped (heterogeneous) - T: Reshaped data.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxResize#
- class mlprodict.npy.xop_auto_import_.OnnxResize(*args, **kwargs)#
Version
name: Resize (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. Each dimension value of the output tensor is: <br/>
output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) <br/>
if input "sizes" is not specified.
Attributes
antialias: If set to 1, “linear” and “cubic” interpolation modes will use an antialiasing filter when downscaling. Antialiasing is achieved by stretching the resampling filter by a factor max(1, 1 / scale), which means that when downsampling, more input pixels contribute to an output pixel. Default value is
0.axes: If provided, it specifies a subset of axes that ‘roi’, ‘scales’ and ‘sizes’ refer to. If not provided, all axes are assumed [0, 1, …, r-1], where r = rank(data). Non-specified dimensions are interpreted as non-resizable. Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
- coordinate_transformation_mode:
This attribute describes how to transform the coordinate in the
resized tensor to the coordinate in the original tensor. <br/> The coordinate of each dimension is transformed individually. Let’s describe a case using axis x as an example. Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, length_original as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input “roi”, scale = length_resized / length_original, <br/> if coordinate_transformation_mode is “half_pixel”, <br/> x_original = (x_resized + 0.5) / scale - 0.5 <br/> if coordinate_transformation_mode is “pytorch_half_pixel”, <br/> x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0 <br/> if coordinate_transformation_mode is “align_corners”, <br/> x_original = x_resized * (length_original - 1) / (length_resized - 1) <br/> if coordinate_transformation_mode is “asymmetric”, <br/> x_original = x_resized / scale <br/> if coordinate_transformation_mode is “tf_crop_and_resize”, <br/> x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1) . Default value is
'half_pixel'.cubic_coeff_a: The coefficient ‘a’ used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if mode is “cubic”. Default value is
-0.75.exclude_outside: If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0. Default value is
0.extrapolation_value: When coordinate_transformation_mode is “tf_crop_and_resize” and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f. Default value is
0.0.- keep_aspect_ratio_policy:
This attribute describes how to interpret the sizes input with
regard to keeping the original aspect ratio of the input, and it is not applicable when the scales input is used. <br/> Given a set of sizes, associated with a subset of axes (explicitly provided or default), and assuming d = axes[i], with i being the index of the provided sizes. <br/> If keep_aspect_ratio_policy is “stretch”, the original aspect ratio is disregarded, and the input is resized to the specified size: <br/> out_size[d] = sizes[i] <br/> If keep_aspect_ratio_policy is “not_larger”, the sizes are adjusted so that no extent of the output is larger than the specified size, while keeping the original aspect ratio: <br/> scale = Min(sizes[i] / in_size[d]) <br/> out_size[d] = round_int(scale * in_size[i]) <br/> If keep_aspect_ratio_policy is “not_smaller”, the sizes are adjusted so that no extent of the output is smaller than the specified size, while keeping the original aspect ratio: <br/> scale = Max(sizes[i] / in_size[d]) <br/> out_size[d] = round_int(scale * in_size[i]) <br/> For non- resizable axes (those not specified in axes), the output size will be equal to the input size. Note: round_int stands for computing the nearest integer value, rounding halfway cases up. Default value is
'stretch'.mode: Three interpolation modes: “nearest” (default), “linear” and “cubic”. The “linear” mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The “cubic” mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor). Default value is
'nearest'.nearest_mode: Four modes: “round_prefer_floor” (default, as known as round half down), “round_prefer_ceil” (as known as round half up), “floor”, “ceil”. Only used by nearest interpolation. It indicates how to get “nearest” pixel in input tensor from x_original, so this attribute is valid only if “mode” is “nearest”. Default value is
'round_prefer_floor'.
Inputs
Between 1 and 4 inputs.
X (heterogeneous) - T1: N-D tensor
roi (optional, heterogeneous) - T2: 1-D tensor given as [start1, …, startN, end1, …, endN], where N is the rank of X or the length of axes, if provided. The RoIs’ coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is “tf_crop_and_resize”
scales (optional, heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than 0. If it’s less than 1, it’s sampling down, otherwise, it’s upsampling. The number of elements of ‘scales’ should be the same as the rank of input ‘X’ or the length of ‘axes’, if provided. One of ‘scales’ and ‘sizes’ MUST be specified and it is an error if both are specified. If ‘sizes’ is needed, the user can use an empty string as the name of ‘scales’ in this operator’s input list.
sizes (optional, heterogeneous) - tensor(int64): Target size of the output tensor. Its interpretation depends on the ‘keep_aspect_ratio_policy’ value.The number of elements of ‘sizes’ should be the same as the rank of input ‘X’, or the length of ‘axes’, if provided. Only one of ‘scales’ and ‘sizes’ can be specified.
Outputs
Y (heterogeneous) - T1: N-D tensor after resizing
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain roi type to float or double.
OnnxResize_10#
- class mlprodict.npy.xop_auto_import_.OnnxResize_10(*args, **kwargs)#
Version
name: Resize (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Resize the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * scale).
Attributes
mode: Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc) Default value is
'nearest'.
Inputs
X (heterogeneous) - T: N-D tensor
scales (heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than 0. If it’s less than 1, it’s sampling down, otherwise, it’s upsampling. The number of elements of ‘scales’ should be the same as the rank of input ‘X’.
Outputs
Y (heterogeneous) - T: N-D tensor after resizing
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
OnnxResize_11#
- class mlprodict.npy.xop_auto_import_.OnnxResize_11(*args, **kwargs)#
Version
name: Resize (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) if input "sizes" is not specified.
Attributes
- coordinate_transformation_mode:
This attribute describes how to transform the coordinate in the
resized tensor to the coordinate in the original tensor. <br/> The coordinate of each dimension is transformed individually. Let’s describe a case using axis x as an example. Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, length_original as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input “roi”, scale = length_resized / length_original, <br/> if coordinate_transformation_mode is “half_pixel”, <br/> x_original = (x_resized + 0.5) / scale - 0.5, <br/> if coordinate_transformation_mode is “pytorch_half_pixel”, <br/> x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0, <br/> if coordinate_transformation_mode is “align_corners”, <br/> x_original = x_resized * (length_original - 1) / (length_resized - 1), <br/> if coordinate_transformation_mode is “asymmetric”, <br/> x_original = x_resized / scale, <br/> if coordinate_transformation_mode is “tf_half_pixel_for_nn”, <br/> x_original = (x_resized + 0.5) / scale, <br/> if coordinate_transformation_mode is “tf_crop_and_resize”, <br/> x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1). Default value is
'half_pixel'.cubic_coeff_a: The coefficient ‘a’ used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if “mode” is “cubic”. Default value is
-0.75.exclude_outside: If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0. Default value is
0.extrapolation_value: When coordinate_transformation_mode is “tf_crop_and_resize” and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f. Default value is
0.0.mode: Three interpolation modes: nearest (default), linear and cubic. The “linear” mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The “cubic” mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor). Default value is
'nearest'.nearest_mode: Four modes: round_prefer_floor (default, as known as round half down), round_prefer_ceil (as known as round half up), floor, ceil. Only used by nearest interpolation. It indicates how to get “nearest” pixel in input tensor from x_original, so this attribute is valid only if “mode” is “nearest”. Default value is
'round_prefer_floor'.
Inputs
Between 3 and 4 inputs.
X (heterogeneous) - T1: N-D tensor
roi (heterogeneous) - T2: 1-D tensor given as [start1, …, startN, end1, …, endN], where N is the rank of X. The RoIs’ coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is “tf_crop_and_resize”
scales (heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than 0. If it’s less than 1, it’s sampling down, otherwise, it’s upsampling. The number of elements of ‘scales’ should be the same as the rank of input ‘X’. If ‘size’ is needed, the user must set ‘scales’ to an empty tensor.
sizes (optional, heterogeneous) - tensor(int64): The size of the output tensor. The number of elements of ‘sizes’ should be the same as the rank of input ‘X’. May only be set if ‘scales’ is set to an empty tensor.
Outputs
Y (heterogeneous) - T1: N-D tensor after resizing
Type Constraints
T1 in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain roi type to float or double.
OnnxResize_13#
- class mlprodict.npy.xop_auto_import_.OnnxResize_13(*args, **kwargs)#
Version
name: Resize (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) if input "sizes" is not specified.
Attributes
- coordinate_transformation_mode:
This attribute describes how to transform the coordinate in the
resized tensor to the coordinate in the original tensor. <br/> The coordinate of each dimension is transformed individually. Let’s describe a case using axis x as an example. Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, length_original as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input “roi”, scale = length_resized / length_original, <br/> if coordinate_transformation_mode is “half_pixel”, <br/> x_original = (x_resized + 0.5) / scale - 0.5, <br/> if coordinate_transformation_mode is “pytorch_half_pixel”, <br/> x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0, <br/> if coordinate_transformation_mode is “align_corners”, <br/> x_original = x_resized * (length_original - 1) / (length_resized - 1), <br/> if coordinate_transformation_mode is “asymmetric”, <br/> x_original = x_resized / scale, <br/> if coordinate_transformation_mode is “tf_crop_and_resize”, <br/> x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1). Default value is
'half_pixel'.cubic_coeff_a: The coefficient ‘a’ used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if “mode” is “cubic”. Default value is
-0.75.exclude_outside: If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0. Default value is
0.extrapolation_value: When coordinate_transformation_mode is “tf_crop_and_resize” and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f. Default value is
0.0.mode: Three interpolation modes: nearest (default), linear and cubic. The “linear” mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The “cubic” mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor). Default value is
'nearest'.nearest_mode: Four modes: round_prefer_floor (default, as known as round half down), round_prefer_ceil (as known as round half up), floor, ceil. Only used by nearest interpolation. It indicates how to get “nearest” pixel in input tensor from x_original, so this attribute is valid only if “mode” is “nearest”. Default value is
'round_prefer_floor'.
Inputs
Between 1 and 4 inputs.
X (heterogeneous) - T1: N-D tensor
roi (optional, heterogeneous) - T2: 1-D tensor given as [start1, …, startN, end1, …, endN], where N is the rank of X. The RoIs’ coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is “tf_crop_and_resize”
scales (optional, heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than 0. If it’s less than 1, it’s sampling down, otherwise, it’s upsampling. The number of elements of ‘scales’ should be the same as the rank of input ‘X’. One of ‘scales’ and ‘sizes’ MUST be specified and it is an error if both are specified. If ‘sizes’ is needed, the user can use an empty string as the name of ‘scales’ in this operator’s input list.
sizes (optional, heterogeneous) - tensor(int64): The size of the output tensor. The number of elements of ‘sizes’ should be the same as the rank of input ‘X’. Only one of ‘scales’ and ‘sizes’ can be specified.
Outputs
Y (heterogeneous) - T1: N-D tensor after resizing
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain roi type to float or double.
OnnxResize_18#
- class mlprodict.npy.xop_auto_import_.OnnxResize_18(*args, **kwargs)#
Version
name: Resize (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Resize the input tensor. In general, it calculates every value in the output tensor as a weighted average of neighborhood (a.k.a. sampling locations) in the input tensor. Each dimension value of the output tensor is: <br/>
output_dimension = floor(input_dimension * (roi_end - roi_start) * scale) <br/>
if input "sizes" is not specified.
Attributes
antialias: If set to 1, “linear” and “cubic” interpolation modes will use an antialiasing filter when downscaling. Antialiasing is achieved by stretching the resampling filter by a factor max(1, 1 / scale), which means that when downsampling, more input pixels contribute to an output pixel. Default value is
0.axes: If provided, it specifies a subset of axes that ‘roi’, ‘scales’ and ‘sizes’ refer to. If not provided, all axes are assumed [0, 1, …, r-1], where r = rank(data). Non-specified dimensions are interpreted as non-resizable. Negative value means counting dimensions from the back. Accepted range is [-r, r-1], where r = rank(data). Behavior is undefined if an axis is repeated.
- coordinate_transformation_mode:
This attribute describes how to transform the coordinate in the
resized tensor to the coordinate in the original tensor. <br/> The coordinate of each dimension is transformed individually. Let’s describe a case using axis x as an example. Denote x_resized as the coordinate of axis x in the resized tensor, x_original as the coordinate of axis x in the original tensor, length_original as the length of the original tensor in axis x, length_resized as the length of the resized tensor in axis x, roi_x = (start_x, end_x) of the axis x in input “roi”, scale = length_resized / length_original, <br/> if coordinate_transformation_mode is “half_pixel”, <br/> x_original = (x_resized + 0.5) / scale - 0.5 <br/> if coordinate_transformation_mode is “pytorch_half_pixel”, <br/> x_original = length_resized > 1 ? (x_resized + 0.5) / scale - 0.5 : 0 <br/> if coordinate_transformation_mode is “align_corners”, <br/> x_original = x_resized * (length_original - 1) / (length_resized - 1) <br/> if coordinate_transformation_mode is “asymmetric”, <br/> x_original = x_resized / scale <br/> if coordinate_transformation_mode is “tf_crop_and_resize”, <br/> x_original = length_resized > 1 ? start_x * (length_original - 1) + x_resized * (end_x - start_x) * (length_original - 1) / (length_resized - 1) : 0.5 * (start_x + end_x) * (length_original - 1) . Default value is
'half_pixel'.cubic_coeff_a: The coefficient ‘a’ used in cubic interpolation. Two common choice are -0.5 (in some cases of TensorFlow) and -0.75 (in PyTorch). Check out Equation (4) in https://ieeexplore.ieee.org/document/1163711 for the details. This attribute is valid only if mode is “cubic”. Default value is
-0.75.exclude_outside: If set to 1, the weight of sampling locations outside the tensor will be set to 0 and the weight will be renormalized so that their sum is 1.0. The default value is 0. Default value is
0.extrapolation_value: When coordinate_transformation_mode is “tf_crop_and_resize” and x_original is outside the range [0, length_original - 1], this value is used as the corresponding output value. Default is 0.0f. Default value is
0.0.- keep_aspect_ratio_policy:
This attribute describes how to interpret the sizes input with
regard to keeping the original aspect ratio of the input, and it is not applicable when the scales input is used. <br/> Given a set of sizes, associated with a subset of axes (explicitly provided or default), and assuming d = axes[i], with i being the index of the provided sizes. <br/> If keep_aspect_ratio_policy is “stretch”, the original aspect ratio is disregarded, and the input is resized to the specified size: <br/> out_size[d] = sizes[i] <br/> If keep_aspect_ratio_policy is “not_larger”, the sizes are adjusted so that no extent of the output is larger than the specified size, while keeping the original aspect ratio: <br/> scale = Min(sizes[i] / in_size[d]) <br/> out_size[d] = round_int(scale * in_size[i]) <br/> If keep_aspect_ratio_policy is “not_smaller”, the sizes are adjusted so that no extent of the output is smaller than the specified size, while keeping the original aspect ratio: <br/> scale = Max(sizes[i] / in_size[d]) <br/> out_size[d] = round_int(scale * in_size[i]) <br/> For non- resizable axes (those not specified in axes), the output size will be equal to the input size. Note: round_int stands for computing the nearest integer value, rounding halfway cases up. Default value is
'stretch'.mode: Three interpolation modes: “nearest” (default), “linear” and “cubic”. The “linear” mode includes linear interpolation for 1D tensor and N-linear interpolation for N-D tensor (for example, bilinear interpolation for 2D tensor). The “cubic” mode includes cubic interpolation for 1D tensor and N-cubic interpolation for N-D tensor (for example, bicubic interpolation for 2D tensor). Default value is
'nearest'.nearest_mode: Four modes: “round_prefer_floor” (default, as known as round half down), “round_prefer_ceil” (as known as round half up), “floor”, “ceil”. Only used by nearest interpolation. It indicates how to get “nearest” pixel in input tensor from x_original, so this attribute is valid only if “mode” is “nearest”. Default value is
'round_prefer_floor'.
Inputs
Between 1 and 4 inputs.
X (heterogeneous) - T1: N-D tensor
roi (optional, heterogeneous) - T2: 1-D tensor given as [start1, …, startN, end1, …, endN], where N is the rank of X or the length of axes, if provided. The RoIs’ coordinates are normalized in the coordinate system of the input image. It only takes effect when coordinate_transformation_mode is “tf_crop_and_resize”
scales (optional, heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than 0. If it’s less than 1, it’s sampling down, otherwise, it’s upsampling. The number of elements of ‘scales’ should be the same as the rank of input ‘X’ or the length of ‘axes’, if provided. One of ‘scales’ and ‘sizes’ MUST be specified and it is an error if both are specified. If ‘sizes’ is needed, the user can use an empty string as the name of ‘scales’ in this operator’s input list.
sizes (optional, heterogeneous) - tensor(int64): Target size of the output tensor. Its interpretation depends on the ‘keep_aspect_ratio_policy’ value.The number of elements of ‘sizes’ should be the same as the rank of input ‘X’, or the length of ‘axes’, if provided. Only one of ‘scales’ and ‘sizes’ can be specified.
Outputs
Y (heterogeneous) - T1: N-D tensor after resizing
Type Constraints
T1 in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
T2 in ( tensor(double), tensor(float), tensor(float16) ): Constrain roi type to float or double.
OnnxReverseSequence#
- class mlprodict.npy.xop_auto_import_.OnnxReverseSequence(*args, **kwargs)#
Version
name: ReverseSequence (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Reverse batch of sequences having different lengths specified by sequence_lens.
For each slice i iterating on batch axis, the operator reverses the first sequence_lens[i] elements on time axis, and copies elements whose index’s beyond sequence_lens[i] to the output. So the output slice i contains reversed sequences on the first sequence_lens[i] elements, then have original values copied for the other elements.
- Example 1:
- input = [[0.0, 4.0, 8.0, 12.0],
[1.0, 5.0, 9.0, 13.0], [2.0, 6.0, 10.0, 14.0], [3.0, 7.0, 11.0, 15.0]]
sequence_lens = [4, 3, 2, 1] time_axis = 0 batch_axis = 1
- output = [[3.0, 6.0, 9.0, 12.0],
[2.0, 5.0, 8.0, 13.0], [1.0, 4.0, 10.0, 14.0], [0.0, 7.0, 11.0, 15.0]]
- Example 2:
- input = [[0.0, 1.0, 2.0, 3.0 ],
[4.0, 5.0, 6.0, 7.0 ], [8.0, 9.0, 10.0, 11.0], [12.0, 13.0, 14.0, 15.0]]
sequence_lens = [1, 2, 3, 4] time_axis = 1 batch_axis = 0
- output = [[0.0, 1.0, 2.0, 3.0 ],
[5.0, 4.0, 6.0, 7.0 ], [10.0, 9.0, 8.0, 11.0], [15.0, 14.0, 13.0, 12.0]]
Attributes
batch_axis: (Optional) Specify which axis is batch axis. Must be one of 1 (default), or 0. Default value is
1.time_axis: (Optional) Specify which axis is time axis. Must be one of 0 (default), or 1. Default value is
0.
Inputs
input (heterogeneous) - T: Tensor of rank r >= 2.
sequence_lens (heterogeneous) - tensor(int64): Tensor specifying lengths of the sequences in a batch. It has shape [batch_size].
Outputs
Y (heterogeneous) - T: Tensor with same shape of input.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
OnnxReverseSequence_10#
- class mlprodict.npy.xop_auto_import_.OnnxReverseSequence_10(*args, **kwargs)#
Version
name: ReverseSequence (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Reverse batch of sequences having different lengths specified by sequence_lens.
For each slice i iterating on batch axis, the operator reverses the first sequence_lens[i] elements on time axis, and copies elements whose index’s beyond sequence_lens[i] to the output. So the output slice i contains reversed sequences on the first sequence_lens[i] elements, then have original values copied for the other elements.
- Example 1:
- input = [[0.0, 4.0, 8.0, 12.0],
[1.0, 5.0, 9.0, 13.0], [2.0, 6.0, 10.0, 14.0], [3.0, 7.0, 11.0, 15.0]]
sequence_lens = [4, 3, 2, 1] time_axis = 0 batch_axis = 1
- output = [[3.0, 6.0, 9.0, 12.0],
[2.0, 5.0, 8.0, 13.0], [1.0, 4.0, 10.0, 14.0], [0.0, 7.0, 11.0, 15.0]]
- Example 2:
- input = [[0.0, 1.0, 2.0, 3.0 ],
[4.0, 5.0, 6.0, 7.0 ], [8.0, 9.0, 10.0, 11.0], [12.0, 13.0, 14.0, 15.0]]
sequence_lens = [1, 2, 3, 4] time_axis = 1 batch_axis = 0
- output = [[0.0, 1.0, 2.0, 3.0 ],
[5.0, 4.0, 6.0, 7.0 ], [10.0, 9.0, 8.0, 11.0], [15.0, 14.0, 13.0, 12.0]]
Attributes
batch_axis: (Optional) Specify which axis is batch axis. Must be one of 1 (default), or 0. Default value is
1.time_axis: (Optional) Specify which axis is time axis. Must be one of 0 (default), or 1. Default value is
0.
Inputs
input (heterogeneous) - T: Tensor of rank r >= 2.
sequence_lens (heterogeneous) - tensor(int64): Tensor specifying lengths of the sequences in a batch. It has shape [batch_size].
Outputs
Y (heterogeneous) - T: Tensor with same shape of input.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
OnnxRoiAlign#
- class mlprodict.npy.xop_auto_import_.OnnxRoiAlign(*args, **kwargs)#
Version
name: RoiAlign (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Region of Interest (RoI) align operation described in the [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). RoiAlign consumes an input tensor X and region of interests (rois) to apply pooling across each RoI; it produces a 4-D tensor of shape (num_rois, C, output_height, output_width).
RoiAlign is proposed to avoid the misalignment by removing quantizations while converting from original image into feature map and from feature map into RoI feature; in each ROI bin, the value of the sampled locations are computed directly through bilinear interpolation.
Attributes
coordinate_transformation_mode: Allowed values are ‘half_pixel’ and ‘output_half_pixel’. Use the value ‘half_pixel’ to pixel shift the input coordinates by -0.5 (the recommended behavior). Use the value ‘output_half_pixel’ to omit the pixel shift for the input (use this for a backward-compatible behavior). Default value is
'half_pixel'.mode: The pooling method. Two modes are supported: ‘avg’ and ‘max’. Default is ‘avg’. Default value is
'avg'.output_height: default 1; Pooled output Y’s height. Default value is
1.output_width: default 1; Pooled output Y’s width. Default value is
1.sampling_ratio: Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0. Default value is
0.spatial_scale: Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f. Default value is
1.0.
Inputs
X (heterogeneous) - T1: Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T1: RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], …]. The RoIs’ coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the ‘batch_indices’ input.
batch_indices (heterogeneous) - T2: 1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
Outputs
Y (heterogeneous) - T1: RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain types to float tensors.
T2 in ( tensor(int64) ): Constrain types to int tensors.
OnnxRoiAlign_10#
- class mlprodict.npy.xop_auto_import_.OnnxRoiAlign_10(*args, **kwargs)#
Version
name: RoiAlign (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Region of Interest (RoI) align operation described in the [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). RoiAlign consumes an input tensor X and region of interests (rois) to apply pooling across each RoI; it produces a 4-D tensor of shape (num_rois, C, output_height, output_width).
RoiAlign is proposed to avoid the misalignment by removing quantizations while converting from original image into feature map and from feature map into RoI feature; in each ROI bin, the value of the sampled locations are computed directly through bilinear interpolation.
Attributes
mode: The pooling method. Two modes are supported: ‘avg’ and ‘max’. Default is ‘avg’. Default value is
'avg'.output_height: default 1; Pooled output Y’s height. Default value is
1.output_width: default 1; Pooled output Y’s width. Default value is
1.sampling_ratio: Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0. Default value is
0.spatial_scale: Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f. Default value is
1.0.
Inputs
X (heterogeneous) - T1: Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T1: RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], …]. The RoIs’ coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the ‘batch_indices’ input.
batch_indices (heterogeneous) - T2: 1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
Outputs
Y (heterogeneous) - T1: RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain types to float tensors.
T2 in ( tensor(int64) ): Constrain types to int tensors.
OnnxRoiAlign_16#
- class mlprodict.npy.xop_auto_import_.OnnxRoiAlign_16(*args, **kwargs)#
Version
name: RoiAlign (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Region of Interest (RoI) align operation described in the [Mask R-CNN paper](https://arxiv.org/abs/1703.06870). RoiAlign consumes an input tensor X and region of interests (rois) to apply pooling across each RoI; it produces a 4-D tensor of shape (num_rois, C, output_height, output_width).
RoiAlign is proposed to avoid the misalignment by removing quantizations while converting from original image into feature map and from feature map into RoI feature; in each ROI bin, the value of the sampled locations are computed directly through bilinear interpolation.
Attributes
coordinate_transformation_mode: Allowed values are ‘half_pixel’ and ‘output_half_pixel’. Use the value ‘half_pixel’ to pixel shift the input coordinates by -0.5 (the recommended behavior). Use the value ‘output_half_pixel’ to omit the pixel shift for the input (use this for a backward-compatible behavior). Default value is
'half_pixel'.mode: The pooling method. Two modes are supported: ‘avg’ and ‘max’. Default is ‘avg’. Default value is
'avg'.output_height: default 1; Pooled output Y’s height. Default value is
1.output_width: default 1; Pooled output Y’s width. Default value is
1.sampling_ratio: Number of sampling points in the interpolation grid used to compute the output value of each pooled output bin. If > 0, then exactly sampling_ratio x sampling_ratio grid points are used. If == 0, then an adaptive number of grid points are used (computed as ceil(roi_width / output_width), and likewise for height). Default is 0. Default value is
0.spatial_scale: Multiplicative spatial scale factor to translate ROI coordinates from their input spatial scale to the scale used when pooling, i.e., spatial scale of the input feature map X relative to the input image. E.g.; default is 1.0f. Default value is
1.0.
Inputs
X (heterogeneous) - T1: Input data tensor from the previous operator; 4-D feature map of shape (N, C, H, W), where N is the batch size, C is the number of channels, and H and W are the height and the width of the data.
rois (heterogeneous) - T1: RoIs (Regions of Interest) to pool over; rois is 2-D input of shape (num_rois, 4) given as [[x1, y1, x2, y2], …]. The RoIs’ coordinates are in the coordinate system of the input image. Each coordinate set has a 1:1 correspondence with the ‘batch_indices’ input.
batch_indices (heterogeneous) - T2: 1-D tensor of shape (num_rois,) with each element denoting the index of the corresponding image in the batch.
Outputs
Y (heterogeneous) - T1: RoI pooled output, 4-D tensor of shape (num_rois, C, output_height, output_width). The r-th batch element Y[r-1] is a pooled feature map corresponding to the r-th RoI X[r-1].
Type Constraints
T1 in ( tensor(double), tensor(float), tensor(float16) ): Constrain types to float tensors.
T2 in ( tensor(int64) ): Constrain types to int tensors.
OnnxRound#
- class mlprodict.npy.xop_auto_import_.OnnxRound(*args, **kwargs)#
Version
name: Round (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Round takes one input Tensor and rounds the values, element-wise, meaning it finds the nearest integer for each value. In case of halfs, the rule is to round them to the nearest even integer. The output tensor has the same shape and type as the input.
Examples:
round([0.9]) = [1.0] round([2.5]) = [2.0] round([2.3]) = [2.0] round([1.5]) = [2.0] round([-4.5]) = [-4.0]
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxRound_11#
- class mlprodict.npy.xop_auto_import_.OnnxRound_11(*args, **kwargs)#
Version
name: Round (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Round takes one input Tensor and rounds the values, element-wise, meaning it finds the nearest integer for each value. In case of halfs, the rule is to round them to the nearest even integer. The output tensor has the same shape and type as the input.
Examples:
round([0.9]) = [1.0] round([2.5]) = [2.0] round([2.3]) = [2.0] round([1.5]) = [2.0] round([-4.5]) = [-4.0]
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSTFT#
- class mlprodict.npy.xop_auto_import_.OnnxSTFT(*args, **kwargs)#
Version
name: STFT (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Computes the Short-time Fourier Transform of the signal.
Attributes
onesided: If onesided is 1, only values for w in [0, 1, 2, …, floor(n_fft/2) + 1] are returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m,w]=X[m,n_fft-w]*. Note if the input or window tensors are complex, then onesided output is not possible. Enabling onesided with real inputs performs a Real-valued fast Fourier transform (RFFT).When invoked with real or complex valued input, the default value is 1. Values can be 0 or 1. Default value is
1.
Inputs
Between 2 and 4 inputs.
signal (heterogeneous) - T1: Input tensor representing a real or complex valued signal. For real input, the following shape is expected: [batch_size][signal_length][1]. For complex input, the following shape is expected: [batch_size][signal_length][2], where [batch_size][signal_length][0] represents the real component and [batch_size][signal_length][1] represents the imaginary component of the signal.
frame_step (heterogeneous) - T2: The number of samples to step between successive DFTs.
window (optional, heterogeneous) - T1: A tensor representing the window that will be slid over the signal.The window must have rank 1 with shape: [window_shape]. It’s an optional value.
frame_length (optional, heterogeneous) - T2: A scalar representing the size of the DFT. It’s an optional value.
Outputs
output (heterogeneous) - T1: The Short-time Fourier Transform of the signals.If onesided is 1, the output has the shape: [batch_size][frames][dft_unique_bins][2], where dft_unique_bins is frame_length // 2 + 1 (the unique components of the DFT) If onesided is 0, the output has the shape: [batch_size][frames][frame_length][2], where frame_length is the length of the DFT.
Type Constraints
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain signal and output to float tensors.
T2 in ( tensor(int32), tensor(int64) ): Constrain scalar length types to int64_t.
OnnxSTFT_17#
- class mlprodict.npy.xop_auto_import_.OnnxSTFT_17(*args, **kwargs)#
Version
name: STFT (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Computes the Short-time Fourier Transform of the signal.
Attributes
onesided: If onesided is 1, only values for w in [0, 1, 2, …, floor(n_fft/2) + 1] are returned because the real-to-complex Fourier transform satisfies the conjugate symmetry, i.e., X[m, w] = X[m,w]=X[m,n_fft-w]*. Note if the input or window tensors are complex, then onesided output is not possible. Enabling onesided with real inputs performs a Real-valued fast Fourier transform (RFFT).When invoked with real or complex valued input, the default value is 1. Values can be 0 or 1. Default value is
1.
Inputs
Between 2 and 4 inputs.
signal (heterogeneous) - T1: Input tensor representing a real or complex valued signal. For real input, the following shape is expected: [batch_size][signal_length][1]. For complex input, the following shape is expected: [batch_size][signal_length][2], where [batch_size][signal_length][0] represents the real component and [batch_size][signal_length][1] represents the imaginary component of the signal.
frame_step (heterogeneous) - T2: The number of samples to step between successive DFTs.
window (optional, heterogeneous) - T1: A tensor representing the window that will be slid over the signal.The window must have rank 1 with shape: [window_shape]. It’s an optional value.
frame_length (optional, heterogeneous) - T2: A scalar representing the size of the DFT. It’s an optional value.
Outputs
output (heterogeneous) - T1: The Short-time Fourier Transform of the signals.If onesided is 1, the output has the shape: [batch_size][frames][dft_unique_bins][2], where dft_unique_bins is frame_length // 2 + 1 (the unique components of the DFT) If onesided is 0, the output has the shape: [batch_size][frames][frame_length][2], where frame_length is the length of the DFT.
Type Constraints
T1 in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain signal and output to float tensors.
T2 in ( tensor(int32), tensor(int64) ): Constrain scalar length types to int64_t.
OnnxScan#
- class mlprodict.npy.xop_auto_import_.OnnxScan(*args, **kwargs)#
Version
name: Scan (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Scan can be used to iterate over one or more scan_input tensors, constructing zero or more scan_output tensors. It combines ideas from general recurrences, functional programming constructs such as scan, fold, map, and zip, and is intended to enable generalizations of RNN-like constructs for sequence-to-sequence processing. Other tensors (referred to as state_variables here) can be used to carry a state when iterating from one element to another (similar to hidden-state in RNNs, also referred to as loop-carried dependences in the context of loops). Many common usages involve a single scan_input tensor (where functionality similar to scan, fold and map can be obtained). When more than one scan_input is used, a behavior similar to zip is obtained.
The attribute body must be a graph, specifying the computation to be performed in every iteration. It takes as input the current values of the state_variables and the current iterated element of the scan_inputs. It must return the (updated) values of the state_variables and zero or more scan_output_element tensors. The values of the scan_output_element tensors are concatenated over all the iterations to produce the scan_output values of the scan construct (similar to the concatenated intermediate hidden-state values of RNN-like constructs). All the output tensors (state_variables as well as scan_output_element tensors) are required to have the same shape in each iteration of the loop (a restriction imposed to enable efficient memory allocation).
Note that the iterated element passed to the body subgraph does not have a sequence axis. It will have a rank one less than the rank of the corresponding scan_input.
The scan operation returns the final values of the state_variables as well as the scan_outputs.
The optional attribute scan_input_directions specifies the direction (forward or backward) for each scan input. If this attribute is omitted, all sequences are scanned in the forward direction. A bidirectional scan may be performed by specifying the same tensor input twice in the scan_inputs, once with a forward direction, and once with a backward direction.
The scan_output of the operation is produced by concatenating the scan_output_element values produced by the body in each iteration. The optional attribute scan_output_directions specifies the direction in which scan_output is constructed (by appending or prepending the scan_output_element to scan_output in each iteration) for each scan_output. If this attribute is omitted, the scan_output_element is appended to the scan_output in each iteration.
The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. Note that scanning a non-zero axis may be less efficient than scanning axis zero.
The optional attribute scan_output_axes specifies the axis along which the scan_outputs are accumulated for each scan_output. For example, if axis 1 is the time axis (to be scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis value of 1.
Note that because of the ONNX restriction that only the last parameter of an operator can be variadic, the initial-states and scan-inputs are listed together as one input parameter. Similarly, the final-states and scan-outputs are listed together as one output parameter. The attribute num_scan_inputs indicates the number M of scan-inputs.
The behavior of
- Scan <
num_scan_inputs = m, body = loop-body, scan_input_axes = [axis_1, …, axis_m]
> (init_1, …, init_n, scan_1, …, scan_m)
is equivalent to the following pseudo-code:
// scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. sequence_length = scan_1.shape[axis_1];
// initialize state-variables st_1 = init_1; … st_n = init_n; // initialize scan-output variables: [] denotes an empty tensor scan_out_1 = []; …; scan_out_k = []; // identify number of iterations:
// execute loop for (int t = 0; t < sequence_length; ++t) {
// generate the scan-input elements: the notation T<axis=k>[t] indicates the sub-tensor // of rank one less than T obtained by indexing T at position t along axis k. si_1 = scan_1<axis=axis_1>[t]; … ; si_m = scan_m<axis=axis_m>[t]; // execute loop-body st_1, …, st_n, so_1, …, so_k = loop-body(st_1, …, st_n, si_1, …, si_m) // accumulate the scan-output elements scan_out_1 = Concat<axis=0>(scan_out_1, so_1); … ; scan_out_k = Concat<axis=0>(scan_out_k, so_k);
}
return st_1, …, st_n, scan_out_1, …, scan_out_k;
Sample usage: Encoding RNN using a Scan
The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these values are computed in the outer graph, they need to be passed in as extra state_variables.
- graph rnn-encoding {
%H_0 = … %X = … %Y_h, %Y = Scan[body = <graph rnn-cell-1>, num_scan_inputs=1](%H_0, %X) return %Y, %Y_h
}
- graph rnn-cell-1 (
%H_tminus1[FLOAT, tensor] %X_t[FLOAT, tensor]
- ) {
%Wi = … %Ri = … %Wbi = … %Rbi = … %t1 = X_t * (Wi^T) %t2 = H_tminus1*(Ri^T) %t3 = Add(%t1, %t2) %t4 = Add(%t3, %Wbi) %t5 = Add(%t4, %Rbi) %Ht = Tanh(%t5) %Accumulate = Identity(%Ht) return %Ht, %Accumulate
}
Attributes
body (required): The graph run each iteration. It has N+M inputs: (loop state variables…, scan_input_elts…). It has N+K outputs: (loop state variables…, scan_output_elts…). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
num_scan_inputs (required): An attribute specifying the number of scan_inputs M.
scan_input_axes: An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
scan_input_directions: An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
scan_output_axes: An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
scan_output_directions: An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
Inputs
Between 1 and 2147483647 inputs.
initial_state_and_scan_inputs (variadic) - V: Initial values of the loop’s N state variables followed by M scan_inputs
Outputs
Between 1 and 2147483647 outputs.
final_state_and_scan_outputs (variadic) - V: Final values of the loop’s N state variables followed by K scan_outputs
Type Constraints
V in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
OnnxScan_11#
- class mlprodict.npy.xop_auto_import_.OnnxScan_11(*args, **kwargs)#
Version
name: Scan (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Scan can be used to iterate over one or more scan_input tensors, constructing zero or more scan_output tensors. It combines ideas from general recurrences, functional programming constructs such as scan, fold, map, and zip, and is intended to enable generalizations of RNN-like constructs for sequence-to-sequence processing. Other tensors (referred to as state_variables here) can be used to carry a state when iterating from one element to another (similar to hidden-state in RNNs, also referred to as loop-carried dependences in the context of loops). Many common usages involve a single scan_input tensor (where functionality similar to scan, fold and map can be obtained). When more than one scan_input is used, a behavior similar to zip is obtained.
The attribute body must be a graph, specifying the computation to be performed in every iteration. It takes as input the current values of the state_variables and the current iterated element of the scan_inputs. It must return the (updated) values of the state_variables and zero or more scan_output_element tensors. The values of the scan_output_element tensors are concatenated over all the iterations to produce the scan_output values of the scan construct (similar to the concatenated intermediate hidden-state values of RNN-like constructs). All the output tensors (state_variables as well as scan_output_element tensors) are required to have the same shape in each iteration of the loop (a restriction imposed to enable efficient memory allocation).
Note that the iterated element passed to the body subgraph does not have a sequence axis. It will have a rank one less than the rank of the corresponding scan_input.
The scan operation returns the final values of the state_variables as well as the scan_outputs.
The optional attribute scan_input_directions specifies the direction (forward or backward) for each scan input. If this attribute is omitted, all sequences are scanned in the forward direction. A bidirectional scan may be performed by specifying the same tensor input twice in the scan_inputs, once with a forward direction, and once with a backward direction.
The scan_output of the operation is produced by concatenating the scan_output_element values produced by the body in each iteration. The optional attribute scan_output_directions specifies the direction in which scan_output is constructed (by appending or prepending the scan_output_element to scan_output in each iteration) for each scan_output. If this attribute is omitted, the scan_output_element is appended to the scan_output in each iteration.
The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. Note that scanning a non-zero axis may be less efficient than scanning axis zero.
The optional attribute scan_output_axes specifies the axis along which the scan_outputs are accumulated for each scan_output. For example, if axis 1 is the time axis (to be scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis value of 1.
Note that because of the ONNX restriction that only the last parameter of an operator can be variadic, the initial-states and scan-inputs are listed together as one input parameter. Similarly, the final-states and scan-outputs are listed together as one output parameter. The attribute num_scan_inputs indicates the number M of scan-inputs.
The behavior of
- Scan <
num_scan_inputs = m, body = loop-body, scan_input_axes = [axis_1, …, axis_m]
> (init_1, …, init_n, scan_1, …, scan_m)
is equivalent to the following pseudo-code:
// scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. sequence_length = scan_1.shape[axis_1];
// initialize state-variables st_1 = init_1; … st_n = init_n; // initialize scan-output variables: [] denotes an empty tensor scan_out_1 = []; …; scan_out_k = []; // identify number of iterations:
// execute loop for (int t = 0; t < sequence_length; ++t) {
// generate the scan-input elements: the notation T<axis=k>[t] indicates the sub-tensor // of rank one less than T obtained by indexing T at position t along axis k. si_1 = scan_1<axis=axis_1>[t]; … ; si_m = scan_m<axis=axis_m>[t]; // execute loop-body st_1, …, st_n, so_1, …, so_k = loop-body(st_1, …, st_n, si_1, …, si_m) // accumulate the scan-output elements scan_out_1 = Concat<axis=0>(scan_out_1, so_1); … ; scan_out_k = Concat<axis=0>(scan_out_k, so_k);
}
return st_1, …, st_n, scan_out_1, …, scan_out_k;
Sample usage: Encoding RNN using a Scan
The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these values are computed in the outer graph, they need to be passed in as extra state_variables.
- graph rnn-encoding {
%H_0 = … %X = … %Y_h, %Y = Scan[body = <graph rnn-cell-1>, num_scan_inputs=1](%H_0, %X) return %Y, %Y_h
}
- graph rnn-cell-1 (
%H_tminus1[FLOAT, tensor] %X_t[FLOAT, tensor]
- ) {
%Wi = … %Ri = … %Wbi = … %Rbi = … %t1 = X_t * (Wi^T) %t2 = H_tminus1*(Ri^T) %t3 = Add(%t1, %t2) %t4 = Add(%t3, %Wbi) %t5 = Add(%t4, %Rbi) %Ht = Tanh(%t5) %Accumulate = Identity(%Ht) return %Ht, %Accumulate
}
Attributes
body (required): The graph run each iteration. It has N+M inputs: (loop state variables…, scan_input_elts…). It has N+K outputs: (loop state variables…, scan_output_elts…). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
num_scan_inputs (required): An attribute specifying the number of scan_inputs M.
scan_input_axes: An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
scan_input_directions: An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
scan_output_axes: An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
scan_output_directions: An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
Inputs
Between 1 and 2147483647 inputs.
initial_state_and_scan_inputs (variadic) - V: Initial values of the loop’s N state variables followed by M scan_inputs
Outputs
Between 1 and 2147483647 outputs.
final_state_and_scan_outputs (variadic) - V: Final values of the loop’s N state variables followed by K scan_outputs
Type Constraints
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
OnnxScan_16#
- class mlprodict.npy.xop_auto_import_.OnnxScan_16(*args, **kwargs)#
Version
name: Scan (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Scan can be used to iterate over one or more scan_input tensors, constructing zero or more scan_output tensors. It combines ideas from general recurrences, functional programming constructs such as scan, fold, map, and zip, and is intended to enable generalizations of RNN-like constructs for sequence-to-sequence processing. Other tensors (referred to as state_variables here) can be used to carry a state when iterating from one element to another (similar to hidden-state in RNNs, also referred to as loop-carried dependences in the context of loops). Many common usages involve a single scan_input tensor (where functionality similar to scan, fold and map can be obtained). When more than one scan_input is used, a behavior similar to zip is obtained.
The attribute body must be a graph, specifying the computation to be performed in every iteration. It takes as input the current values of the state_variables and the current iterated element of the scan_inputs. It must return the (updated) values of the state_variables and zero or more scan_output_element tensors. The values of the scan_output_element tensors are concatenated over all the iterations to produce the scan_output values of the scan construct (similar to the concatenated intermediate hidden-state values of RNN-like constructs). All the output tensors (state_variables as well as scan_output_element tensors) are required to have the same shape in each iteration of the loop (a restriction imposed to enable efficient memory allocation).
Note that the iterated element passed to the body subgraph does not have a sequence axis. It will have a rank one less than the rank of the corresponding scan_input.
The scan operation returns the final values of the state_variables as well as the scan_outputs.
The optional attribute scan_input_directions specifies the direction (forward or backward) for each scan input. If this attribute is omitted, all sequences are scanned in the forward direction. A bidirectional scan may be performed by specifying the same tensor input twice in the scan_inputs, once with a forward direction, and once with a backward direction.
The scan_output of the operation is produced by concatenating the scan_output_element values produced by the body in each iteration. The optional attribute scan_output_directions specifies the direction in which scan_output is constructed (by appending or prepending the scan_output_element to scan_output in each iteration) for each scan_output. If this attribute is omitted, the scan_output_element is appended to the scan_output in each iteration.
The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. Note that scanning a non-zero axis may be less efficient than scanning axis zero.
The optional attribute scan_output_axes specifies the axis along which the scan_outputs are accumulated for each scan_output. For example, if axis 1 is the time axis (to be scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis value of 1.
Note that because of the ONNX restriction that only the last parameter of an operator can be variadic, the initial-states and scan-inputs are listed together as one input parameter. Similarly, the final-states and scan-outputs are listed together as one output parameter. The attribute num_scan_inputs indicates the number M of scan-inputs.
The behavior of
- Scan <
num_scan_inputs = m, body = loop-body, scan_input_axes = [axis_1, …, axis_m]
> (init_1, …, init_n, scan_1, …, scan_m)
is equivalent to the following pseudo-code:
// scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. sequence_length = scan_1.shape[axis_1];
// initialize state-variables st_1 = init_1; … st_n = init_n; // initialize scan-output variables: [] denotes an empty tensor scan_out_1 = []; …; scan_out_k = []; // identify number of iterations:
// execute loop for (int t = 0; t < sequence_length; ++t) {
// generate the scan-input elements: the notation T<axis=k>[t] indicates the sub-tensor // of rank one less than T obtained by indexing T at position t along axis k. si_1 = scan_1<axis=axis_1>[t]; … ; si_m = scan_m<axis=axis_m>[t]; // execute loop-body st_1, …, st_n, so_1, …, so_k = loop-body(st_1, …, st_n, si_1, …, si_m) // accumulate the scan-output elements scan_out_1 = Concat<axis=0>(scan_out_1, so_1); … ; scan_out_k = Concat<axis=0>(scan_out_k, so_k);
}
return st_1, …, st_n, scan_out_1, …, scan_out_k;
Sample usage: Encoding RNN using a Scan
The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these values are computed in the outer graph, they need to be passed in as extra state_variables.
- graph rnn-encoding {
%H_0 = … %X = … %Y_h, %Y = Scan[body = <graph rnn-cell-1>, num_scan_inputs=1](%H_0, %X) return %Y, %Y_h
}
- graph rnn-cell-1 (
%H_tminus1[FLOAT, tensor] %X_t[FLOAT, tensor]
- ) {
%Wi = … %Ri = … %Wbi = … %Rbi = … %t1 = X_t * (Wi^T) %t2 = H_tminus1*(Ri^T) %t3 = Add(%t1, %t2) %t4 = Add(%t3, %Wbi) %t5 = Add(%t4, %Rbi) %Ht = Tanh(%t5) %Accumulate = Identity(%Ht) return %Ht, %Accumulate
}
Attributes
body (required): The graph run each iteration. It has N+M inputs: (loop state variables…, scan_input_elts…). It has N+K outputs: (loop state variables…, scan_output_elts…). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
num_scan_inputs (required): An attribute specifying the number of scan_inputs M.
scan_input_axes: An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
scan_input_directions: An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
scan_output_axes: An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output. Negative value for an axis means counting dimensions from the back. Accepted range is [-r, r-1].
scan_output_directions: An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
Inputs
Between 1 and 2147483647 inputs.
initial_state_and_scan_inputs (variadic) - V: Initial values of the loop’s N state variables followed by M scan_inputs
Outputs
Between 1 and 2147483647 outputs.
final_state_and_scan_outputs (variadic) - V: Final values of the loop’s N state variables followed by K scan_outputs
Type Constraints
V in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
OnnxScan_8#
- class mlprodict.npy.xop_auto_import_.OnnxScan_8(*args, **kwargs)#
Version
name: Scan (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
Scan can be used to iterate over one or more scan_input tensors, constructing zero or more scan_output tensors. It combines ideas from general recurrences, functional programming constructs such as scan, fold, map, and zip, and is intended to enable generalizations of RNN-like constructs for sequence-to-sequence processing. Other tensors (referred to as state_variables here) can be used to carry a state when iterating from one element to another (similar to hidden-state in RNNs, also referred to as loop-carried dependences in the context of loops). All these tensors are required to have the same shape in each iteration of the loop (a restriction imposed to enable efficient memory allocation). Many common usages involve a single scan_input tensor (where functionality similar to scan, fold and map can be obtained). When more than one scan_input is used, a behavior similar to zip is obtained.
The attribute body must be a graph, specifying the computation to be performed in every iteration. It takes as input the current values of the state_variables and the current iterated element of the scan_inputs. It must return the (updated) values of the state_variables and zero or more scan_output_element tensors. The values of the scan_output_element tensors are concatenated over all the iterations to produce the scan_output values of the scan construct (similar to the concatenated intermediate hidden-state values of RNN-like constructs).
The scan operation returns the final values of the state_variables as well as the scan_outputs.
The operation supports batching, and the batch-axis is required to be 0. When multiple scan_input tensors are used, they must all have the same batch-size, and they must all have the same maximum-sequence-length (the dimensionality of the sequence axis or scan axis). The sequence axis or scan axis is required to be 1.
The operation has an optional sequence_lens input (of shape [BATCH_SIZE]) to allow variable length sequences of length <= the maximum-sequence-length. If this input is not specified, all sequences are assumed to be of length equal to maximum-sequence-length. For variable length input sequences, the scan_outputs will consist of a sequence of same length as the input, padded to the maximum-sequence-length.
The optional attribute directions can be used to scan a sequence in the reverse direction. If this attribute is omitted, all sequences are scanned in the forward direction. A bidirectional scan be performed by specifying the same tensor input twice in the scan_inputs, once with a forward direction, and once with a backward direction.
Note that because of the ONNX restriction that only the last parameter of an operator can be variadic, the initial-states and scan-inputs are listed together as one input parameter. Similarly, the final-states and scan-outputs are listed together as one output parameter. The attribute num_scan_inputs indicates the number M of scan-inputs.
The behavior of
- Scan <
num_scan_inputs = m, body = loop-body
> (sequence_lengths, init_1, …, init_n, scan_1, …, scan_m)
is equivalent to the following pseudo-code:
// T.shape[0] denotes the batch-size of T // The batch-size of scan_1, …, scan_m are all required to be equal batch_size = scan_1.shape[0];
// scan_i.shape[1] denotes the (max) sequence-length of scan_i // scan_i.shape[1] is required to be equal to scan_j.shape[1] for all i,j. max_sequence_length = scan_1.shape[1];
- for (int batch = 0; batch < batch_size; ++batch) {
// initialize state-variables st_1 = init_1; … st_n = init_n; // initialize scan-output variables: [] denotes an empty tensor scan_out_1 = []; …; scan_out_k = []; // identify number of iterations: N = (sequence_lengths specified) ? sequence_lengths[batch] : max_sequence_length;
// execute loop for (int t = 0; t < N; ++t) {
// generate the scan-input elements: the notation T<axis=k>[t] indicates the sub-tensor // of rank one less than T obtained by indexing T at position t along axis k. si_1 = (scan_1<axis=0>[batch])<axis=1>[t]; … ; si_m = (scan_m<axis=0>[batch])<axis=1>[t]; // execute loop-body st_1, …, st_n, so_1, …, so_k = loop-body(st_1, …, st_n, si_1, …, si_m) // accumulate the scan-output elements scan_out_1 = Concat<axis=0>(scan_out_1, so_1); … ; scan_out_k = Concat<axis=0>(scan_out_k, so_k);
} // accumulate the outputs for this batch: bst_1[batch] = st_1; …, bst_n[batch] = st_n; // Note scan-outputs will have size max_sequence_length, but only first N values will be meaningful. // The remaining values have an undefined value. b_scan_out_1[batch] = scan_out_1; …; b_scan_out_k[batch] = scan_out_k;
} return bst_1, …, bst_n, b_scan_out_1, …, b_scan_out_k;
Sample usage: Encoding RNN using a Scan
The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these values are computed in the outer graph, they need to be passed in as extra state_variables.
- graph rnn-encoding {
%H_0 = … %X = … %Y_h, %Y = Scan[body = <graph rnn-cell-1>, num_scan_inputs=1](“”, %H_0, %X) return %Y, %Y_h
}
- graph rnn-cell-1 (
%H_tminus1[FLOAT, tensor] %X_t[FLOAT, tensor]
- ) {
%Wi = … %Ri = … %Wbi = … %Rbi = … %t1 = X_t * (Wi^T) %t2 = H_tminus1*(Ri^T) %t3 = Add(%t1, %t2) %t4 = Add(%t3, %Wbi) %t5 = Add(%t4, %Rbi) %Ht = Tanh(%t5) %Accumulate = Identity(%Ht) return %Ht, %Accumulate
}
Attributes
body (required): The graph run each iteration. It has N+M inputs: (loop state variables…, scan_input_elts…). It has N+K outputs: (loop state variables…, scan_output_elts…). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
directions: An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
num_scan_inputs (required): An attribute specifying the number of scan_inputs M.
Inputs
Between 2 and 2147483647 inputs.
sequence_lens (optional, heterogeneous) - I: Optional tensor specifying lengths of the sequences in a batch. If this input is not specified, all sequences are assumed to be of the maximum sequence length (the dimension of the sequence axis of the scan_input tensors).
initial_state_and_scan_inputs (variadic) - V: Initial values of the loop’s N state variables followed by M scan_inputs
Outputs
Between 1 and 2147483647 outputs.
final_state_and_scan_outputs (variadic) - V: Final values of the loop’s N state variables followed by K scan_outputs
Type Constraints
I in ( tensor(int64) ): Int64 tensor
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
OnnxScan_9#
- class mlprodict.npy.xop_auto_import_.OnnxScan_9(*args, **kwargs)#
Version
name: Scan (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Scan can be used to iterate over one or more scan_input tensors, constructing zero or more scan_output tensors. It combines ideas from general recurrences, functional programming constructs such as scan, fold, map, and zip, and is intended to enable generalizations of RNN-like constructs for sequence-to-sequence processing. Other tensors (referred to as state_variables here) can be used to carry a state when iterating from one element to another (similar to hidden-state in RNNs, also referred to as loop-carried dependences in the context of loops). Many common usages involve a single scan_input tensor (where functionality similar to scan, fold and map can be obtained). When more than one scan_input is used, a behavior similar to zip is obtained.
The attribute body must be a graph, specifying the computation to be performed in every iteration. It takes as input the current values of the state_variables and the current iterated element of the scan_inputs. It must return the (updated) values of the state_variables and zero or more scan_output_element tensors. The values of the scan_output_element tensors are concatenated over all the iterations to produce the scan_output values of the scan construct (similar to the concatenated intermediate hidden-state values of RNN-like constructs). All the output tensors (state_variables as well as scan_output_element tensors) are required to have the same shape in each iteration of the loop (a restriction imposed to enable efficient memory allocation).
Note that the iterated element passed to the body subgraph does not have a sequence axis. It will have a rank one less than the rank of the corresponding scan_input.
The scan operation returns the final values of the state_variables as well as the scan_outputs.
The optional attribute scan_input_directions specifies the direction (forward or backward) for each scan input. If this attribute is omitted, all sequences are scanned in the forward direction. A bidirectional scan may be performed by specifying the same tensor input twice in the scan_inputs, once with a forward direction, and once with a backward direction.
The scan_output of the operation is produced by concatenating the scan_output_element values produced by the body in each iteration. The optional attribute scan_output_directions specifies the direction in which scan_output is constructed (by appending or prepending the scan_output_element to scan_output in each iteration) for each scan_output. If this attribute is omitted, the scan_output_element is appended to the scan_output in each iteration.
The optional attribute scan_input_axes specifies the axis to be scanned for each scan_input. If omitted, every scan_input will be scanned in axis 0. For example, if axis 0 is the batch axis and axis 1 is the time axis (to be scanned), specify an axis value of 1. Note that scanning a non-zero axis may be less efficient than scanning axis zero.
The optional attribute scan_output_axes specifies the axis along which the scan_outputs are accumulated for each scan_output. For example, if axis 1 is the time axis (to be scanned) for both inputs and outputs, specify a scan_input axis and scan_output axis value of 1.
Note that because of the ONNX restriction that only the last parameter of an operator can be variadic, the initial-states and scan-inputs are listed together as one input parameter. Similarly, the final-states and scan-outputs are listed together as one output parameter. The attribute num_scan_inputs indicates the number M of scan-inputs.
The behavior of
- Scan <
num_scan_inputs = m, body = loop-body, scan_input_axes = [axis_1, …, axis_m]
> (init_1, …, init_n, scan_1, …, scan_m)
is equivalent to the following pseudo-code:
// scan_i.shape[axis_i] denotes the (max) sequence-length of scan_i // scan_i.shape[axis_i] is required to be equal to scan_j.shape[axis_j] for all i,j. sequence_length = scan_1.shape[axis_1];
// initialize state-variables st_1 = init_1; … st_n = init_n; // initialize scan-output variables: [] denotes an empty tensor scan_out_1 = []; …; scan_out_k = []; // identify number of iterations:
// execute loop for (int t = 0; t < sequence_length; ++t) {
// generate the scan-input elements: the notation T<axis=k>[t] indicates the sub-tensor // of rank one less than T obtained by indexing T at position t along axis k. si_1 = scan_1<axis=axis_1>[t]; … ; si_m = scan_m<axis=axis_m>[t]; // execute loop-body st_1, …, st_n, so_1, …, so_k = loop-body(st_1, …, st_n, si_1, …, si_m) // accumulate the scan-output elements scan_out_1 = Concat<axis=0>(scan_out_1, so_1); … ; scan_out_k = Concat<axis=0>(scan_out_k, so_k);
}
return st_1, …, st_n, scan_out_1, …, scan_out_k;
Sample usage: Encoding RNN using a Scan
The following example shows how a simple RNN over an input tensor %X, with weight tensor %Wi, recurrence weight tensor %Ri, bias tensors %Wbi and %Rbi, and initial hidden-state %H_0 can be encoded as a ScanLoop. Note that the loop-body is a nested graph, and it directly computes %Wi, %Ri, %Wbi, and %Rbi (typically constants or initializers in the body graph). If these values are computed in the outer graph, they need to be passed in as extra state_variables.
- graph rnn-encoding {
%H_0 = … %X = … %Y_h, %Y = Scan[body = <graph rnn-cell-1>, num_scan_inputs=1](%H_0, %X) return %Y, %Y_h
}
- graph rnn-cell-1 (
%H_tminus1[FLOAT, tensor] %X_t[FLOAT, tensor]
- ) {
%Wi = … %Ri = … %Wbi = … %Rbi = … %t1 = X_t * (Wi^T) %t2 = H_tminus1*(Ri^T) %t3 = Add(%t1, %t2) %t4 = Add(%t3, %Wbi) %t5 = Add(%t4, %Rbi) %Ht = Tanh(%t5) %Accumulate = Identity(%Ht) return %Ht, %Accumulate
}
Attributes
body (required): The graph run each iteration. It has N+M inputs: (loop state variables…, scan_input_elts…). It has N+K outputs: (loop state variables…, scan_output_elts…). Each scan_output is created by concatenating the value of the specified scan_output_elt value at the end of each iteration of the loop. It is an error if the dimensions of these values change across loop iterations.
num_scan_inputs (required): An attribute specifying the number of scan_inputs M.
scan_input_axes: An optional list of M flags. The i-th element of the list specifies the axis to be scanned (the sequence axis) for the i-th scan_input. If omitted, 0 will be used as the scan axis for every scan_input.
scan_input_directions: An optional list of M flags. The i-th element of the list specifies the direction to be scanned for the i-th scan_input tensor: 0 indicates forward direction and 1 indicates reverse direction. If omitted, all scan_input tensors will be scanned in the forward direction.
scan_output_axes: An optional list of K flags. The i-th element of the list specifies the axis for the i-th scan_output. The scan outputs are accumulated along the specified axis. If omitted, 0 will be used as the scan axis for every scan_output.
scan_output_directions: An optional list of K flags, one for each scan_output. The i-th element of the list specifies whether the i-th scan_output should be constructed by appending or prepending a new value in each iteration: 0 indicates appending and 1 indicates prepending. If omitted, all scan_output tensors will be produced by appending a value in each iteration.
Inputs
Between 1 and 2147483647 inputs.
initial_state_and_scan_inputs (variadic) - V: Initial values of the loop’s N state variables followed by M scan_inputs
Outputs
Between 1 and 2147483647 outputs.
final_state_and_scan_outputs (variadic) - V: Final values of the loop’s N state variables followed by K scan_outputs
Type Constraints
V in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): All Tensor types
OnnxScatter#
- class mlprodict.npy.xop_auto_import_.OnnxScatter(*args, **kwargs)#
Version
name: Scatter (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been deprecated since version 11.
Summary
This operator is deprecated. Please use ScatterElements, which provides the same functionality.
Scatter takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself.
For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation.
Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatterElements#
- class mlprodict.npy.xop_auto_import_.OnnxScatterElements(*args, **kwargs)#
Version
name: ScatterElements (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
ScatterElements takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself.
reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
When reduction is set to some reduction function f, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] += f(output[indices[i][j]][j], updates[i][j]) if axis = 0, output[i][indices[i][j]] += f(output[i][indices[i][j]], updates[i][j]) if axis = 1,
where the f is +/*/max/min as specified.
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation.
(Opset 18 change): Adds max/min to the set of allowed reduction ops.
Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.reduction: Type of reduction to apply: none (default), add, mul, max, min. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the multiplication operation.’max’: reduction using the maximum operation.’min’: reduction using the minimum operation. Default value is
'none'.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatterElements_11#
- class mlprodict.npy.xop_auto_import_.OnnxScatterElements_11(*args, **kwargs)#
Version
name: ScatterElements (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
ScatterElements takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself.
For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation.
Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatterElements_13#
- class mlprodict.npy.xop_auto_import_.OnnxScatterElements_13(*args, **kwargs)#
Version
name: ScatterElements (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
ScatterElements takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself.
For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation.
Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatterElements_16#
- class mlprodict.npy.xop_auto_import_.OnnxScatterElements_16(*args, **kwargs)#
Version
name: ScatterElements (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
ScatterElements takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data. For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself. reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
When reduction is set to “add”, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] += updates[i][j] if axis = 0, output[i][indices[i][j]] += updates[i][j] if axis = 1,
When reduction is set to “mul”, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] *= updates[i][j] if axis = 0, output[i][indices[i][j]] *= updates[i][j] if axis = 1,
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation. Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.reduction: Type of reduction to apply: none (default), add, mul. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the multiplication operation. Default value is
'none'.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatterElements_18#
- class mlprodict.npy.xop_auto_import_.OnnxScatterElements_18(*args, **kwargs)#
Version
name: ScatterElements (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
ScatterElements takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself.
reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
When reduction is set to some reduction function f, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] += f(output[indices[i][j]][j], updates[i][j]) if axis = 0, output[i][indices[i][j]] += f(output[i][indices[i][j]], updates[i][j]) if axis = 1,
where the f is +/*/max/min as specified.
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation.
(Opset 18 change): Adds max/min to the set of allowed reduction ops.
Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.reduction: Type of reduction to apply: none (default), add, mul, max, min. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the multiplication operation.’max’: reduction using the maximum operation.’min’: reduction using the minimum operation. Default value is
'none'.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatterND#
- class mlprodict.npy.xop_auto_import_.OnnxScatterND(*args, **kwargs)#
Version
name: ScatterND (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates tensor of rank q + r - indices.shape[-1] - 1. The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
- indices is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of indices.
indices is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into data.
Hence, k can be a value at most the rank of data. When k equals rank(data), each update entry specifies an update to a single element of the tensor. When k is less than rank(data) each update entry specifies an update to a slice of the tensor. Index values are allowed to be negative, as per the usual convention for counting backwards from the end, but are expected in the valid range.
updates is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. The remaining dimensions of updates correspond to the dimensions of the replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, corresponding to the trailing (r-k) dimensions of data. Thus, the shape of updates must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation of shapes.
The output is calculated via the following equation:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = updates[idx]
The order of iteration in the above loop is not specified. In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order.
reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. When reduction is set to some reduction function f, output is calculated as follows:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = f(output[indices[idx]], updates[idx])
where the f is +/*/max/min as specified.
This operator is the inverse of GatherND.
(Opset 18 change): Adds max/min to the set of allowed reduction ops.
Example 1:
data = [1, 2, 3, 4, 5, 6, 7, 8] indices = [[4], [3], [1], [7]] updates = [9, 10, 11, 12] output = [1, 11, 3, 10, 9, 6, 7, 12]
Example 2:
data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] indices = [[0], [2]] updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]
Attributes
reduction: Type of reduction to apply: none (default), add, mul, max, min. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the addition operation. ‘max’: reduction using the maximum operation.’min’: reduction using the minimum operation. Default value is
'none'.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1.
updates (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxScatterND_11#
- class mlprodict.npy.xop_auto_import_.OnnxScatterND_11(*args, **kwargs)#
Version
name: ScatterND (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates tensor of rank q + r - indices.shape[-1] - 1. The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data. Note that indices should not have duplicate entries. That is, two or more updates for the same index-location is not supported.
- indices is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of indices.
indices is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into data.
Hence, k can be a value at most the rank of data. When k equals rank(data), each update entry specifies an update to a single element of the tensor. When k is less than rank(data) each update entry specifies an update to a slice of the tensor. Index values are allowed to be negative, as per the usual convention for counting backwards from the end, but are expected in the valid range.
updates is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. The remaining dimensions of updates correspond to the dimensions of the replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, corresponding to the trailing (r-k) dimensions of data. Thus, the shape of updates must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation of shapes.
The output is calculated via the following equation:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = updates[idx]
The order of iteration in the above loop is not specified. In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order.
This operator is the inverse of GatherND.
Example 1:
data = [1, 2, 3, 4, 5, 6, 7, 8] indices = [[4], [3], [1], [7]] updates = [9, 10, 11, 12] output = [1, 11, 3, 10, 9, 6, 7, 12]
Example 2:
data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] indices = [[0], [2]] updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1.
updates (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxScatterND_13#
- class mlprodict.npy.xop_auto_import_.OnnxScatterND_13(*args, **kwargs)#
Version
name: ScatterND (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates tensor of rank q + r - indices.shape[-1] - 1. The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data. Note that indices should not have duplicate entries. That is, two or more updates for the same index-location is not supported.
- indices is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of indices.
indices is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into data.
Hence, k can be a value at most the rank of data. When k equals rank(data), each update entry specifies an update to a single element of the tensor. When k is less than rank(data) each update entry specifies an update to a slice of the tensor. Index values are allowed to be negative, as per the usual convention for counting backwards from the end, but are expected in the valid range.
updates is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. The remaining dimensions of updates correspond to the dimensions of the replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, corresponding to the trailing (r-k) dimensions of data. Thus, the shape of updates must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation of shapes.
The output is calculated via the following equation:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = updates[idx]
The order of iteration in the above loop is not specified. In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order.
This operator is the inverse of GatherND.
Example 1:
data = [1, 2, 3, 4, 5, 6, 7, 8] indices = [[4], [3], [1], [7]] updates = [9, 10, 11, 12] output = [1, 11, 3, 10, 9, 6, 7, 12]
Example 2:
data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] indices = [[0], [2]] updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1.
updates (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxScatterND_16#
- class mlprodict.npy.xop_auto_import_.OnnxScatterND_16(*args, **kwargs)#
Version
name: ScatterND (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates tensor of rank q + r - indices.shape[-1] - 1. The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
- indices is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of indices.
indices is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into data.
Hence, k can be a value at most the rank of data. When k equals rank(data), each update entry specifies an update to a single element of the tensor. When k is less than rank(data) each update entry specifies an update to a slice of the tensor. Index values are allowed to be negative, as per the usual convention for counting backwards from the end, but are expected in the valid range.
updates is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. The remaining dimensions of updates correspond to the dimensions of the replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, corresponding to the trailing (r-k) dimensions of data. Thus, the shape of updates must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation of shapes.
- The output is calculated via the following equation:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = updates[idx]
The order of iteration in the above loop is not specified. In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order.
reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. When reduction is set to “add”, output is calculated as follows:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] += updates[idx]
- When reduction is set to “mul”, output is calculated as follows:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] *= updates[idx]
This operator is the inverse of GatherND. Example 1:
data = [1, 2, 3, 4, 5, 6, 7, 8] indices = [[4], [3], [1], [7]] updates = [9, 10, 11, 12] output = [1, 11, 3, 10, 9, 6, 7, 12]
Example 2:
data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] indices = [[0], [2]] updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]
Attributes
reduction: Type of reduction to apply: none (default), add, mul. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the multiplication operation. Default value is
'none'.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1.
updates (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxScatterND_18#
- class mlprodict.npy.xop_auto_import_.OnnxScatterND_18(*args, **kwargs)#
Version
name: ScatterND (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
ScatterND takes three inputs data tensor of rank r >= 1, indices tensor of rank q >= 1, and updates tensor of rank q + r - indices.shape[-1] - 1. The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
- indices is an integer tensor. Let k denote indices.shape[-1], the last dimension in the shape of indices.
indices is treated as a (q-1)-dimensional tensor of k-tuples, where each k-tuple is a partial-index into data.
Hence, k can be a value at most the rank of data. When k equals rank(data), each update entry specifies an update to a single element of the tensor. When k is less than rank(data) each update entry specifies an update to a slice of the tensor. Index values are allowed to be negative, as per the usual convention for counting backwards from the end, but are expected in the valid range.
updates is treated as a (q-1)-dimensional tensor of replacement-slice-values. Thus, the first (q-1) dimensions of updates.shape must match the first (q-1) dimensions of indices.shape. The remaining dimensions of updates correspond to the dimensions of the replacement-slice-values. Each replacement-slice-value is a (r-k) dimensional tensor, corresponding to the trailing (r-k) dimensions of data. Thus, the shape of updates must equal indices.shape[0:q-1] ++ data.shape[k:r-1], where ++ denotes the concatenation of shapes.
The output is calculated via the following equation:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = updates[idx]
The order of iteration in the above loop is not specified. In particular, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order.
reduction allows specification of an optional reduction operation, which is applied to all values in updates tensor into output at the specified indices. In cases where reduction is set to “none”, indices should not have duplicate entries: that is, if idx1 != idx2, then indices[idx1] != indices[idx2]. This ensures that the output value does not depend on the iteration order. When reduction is set to some reduction function f, output is calculated as follows:
output = np.copy(data) update_indices = indices.shape[:-1] for idx in np.ndindex(update_indices):
output[indices[idx]] = f(output[indices[idx]], updates[idx])
where the f is +/*/max/min as specified.
This operator is the inverse of GatherND.
(Opset 18 change): Adds max/min to the set of allowed reduction ops.
Example 1:
data = [1, 2, 3, 4, 5, 6, 7, 8] indices = [[4], [3], [1], [7]] updates = [9, 10, 11, 12] output = [1, 11, 3, 10, 9, 6, 7, 12]
Example 2:
data = [[[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]] indices = [[0], [2]] updates = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]]] output = [[[5, 5, 5, 5], [6, 6, 6, 6], [7, 7, 7, 7], [8, 8, 8, 8]], [[1, 2, 3, 4], [5, 6, 7, 8], [8, 7, 6, 5], [4, 3, 2, 1]], [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], [[8, 7, 6, 5], [4, 3, 2, 1], [1, 2, 3, 4], [5, 6, 7, 8]]]
Attributes
reduction: Type of reduction to apply: none (default), add, mul, max, min. ‘none’: no reduction applied. ‘add’: reduction using the addition operation. ‘mul’: reduction using the addition operation. ‘max’: reduction using the maximum operation.’min’: reduction using the minimum operation. Default value is
'none'.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - tensor(int64): Tensor of rank q >= 1.
updates (heterogeneous) - T: Tensor of rank q + r - indices_shape[-1] - 1.
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to any tensor type.
OnnxScatter_11#
- class mlprodict.npy.xop_auto_import_.OnnxScatter_11(*args, **kwargs)#
Version
name: Scatter (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been deprecated since version 11.
Summary
This operator is deprecated. Please use ScatterElements, which provides the same functionality.
Scatter takes three inputs data, updates, and indices of the same rank r >= 1 and an optional attribute axis that identifies an axis of data (by default, the outer-most axis, that is axis 0). The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data.
For each entry in updates, the target index in data is obtained by combining the corresponding entry in indices with the index of the entry itself: the index-value for dimension = axis is obtained from the value of the corresponding entry in indices and the index-value for dimension != axis is obtained from the index of the entry itself.
For instance, in a 2-D tensor case, the update corresponding to the [i][j] entry is performed as below:
output[indices[i][j]][j] = updates[i][j] if axis = 0, output[i][indices[i][j]] = updates[i][j] if axis = 1,
This operator is the inverse of GatherElements. It is similar to Torch’s Scatter operation.
Example 1:
data = [ [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0], ] indices = [ [1, 0, 2], [0, 2, 1], ] updates = [ [1.0, 1.1, 1.2], [2.0, 2.1, 2.2], ] output = [ [2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2] ]
Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input). All index values are expected to be within bounds [-s, s-1] along axis of size s. It is an error if any of the index values are out of bounds.
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxScatter_9#
- class mlprodict.npy.xop_auto_import_.OnnxScatter_9(*args, **kwargs)#
Version
name: Scatter (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Given data, updates and indices input tensors of rank r >= 1, write the values provided by updates into the first input, data, along axis dimension of data (by default outer-most one as axis=0) at corresponding indices. For each entry in updates, the target index in data is specified by corresponding entry in indices for dimension = axis, and index in source for dimension != axis. For instance, in a 2-D tensor case, data[indices[i][j]][j] = updates[i][j] if axis = 0, or data[i][indices[i][j]] = updates[i][j] if axis = 1, where i and j are loop counters from 0 up to the respective size in updates - 1. Example 1:
- data = [
[0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 0.0],
] indices = [
[1, 0, 2], [0, 2, 1],
] updates = [
[1.0, 1.1, 1.2], [2.0, 2.1, 2.2],
] output = [
[2.0, 1.1, 0.0] [1.0, 0.0, 2.2] [0.0, 2.1, 1.2]
]
- Example 2:
data = [[1.0, 2.0, 3.0, 4.0, 5.0]] indices = [[1, 3]] updates = [[1.1, 2.1]] axis = 1 output = [[1.0, 1.1, 3.0, 2.1, 5.0]]
Attributes
axis: Which axis to scatter on. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] Default value is
0.
Inputs
data (heterogeneous) - T: Tensor of rank r >= 1.
indices (heterogeneous) - Tind: Tensor of int32/int64 indices, of r >= 1 (same rank as input).
updates (heterogeneous) - T: Tensor of rank r >=1 (same rank and shape as indices)
Outputs
output (heterogeneous) - T: Tensor of rank r >= 1 (same rank as input).
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input and output types can be of any tensor type.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxSelu#
- class mlprodict.npy.xop_auto_import_.OnnxSelu(*args, **kwargs)#
Version
name: Selu (GitHub)
domain: main
since_version: 6
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Selu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the scaled exponential linear unit function, y = gamma * (alpha * e^x - alpha) for x <= 0, y = gamma * x for x > 0, is applied to the tensor elementwise.
Attributes
alpha: Coefficient of SELU default to 1.67326319217681884765625 (i.e., float32 approximation of 1.6732632423543772848170429916717). Default value is
1.6732631921768188.gamma: Coefficient of SELU default to 1.05070102214813232421875 (i.e., float32 approximation of 1.0507009873554804934193349852946). Default value is
1.0507010221481323.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSelu_1#
- class mlprodict.npy.xop_auto_import_.OnnxSelu_1(*args, **kwargs)#
Version
name: Selu (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Selu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the scaled exponential linear unit function, y = gamma * (alpha * e^x - alpha) for x <= 0, y = gamma * x for x > 0, is applied to the tensor elementwise.
Attributes
alpha: Coefficient of SELU default to 1.6732. Default value is
1.673200011253357.consumed_inputs: legacy optimization attribute.
gamma: Coefficient of SELU default to 1.0507. Default value is
1.0506999492645264.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSelu_6#
- class mlprodict.npy.xop_auto_import_.OnnxSelu_6(*args, **kwargs)#
Version
name: Selu (GitHub)
domain: main
since_version: 6
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Selu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the scaled exponential linear unit function, y = gamma * (alpha * e^x - alpha) for x <= 0, y = gamma * x for x > 0, is applied to the tensor elementwise.
Attributes
alpha: Coefficient of SELU default to 1.67326319217681884765625 (i.e., float32 approximation of 1.6732632423543772848170429916717). Default value is
1.6732631921768188.gamma: Coefficient of SELU default to 1.05070102214813232421875 (i.e., float32 approximation of 1.0507009873554804934193349852946). Default value is
1.0507010221481323.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSequenceAt#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceAt(*args, **kwargs)#
Version
name: SequenceAt (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Outputs a tensor copy from the tensor at ‘position’ in ‘input_sequence’. Accepted range for ‘position’ is in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. Negative value means counting positions from the back.
Inputs
input_sequence (heterogeneous) - S: Input sequence.
position (heterogeneous) - I: Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
Outputs
tensor (heterogeneous) - T: Output tensor at the specified position in the input sequence.
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
I in ( tensor(int32), tensor(int64) ): Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceAt_11#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceAt_11(*args, **kwargs)#
Version
name: SequenceAt (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Outputs a tensor copy from the tensor at ‘position’ in ‘input_sequence’. Accepted range for ‘position’ is in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. Negative value means counting positions from the back.
Inputs
input_sequence (heterogeneous) - S: Input sequence.
position (heterogeneous) - I: Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
Outputs
tensor (heterogeneous) - T: Output tensor at the specified position in the input sequence.
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
I in ( tensor(int32), tensor(int64) ): Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceConstruct#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceConstruct(*args, **kwargs)#
Version
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Construct a tensor sequence containing ‘inputs’ tensors. All tensors in ‘inputs’ must have the same data type.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: Tensors.
Outputs
output_sequence (heterogeneous) - S: Sequence enclosing the input tensors.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to any tensor type.
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain output types to any tensor type.
OnnxSequenceConstruct_11#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceConstruct_11(*args, **kwargs)#
Version
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Construct a tensor sequence containing ‘inputs’ tensors. All tensors in ‘inputs’ must have the same data type.
Inputs
Between 1 and 2147483647 inputs.
inputs (variadic, heterogeneous) - T: Tensors.
Outputs
output_sequence (heterogeneous) - S: Sequence enclosing the input tensors.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to any tensor type.
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain output types to any tensor type.
OnnxSequenceEmpty#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceEmpty(*args, **kwargs)#
Version
name: SequenceEmpty (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Construct an empty tensor sequence, with given data type.
Attributes
dtype: (Optional) The data type of the tensors in the output sequence. The default type is ‘float’.
Outputs
output (heterogeneous) - S: Empty sequence.
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain output types to any tensor type.
OnnxSequenceEmpty_11#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceEmpty_11(*args, **kwargs)#
Version
name: SequenceEmpty (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Construct an empty tensor sequence, with given data type.
Attributes
dtype: (Optional) The data type of the tensors in the output sequence. The default type is ‘float’.
Outputs
output (heterogeneous) - S: Empty sequence.
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain output types to any tensor type.
OnnxSequenceErase#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceErase(*args, **kwargs)#
Version
name: SequenceErase (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Outputs a tensor sequence that removes the tensor at ‘position’ from ‘input_sequence’. Accepted range for ‘position’ is in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. Negative value means counting positions from the back. ‘position’ is optional, by default it erases the last tensor from ‘input_sequence’.
Inputs
Between 1 and 2 inputs.
input_sequence (heterogeneous) - S: Input sequence.
position (optional, heterogeneous) - I: Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
Outputs
output_sequence (heterogeneous) - S: Output sequence that has the tensor at the specified position removed.
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
I in ( tensor(int32), tensor(int64) ): Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceErase_11#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceErase_11(*args, **kwargs)#
Version
name: SequenceErase (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Outputs a tensor sequence that removes the tensor at ‘position’ from ‘input_sequence’. Accepted range for ‘position’ is in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. Negative value means counting positions from the back. ‘position’ is optional, by default it erases the last tensor from ‘input_sequence’.
Inputs
Between 1 and 2 inputs.
input_sequence (heterogeneous) - S: Input sequence.
position (optional, heterogeneous) - I: Position of the tensor in the sequence. Negative value means counting positions from the back. Accepted range in [-n, n - 1], where n is the number of tensors in ‘input_sequence’. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
Outputs
output_sequence (heterogeneous) - S: Output sequence that has the tensor at the specified position removed.
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
I in ( tensor(int32), tensor(int64) ): Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceInsert#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceInsert(*args, **kwargs)#
Version
name: SequenceInsert (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Outputs a tensor sequence that inserts ‘tensor’ into ‘input_sequence’ at ‘position’. ‘tensor’ must have the same data type as ‘input_sequence’. Accepted range for ‘position’ is in [-n, n], where n is the number of tensors in ‘input_sequence’. Negative value means counting positions from the back. ‘position’ is optional, by default it inserts ‘tensor’ to the back of ‘input_sequence’.
Inputs
Between 2 and 3 inputs.
input_sequence (heterogeneous) - S: Input sequence.
tensor (heterogeneous) - T: Input tensor to be inserted into the input sequence.
position (optional, heterogeneous) - I: Position in the sequence where the new tensor is inserted. It is optional and default is to insert to the back of the sequence. Negative value means counting positions from the back. Accepted range in [-n, n], where n is the number of tensors in ‘input_sequence’. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
Outputs
output_sequence (heterogeneous) - S: Output sequence that contains the inserted tensor at given position.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
I in ( tensor(int32), tensor(int64) ): Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceInsert_11#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceInsert_11(*args, **kwargs)#
Version
name: SequenceInsert (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Outputs a tensor sequence that inserts ‘tensor’ into ‘input_sequence’ at ‘position’. ‘tensor’ must have the same data type as ‘input_sequence’. Accepted range for ‘position’ is in [-n, n], where n is the number of tensors in ‘input_sequence’. Negative value means counting positions from the back. ‘position’ is optional, by default it inserts ‘tensor’ to the back of ‘input_sequence’.
Inputs
Between 2 and 3 inputs.
input_sequence (heterogeneous) - S: Input sequence.
tensor (heterogeneous) - T: Input tensor to be inserted into the input sequence.
position (optional, heterogeneous) - I: Position in the sequence where the new tensor is inserted. It is optional and default is to insert to the back of the sequence. Negative value means counting positions from the back. Accepted range in [-n, n], where n is the number of tensors in ‘input_sequence’. It is an error if any of the index values are out of bounds. It must be a scalar(tensor of empty shape).
Outputs
output_sequence (heterogeneous) - S: Output sequence that contains the inserted tensor at given position.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor type.
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
I in ( tensor(int32), tensor(int64) ): Constrain position to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceLength#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceLength(*args, **kwargs)#
Version
name: SequenceLength (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Produces a scalar(tensor of empty shape) containing the number of tensors in ‘input_sequence’.
Inputs
input_sequence (heterogeneous) - S: Input sequence.
Outputs
length (heterogeneous) - I: Length of input sequence. It must be a scalar(tensor of empty shape).
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
I in ( tensor(int64) ): Constrain output to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceLength_11#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceLength_11(*args, **kwargs)#
Version
name: SequenceLength (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Produces a scalar(tensor of empty shape) containing the number of tensors in ‘input_sequence’.
Inputs
input_sequence (heterogeneous) - S: Input sequence.
Outputs
length (heterogeneous) - I: Length of input sequence. It must be a scalar(tensor of empty shape).
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain to any tensor type.
I in ( tensor(int64) ): Constrain output to integral tensor. It must be a scalar(tensor of empty shape).
OnnxSequenceMap#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceMap(*args, **kwargs)#
Version
name: SequenceMap (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Applies a sub-graph to each sample in the input sequence(s).
Inputs can be either tensors or sequences, with the exception of the first input which must be a sequence. The length of the first input sequence will determine the number of samples in the outputs. Any other sequence inputs should have the same number of samples. The number of inputs and outputs, should match the one of the subgraph.
For each i-th element in the output, a sample will be extracted from the input sequence(s) at the i-th position and the sub-graph will be applied to it. The outputs will contain the outputs of the sub-graph for each sample, in the same order as in the input.
This operator assumes that processing each sample is independent and could executed in parallel or in any order. Users cannot expect any specific ordering in which each subgraph is computed.
Attributes
body (required): The graph to be run for each sample in the sequence(s). It should have as many inputs and outputs as inputs and outputs to the SequenceMap function.
Inputs
Between 1 and 2147483647 inputs.
input_sequence (heterogeneous) - S: Input sequence.
additional_inputs (variadic) - V: Additional inputs to the graph
Outputs
Between 1 and 2147483647 outputs.
out_sequence (variadic) - S: Output sequence(s)
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain input types to any sequence type.
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor or sequence type.
OnnxSequenceMap_17#
- class mlprodict.npy.xop_auto_import_.OnnxSequenceMap_17(*args, **kwargs)#
Version
name: SequenceMap (GitHub)
domain: main
since_version: 17
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 17.
Summary
Applies a sub-graph to each sample in the input sequence(s).
Inputs can be either tensors or sequences, with the exception of the first input which must be a sequence. The length of the first input sequence will determine the number of samples in the outputs. Any other sequence inputs should have the same number of samples. The number of inputs and outputs, should match the one of the subgraph.
For each i-th element in the output, a sample will be extracted from the input sequence(s) at the i-th position and the sub-graph will be applied to it. The outputs will contain the outputs of the sub-graph for each sample, in the same order as in the input.
This operator assumes that processing each sample is independent and could executed in parallel or in any order. Users cannot expect any specific ordering in which each subgraph is computed.
Attributes
body (required): The graph to be run for each sample in the sequence(s). It should have as many inputs and outputs as inputs and outputs to the SequenceMap function.
Inputs
Between 1 and 2147483647 inputs.
input_sequence (heterogeneous) - S: Input sequence.
additional_inputs (variadic) - V: Additional inputs to the graph
Outputs
Between 1 and 2147483647 outputs.
out_sequence (variadic) - S: Output sequence(s)
Type Constraints
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain input types to any sequence type.
V in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain to any tensor or sequence type.
OnnxShape#
- class mlprodict.npy.xop_auto_import_.OnnxShape(*args, **kwargs)#
Version
name: Shape (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. Optional attributes start and end can be used to compute a slice of the input tensor’s shape. If start axis is omitted, the slice starts from axis 0. The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). If the end axis is omitted, the axes upto the last one will be included. Negative axes indicate counting back from the last axis. Note that axes will be clamped to the range [0, r-1], where r is the rank of the input tensor if they are out-of-range (after adding r in the case of negative axis). Thus, specifying any end value > r is equivalent to specifying an end value of r, and specifying any start value < -r is equivalent to specifying a start value of 0.
For example: Input tensor with shape: [2, 3, 4] No attributes specified. Output: [2, 3, 4]
Input tensor with shape: [2, 3, 4] start: -1 Output: [4]
Input tensor with shape: [2, 3, 4] end: -1 Output: [2, 3]
Input tensor with shape: [2, 3, 4] start: 1 end: 2 Output: [3]
Attributes
end: (Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
start: (Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back. Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
shape (heterogeneous) - T1: Shape of the input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor.
OnnxShape_1#
- class mlprodict.npy.xop_auto_import_.OnnxShape_1(*args, **kwargs)#
Version
name: Shape (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
shape (heterogeneous) - T1: Shape of the input tensor
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor.
OnnxShape_13#
- class mlprodict.npy.xop_auto_import_.OnnxShape_13(*args, **kwargs)#
Version
name: Shape (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
shape (heterogeneous) - T1: Shape of the input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor.
OnnxShape_15#
- class mlprodict.npy.xop_auto_import_.OnnxShape_15(*args, **kwargs)#
Version
name: Shape (GitHub)
domain: main
since_version: 15
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 15.
Summary
Takes a tensor as input and outputs an 1D int64 tensor containing the shape of the input tensor. Optional attributes start and end can be used to compute a slice of the input tensor’s shape. If start axis is omitted, the slice starts from axis 0. The end axis, if specified, is exclusive (and the returned value will not include the size of that axis). If the end axis is omitted, the axes upto the last one will be included. Negative axes indicate counting back from the last axis. Note that axes will be clamped to the range [0, r-1], where r is the rank of the input tensor if they are out-of-range (after adding r in the case of negative axis). Thus, specifying any end value > r is equivalent to specifying an end value of r, and specifying any start value < -r is equivalent to specifying a start value of 0.
For example: Input tensor with shape: [2, 3, 4] No attributes specified. Output: [2, 3, 4]
Input tensor with shape: [2, 3, 4] start: -1 Output: [4]
Input tensor with shape: [2, 3, 4] end: -1 Output: [2, 3]
Input tensor with shape: [2, 3, 4] start: 1 end: 2 Output: [3]
Attributes
end: (Optional) Ending axis for slicing the shape. Negative value means counting dimensions from the back. If omitted, sizes of all axes upto (including) the last one will be included.
start: (Optional) Starting axis for slicing the shape. Default value is 0.Negative value means counting dimensions from the back. Default value is
0.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
shape (heterogeneous) - T1: Shape of the input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor.
OnnxShrink#
- class mlprodict.npy.xop_auto_import_.OnnxShrink(*args, **kwargs)#
Version
name: Shrink (GitHub)
domain: main
since_version: 9
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Shrink takes one input data (Tensor<numeric>) and produces one Tensor output, having same datatype and shape with input. It has two attributes, lambd and bias. The formula of this operator is: If x < -lambd, y = x + bias; If x > lambd, y = x - bias; Otherwise, y = 0.
Attributes
bias: The bias value added to output. Default is 0. Default value is
0.0.lambd: The lambd value for the Shrink formulation. Default is 0.5. Default value is
0.5.
Inputs
input (heterogeneous) - T: The input data as Tensor.
Outputs
output (heterogeneous) - T: The output.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
OnnxShrink_9#
- class mlprodict.npy.xop_auto_import_.OnnxShrink_9(*args, **kwargs)#
Version
name: Shrink (GitHub)
domain: main
since_version: 9
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Shrink takes one input data (Tensor<numeric>) and produces one Tensor output, having same datatype and shape with input. It has two attributes, lambd and bias. The formula of this operator is: If x < -lambd, y = x + bias; If x > lambd, y = x - bias; Otherwise, y = 0.
Attributes
bias: The bias value added to output. Default is 0. Default value is
0.0.lambd: The lambd value for the Shrink formulation. Default is 0.5. Default value is
0.5.
Inputs
input (heterogeneous) - T: The input data as Tensor.
Outputs
output (heterogeneous) - T: The output.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input to only numeric types.
OnnxSigmoid#
- class mlprodict.npy.xop_auto_import_.OnnxSigmoid(*args, **kwargs)#
Version
name: Sigmoid (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Sigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSigmoid_1#
- class mlprodict.npy.xop_auto_import_.OnnxSigmoid_1(*args, **kwargs)#
Version
name: Sigmoid (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Sigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the tensor elementwise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSigmoid_13#
- class mlprodict.npy.xop_auto_import_.OnnxSigmoid_13(*args, **kwargs)#
Version
name: Sigmoid (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Sigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSigmoid_6#
- class mlprodict.npy.xop_auto_import_.OnnxSigmoid_6(*args, **kwargs)#
Version
name: Sigmoid (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Sigmoid takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the sigmoid function, y = 1 / (1 + exp(-x)), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSign#
- class mlprodict.npy.xop_auto_import_.OnnxSign(*args, **kwargs)#
Version
name: Sign (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculate the sign of the given input tensor element-wise. If input > 0, output 1. if input < 0, output -1. if input == 0, output 0.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The sign of the input tensor computed element-wise. It has the same shape and type of the input.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxSign_13#
- class mlprodict.npy.xop_auto_import_.OnnxSign_13(*args, **kwargs)#
Version
name: Sign (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculate the sign of the given input tensor element-wise. If input > 0, output 1. if input < 0, output -1. if input == 0, output 0.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The sign of the input tensor computed element-wise. It has the same shape and type of the input.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxSign_9#
- class mlprodict.npy.xop_auto_import_.OnnxSign_9(*args, **kwargs)#
Version
name: Sign (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculate the sign of the given input tensor element-wise. If input > 0, output 1. if input < 0, output -1. if input == 0, output 0.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The sign of the input tensor computed element-wise. It has the same shape and type of the input.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxSin#
- class mlprodict.npy.xop_auto_import_.OnnxSin(*args, **kwargs)#
Version
name: Sin (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the sine of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The sine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSin_7#
- class mlprodict.npy.xop_auto_import_.OnnxSin_7(*args, **kwargs)#
Version
name: Sin (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the sine of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The sine of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSinh#
- class mlprodict.npy.xop_auto_import_.OnnxSinh(*args, **kwargs)#
Version
name: Sinh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic sine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic sine values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSinh_9#
- class mlprodict.npy.xop_auto_import_.OnnxSinh_9(*args, **kwargs)#
Version
name: Sinh (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Calculates the hyperbolic sine of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic sine values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSize#
- class mlprodict.npy.xop_auto_import_.OnnxSize(*args, **kwargs)#
Version
name: Size (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
size (heterogeneous) - T1: Total number of elements of the input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor, which should be a scalar though.
OnnxSize_1#
- class mlprodict.npy.xop_auto_import_.OnnxSize_1(*args, **kwargs)#
Version
name: Size (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
size (heterogeneous) - T1: Total number of elements of the input tensor
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor, which should be a scalar though.
OnnxSize_13#
- class mlprodict.npy.xop_auto_import_.OnnxSize_13(*args, **kwargs)#
Version
name: Size (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Takes a tensor as input and outputs a int64 scalar that equals to the total number of elements of the input tensor.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
size (heterogeneous) - T1: Total number of elements of the input tensor
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input tensor can be of arbitrary type.
T1 in ( tensor(int64) ): Constrain output to int64 tensor, which should be a scalar though.
OnnxSlice#
- class mlprodict.npy.xop_auto_import_.OnnxSlice(*args, **kwargs)#
Version
name: Slice (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Produces a slice of the input tensor along multiple axes. Similar to numpy: https://numpy.org/doc/stable/user/basics.indexing.html?highlight=slice#slicing-and-striding
Slice uses the starts, ends, axes and steps inputs to select a sub-tensor of its input data tensor.
An effective start[i], end[i], and step[i] must be computed for each i in [0, … r-1] where r = rank(input) as follows:
If axes are omitted, they are set to [0, …, r-1]. If steps are omitted, they are set to [1, …, 1] of length len(starts)
The effective values are initialized as start[i] = 0, end[i] = dims[i] where dims are the dimensions of input and `step[i] = `1.
All negative elements of axes are made non-negatve by adding r to them, where r =rank(input).
All negative values in starts[i] and ends[i] have dims[axes[i]] added to them, where dims are the dimensions of input. Then start[axes[i]] is the adjusted starts[i] is clamped into the range [0, dims[axes[i]]] for positive stepping and [0, dims[axes[i]]-1] for negative stepping.
The clamping for the adjusted ends[i] depends on the sign of steps[i] and must accommodate copying 0 through dims[axes[i]] elements, so for positive stepping end[axes[i]] is clamped to [0, dims[axes[i]]], while for negative stepping it is clamped to [-1, dims[axes[i]]-1].
Finally, step[axes[i]] = steps[i].
For slicing to the end of a dimension with unknown size, it is recommended to pass in INT_MAX when slicing forward and ‘INT_MIN’ when slicing backward.
- Example 1:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] axes = [0, 1] starts = [1, 0] ends = [2, 3] steps = [1, 2] result = [
[5, 7],
]
- Example 2:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] starts = [0, 1] ends = [-1, 1000] result = [
[2, 3, 4],
]
Inputs
Between 3 and 5 inputs.
data (heterogeneous) - T: Tensor of data to extract slices from.
starts (heterogeneous) - Tind: 1-D tensor of starting indices of corresponding axis in axes
ends (heterogeneous) - Tind: 1-D tensor of ending indices (exclusive) of corresponding axis in axes
axes (optional, heterogeneous) - Tind: 1-D tensor of axes that starts and ends apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated.
steps (optional, heterogeneous) - Tind: 1-D tensor of slice step of corresponding axis in axes. Negative value means slicing backward. ‘steps’ cannot be 0. Defaults to 1s.
Outputs
output (heterogeneous) - T: Sliced data tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxSlice_1#
- class mlprodict.npy.xop_auto_import_.OnnxSlice_1(*args, **kwargs)#
Version
name: Slice (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Produces a slice of the input tensor along multiple axes. Similar to numpy: https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html Slices uses axes, starts and ends attributes to specify the start and end dimension for each axis in the list of axes, it uses this information to slice the input data tensor. If a negative value is passed for any of the start or end indices, it represent number of elements before the end of that dimension. If the value passed to start or end is larger than the n (the number of elements in this dimension), it represents n. For slicing to the end of a dimension with unknown size, it is recommended to pass in INT_MAX. If axes are omitted, they are set to [0, …, ndim-1]. Example 1:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] axes = [0, 1] starts = [1, 0] ends = [2, 3] result = [
[5, 6, 7],
]
- Example 2:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] starts = [0, 1] ends = [-1, 1000] result = [
[2, 3, 4],
]
Attributes
axes: Axes that starts and ends apply to. It’s optional. If not present, will be treated as [0, 1, …, len(starts) - 1].
ends (required): Ending indices (exclusive) of corresponding axis in axes`
starts (required): Starting indices of corresponding axis in axes
Inputs
data (heterogeneous) - T: Tensor of data to extract slices from.
Outputs
output (heterogeneous) - T: Sliced data tensor.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSlice_10#
- class mlprodict.npy.xop_auto_import_.OnnxSlice_10(*args, **kwargs)#
Version
name: Slice (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Produces a slice of the input tensor along multiple axes. Similar to numpy: https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html Slices uses starts, ends, axes and steps inputs to specify the start and end dimension and step for each axis in the list of axes, it uses this information to slice the input data tensor. If a negative value is passed for any of the start or end indices, it represent number of elements before the end of that dimension. If the value passed to start or end is larger than the n (the number of elements in this dimension), it represents n. For slicing to the end of a dimension with unknown size, it is recommended to pass in INT_MAX. If a negative value is passed for step, it represents slicing backward. If axes are omitted, they are set to [0, …, ndim-1]. If steps are omitted, they are set to [1, …, 1] of length len(starts) Example 1:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] axes = [0, 1] starts = [1, 0] ends = [2, 3] steps = [1, 2] result = [
[5, 7],
]
- Example 2:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] starts = [0, 1] ends = [-1, 1000] result = [
[2, 3, 4],
]
Inputs
Between 3 and 5 inputs.
data (heterogeneous) - T: Tensor of data to extract slices from.
starts (heterogeneous) - Tind: 1-D tensor of starting indices of corresponding axis in axes
ends (heterogeneous) - Tind: 1-D tensor of ending indices (exclusive) of corresponding axis in axes
axes (optional, heterogeneous) - Tind: 1-D tensor of axes that starts and ends apply to.
steps (optional, heterogeneous) - Tind: 1-D tensor of slice step of corresponding axis in axes. Default to 1.
Outputs
output (heterogeneous) - T: Sliced data tensor.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxSlice_11#
- class mlprodict.npy.xop_auto_import_.OnnxSlice_11(*args, **kwargs)#
Version
name: Slice (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Produces a slice of the input tensor along multiple axes. Similar to numpy: https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html Slices uses starts, ends, axes and steps inputs to specify the start and end dimension and step for each axis in the list of axes, it uses this information to slice the input data tensor. If a negative value is passed for any of the start or end indices, it represents number of elements before the end of that dimension. If the value passed to start or end is larger than the n (the number of elements in this dimension), it represents n. For slicing to the end of a dimension with unknown size, it is recommended to pass in INT_MAX when slicing forward and ‘INT_MIN’ when slicing backward. If a negative value is passed for step, it represents slicing backward. However step value cannot be 0. If axes are omitted, they are set to [0, …, ndim-1]. If steps are omitted, they are set to [1, …, 1] of length len(starts) Example 1:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] axes = [0, 1] starts = [1, 0] ends = [2, 3] steps = [1, 2] result = [
[5, 7],
]
- Example 2:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] starts = [0, 1] ends = [-1, 1000] result = [
[2, 3, 4],
]
Inputs
Between 3 and 5 inputs.
data (heterogeneous) - T: Tensor of data to extract slices from.
starts (heterogeneous) - Tind: 1-D tensor of starting indices of corresponding axis in axes
ends (heterogeneous) - Tind: 1-D tensor of ending indices (exclusive) of corresponding axis in axes
axes (optional, heterogeneous) - Tind: 1-D tensor of axes that starts and ends apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
steps (optional, heterogeneous) - Tind: 1-D tensor of slice step of corresponding axis in axes. Negative value means slicing backward. ‘steps’ cannot be 0. Defaults to 1.
Outputs
output (heterogeneous) - T: Sliced data tensor.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxSlice_13#
- class mlprodict.npy.xop_auto_import_.OnnxSlice_13(*args, **kwargs)#
Version
name: Slice (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Produces a slice of the input tensor along multiple axes. Similar to numpy: https://numpy.org/doc/stable/user/basics.indexing.html?highlight=slice#slicing-and-striding
Slice uses the starts, ends, axes and steps inputs to select a sub-tensor of its input data tensor.
An effective start[i], end[i], and step[i] must be computed for each i in [0, … r-1] where r = rank(input) as follows:
If axes are omitted, they are set to [0, …, r-1]. If steps are omitted, they are set to [1, …, 1] of length len(starts)
The effective values are initialized as start[i] = 0, end[i] = dims[i] where dims are the dimensions of input and `step[i] = `1.
All negative elements of axes are made non-negatve by adding r to them, where r =rank(input).
All negative values in starts[i] and ends[i] have dims[axes[i]] added to them, where dims are the dimensions of input. Then start[axes[i]] is the adjusted starts[i] is clamped into the range [0, dims[axes[i]]] for positive stepping and [0, dims[axes[i]]-1] for negative stepping.
The clamping for the adjusted ends[i] depends on the sign of steps[i] and must accommodate copying 0 through dims[axes[i]] elements, so for positive stepping end[axes[i]] is clamped to [0, dims[axes[i]]], while for negative stepping it is clamped to [-1, dims[axes[i]]-1].
Finally, step[axes[i]] = steps[i].
For slicing to the end of a dimension with unknown size, it is recommended to pass in INT_MAX when slicing forward and ‘INT_MIN’ when slicing backward.
- Example 1:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] axes = [0, 1] starts = [1, 0] ends = [2, 3] steps = [1, 2] result = [
[5, 7],
]
- Example 2:
- data = [
[1, 2, 3, 4], [5, 6, 7, 8],
] starts = [0, 1] ends = [-1, 1000] result = [
[2, 3, 4],
]
Inputs
Between 3 and 5 inputs.
data (heterogeneous) - T: Tensor of data to extract slices from.
starts (heterogeneous) - Tind: 1-D tensor of starting indices of corresponding axis in axes
ends (heterogeneous) - Tind: 1-D tensor of ending indices (exclusive) of corresponding axis in axes
axes (optional, heterogeneous) - Tind: 1-D tensor of axes that starts and ends apply to. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data). Behavior is undefined if an axis is repeated.
steps (optional, heterogeneous) - Tind: 1-D tensor of slice step of corresponding axis in axes. Negative value means slicing backward. ‘steps’ cannot be 0. Defaults to 1s.
Outputs
output (heterogeneous) - T: Sliced data tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
Tind in ( tensor(int32), tensor(int64) ): Constrain indices to integer types
OnnxSoftmax#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmax(*args, **kwargs)#
Version
name: Softmax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator computes the normalized exponential values for the given input:
Softmax(input, axis) = Exp(input) / ReduceSum(Exp(input), axis=axis, keepdims=1)
The “axis” attribute indicates the dimension along which Softmax will be performed. The output tensor has the same shape and contains the Softmax values of the corresponding input.
Attributes
- axis:
Describes the dimension Softmax will be performed on. Negative
value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.
Inputs
input (heterogeneous) - T: The input tensor of rank >= axis.
Outputs
output (heterogeneous) - T: The output values with the same shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftmaxCrossEntropyLoss#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmaxCrossEntropyLoss(*args, **kwargs)#
Version
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Loss function that measures the softmax cross entropy between ‘scores’ and ‘labels’. This operator first computes a loss tensor whose shape is identical to the labels input. If the input is 2-D with shape (N, C), the loss tensor may be a N-element vector L = (l_1, l_2, …, l_N). If the input is N-D tensor with shape (N, C, D1, D2, …, Dk), the loss tensor L may have (N, D1, D2, …, Dk) as its shape and L[i,][j_1][j_2]…[j_k] denotes a scalar element in L. After L is available, this operator can optionally do a reduction operator.
- shape(scores): (N, C) where C is the number of classes, or (N, C, D1, D2,…, Dk),
with K >= 1 in case of K-dimensional loss.
- shape(labels): (N) where each value is 0 <= labels[i] <= C-1, or (N, D1, D2,…, Dk),
with K >= 1 in case of K-dimensional loss.
- The loss for one sample, l_i, can caculated as follows:
l[i][d1][d2]…[dk] = -y[i][c][d1][d2]..[dk], where i is the index of classes.
- or
l[i][d1][d2]…[dk] = -y[i][c][d1][d2]..[dk] * weights[c], if ‘weights’ is provided.
- loss is zero for the case when label-value equals ignore_index.
l[i][d1][d2]…[dk] = 0, when labels[n][d1][d2]…[dk] = ignore_index
- where:
p = Softmax(scores) y = Log(p) c = labels[i][d1][d2]…[dk]
Finally, L is optionally reduced: If reduction = ‘none’, the output is L with shape (N, D1, D2, …, Dk). If reduction = ‘sum’, the output is scalar: Sum(L). If reduction = ‘mean’, the output is scalar: ReduceMean(L), or if weight is provided: ReduceSum(L) / ReduceSum(W), where tensor W is of shape (N, D1, D2, …, Dk) and W[n][d1][d2]…[dk] = weights[labels[i][d1][d2]…[dk]].
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. It’s an optional value.
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: no reduction will be applied, ‘sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the number of elements in the output. Default value is
'mean'.
Inputs
Between 2 and 3 inputs.
scores (heterogeneous) - T: The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , …, Dk], where K is the number of dimensions.
labels (heterogeneous) - Tind: The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, …, Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
weights (optional, heterogeneous) - T: A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: Weighted loss float Tensor. If reduction is ‘none’, this has the shape of [batch_size], or [batch_size, D1, D2, …, Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
log_prob (optional, heterogeneous) - T: Log probability tensor. If the output of softmax is prob, its value is log(prob).
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
Tind in ( tensor(int32), tensor(int64) ): Constrain target to integer types
OnnxSoftmaxCrossEntropyLoss_12#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmaxCrossEntropyLoss_12(*args, **kwargs)#
Version
domain: main
since_version: 12
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 12.
Summary
Loss function that measures the softmax cross entropy between ‘scores’ and ‘labels’. This operator first computes a loss tensor whose shape is identical to the labels input. If the input is 2-D with shape (N, C), the loss tensor may be a N-element vector L = (l_1, l_2, …, l_N). If the input is N-D tensor with shape (N, C, D1, D2, …, Dk), the loss tensor L may have (N, D1, D2, …, Dk) as its shape and L[i,][j_1][j_2]…[j_k] denotes a scalar element in L. After L is available, this operator can optionally do a reduction operator.
- shape(scores): (N, C) where C is the number of classes, or (N, C, D1, D2,…, Dk),
with K >= 1 in case of K-dimensional loss.
- shape(labels): (N) where each value is 0 <= labels[i] <= C-1, or (N, D1, D2,…, Dk),
with K >= 1 in case of K-dimensional loss.
- The loss for one sample, l_i, can caculated as follows:
l[i][d1][d2]…[dk] = -y[i][c][d1][d2]..[dk], where i is the index of classes.
- or
l[i][d1][d2]…[dk] = -y[i][c][d1][d2]..[dk] * weights[c], if ‘weights’ is provided.
- loss is zero for the case when label-value equals ignore_index.
l[i][d1][d2]…[dk] = 0, when labels[n][d1][d2]…[dk] = ignore_index
- where:
p = Softmax(scores) y = Log(p) c = labels[i][d1][d2]…[dk]
Finally, L is optionally reduced: If reduction = ‘none’, the output is L with shape (N, D1, D2, …, Dk). If reduction = ‘sum’, the output is scalar: Sum(L). If reduction = ‘mean’, the output is scalar: ReduceMean(L), or if weight is provided: ReduceSum(L) / ReduceSum(W), where tensor W is of shape (N, D1, D2, …, Dk) and W[n][d1][d2]…[dk] = weights[labels[i][d1][d2]…[dk]].
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. It’s an optional value.
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: no reduction will be applied, ‘sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the number of elements in the output. Default value is
'mean'.
Inputs
Between 2 and 3 inputs.
scores (heterogeneous) - T: The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , …, Dk], where K is the number of dimensions.
labels (heterogeneous) - Tind: The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, …, Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
weights (optional, heterogeneous) - T: A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: Weighted loss float Tensor. If reduction is ‘none’, this has the shape of [batch_size], or [batch_size, D1, D2, …, Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
log_prob (optional, heterogeneous) - T: Log probability tensor. If the output of softmax is prob, its value is log(prob).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
Tind in ( tensor(int32), tensor(int64) ): Constrain target to integer types
OnnxSoftmaxCrossEntropyLoss_13#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmaxCrossEntropyLoss_13(*args, **kwargs)#
Version
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Loss function that measures the softmax cross entropy between ‘scores’ and ‘labels’. This operator first computes a loss tensor whose shape is identical to the labels input. If the input is 2-D with shape (N, C), the loss tensor may be a N-element vector L = (l_1, l_2, …, l_N). If the input is N-D tensor with shape (N, C, D1, D2, …, Dk), the loss tensor L may have (N, D1, D2, …, Dk) as its shape and L[i,][j_1][j_2]…[j_k] denotes a scalar element in L. After L is available, this operator can optionally do a reduction operator.
- shape(scores): (N, C) where C is the number of classes, or (N, C, D1, D2,…, Dk),
with K >= 1 in case of K-dimensional loss.
- shape(labels): (N) where each value is 0 <= labels[i] <= C-1, or (N, D1, D2,…, Dk),
with K >= 1 in case of K-dimensional loss.
- The loss for one sample, l_i, can caculated as follows:
l[i][d1][d2]…[dk] = -y[i][c][d1][d2]..[dk], where i is the index of classes.
- or
l[i][d1][d2]…[dk] = -y[i][c][d1][d2]..[dk] * weights[c], if ‘weights’ is provided.
- loss is zero for the case when label-value equals ignore_index.
l[i][d1][d2]…[dk] = 0, when labels[n][d1][d2]…[dk] = ignore_index
- where:
p = Softmax(scores) y = Log(p) c = labels[i][d1][d2]…[dk]
Finally, L is optionally reduced: If reduction = ‘none’, the output is L with shape (N, D1, D2, …, Dk). If reduction = ‘sum’, the output is scalar: Sum(L). If reduction = ‘mean’, the output is scalar: ReduceMean(L), or if weight is provided: ReduceSum(L) / ReduceSum(W), where tensor W is of shape (N, D1, D2, …, Dk) and W[n][d1][d2]…[dk] = weights[labels[i][d1][d2]…[dk]].
Attributes
ignore_index: Specifies a target value that is ignored and does not contribute to the input gradient. It’s an optional value.
reduction: Type of reduction to apply to loss: none, sum, mean(default). ‘none’: no reduction will be applied, ‘sum’: the output will be summed. ‘mean’: the sum of the output will be divided by the number of elements in the output. Default value is
'mean'.
Inputs
Between 2 and 3 inputs.
scores (heterogeneous) - T: The predicted outputs with shape [batch_size, class_size], or [batch_size, class_size, D1, D2 , …, Dk], where K is the number of dimensions.
labels (heterogeneous) - Tind: The ground truth output tensor, with shape [batch_size], or [batch_size, D1, D2, …, Dk], where K is the number of dimensions. Labels element value shall be in range of [0, C). If ignore_index is specified, it may have a value outside [0, C) and the label values should either be in the range [0, C) or have the value ignore_index.
weights (optional, heterogeneous) - T: A manual rescaling weight given to each class. If given, it has to be a 1D Tensor assigning weight to each of the classes. Otherwise, it is treated as if having all ones.
Outputs
Between 1 and 2 outputs.
output (heterogeneous) - T: Weighted loss float Tensor. If reduction is ‘none’, this has the shape of [batch_size], or [batch_size, D1, D2, …, Dk] in case of K-dimensional loss. Otherwise, it is a scalar.
log_prob (optional, heterogeneous) - T: Log probability tensor. If the output of softmax is prob, its value is log(prob).
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
Tind in ( tensor(int32), tensor(int64) ): Constrain target to integer types
OnnxSoftmax_1#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmax_1(*args, **kwargs)#
Version
name: Softmax (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
- The operator computes the softmax (normalized exponential) values for each layer in the batch
of the given input. The input is a 2-D tensor (Tensor<float>) of size
(batch_size x input_feature_dimensions). The output tensor has the same shape and contains the softmax values of the corresponding input.
Input does not need to explicitly be a 2D vector; rather, it will be coerced into one. For an arbitrary n-dimensional tensor input in [a_0, a_1, …, a_{k-1}, a_k, …, a_{n-1}] and k is the axis provided, then input will be coerced into a 2-dimensional tensor with dimensions [a_0 * … * a_{k-1}, a_k * … * a_{n-1}]. For the default case where axis=1, this means the input tensor will be coerced into a 2D tensor of dimensions [a_0, a_1 * … * a_{n-1}], where a_0 is often the batch size. In this situation, we must have a_0 = N and a_1 * … * a_{n-1} = D. Each of these dimensions must be matched correctly, or else the operator will throw errors.
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size Default value is
1.
Inputs
input (heterogeneous) - T: The input tensor that’s coerced into a 2D matrix of size (NxD) as described above.
Outputs
output (heterogeneous) - T: The output values with the same shape as input tensor (the original size without coercion).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftmax_11#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmax_11(*args, **kwargs)#
Version
name: Softmax (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
- The operator computes the softmax (normalized exponential) values for each layer in the batch
of the given input.
The input does not need to explicitly be a 2D vector; rather, it will be coerced into one. For an arbitrary n-dimensional tensor input in [a_0, a_1, …, a_{k-1}, a_k, …, a_{n-1}] and k is the axis provided, then input will be coerced into a 2-dimensional tensor with dimensions [a_0 * … * a_{k-1}, a_k * … * a_{n-1}]. For the default case where axis=1, this means the input tensor will be coerced into a 2D tensor of dimensions [a_0, a_1 * … * a_{n-1}], where a_0 is often the batch size. In this situation, we must have a_0 = N and a_1 * … * a_{n-1} = D. Each of these dimensions must be matched correctly, or else the operator will throw errors. The output tensor has the same shape and contains the softmax values of the corresponding input.
Attributes
axis: Describes the axis of the inputs when coerced to 2D; defaults to one because the 0th axis most likely describes the batch_size. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
1.
Inputs
input (heterogeneous) - T: The input tensor that’s coerced into a 2D matrix of size (NxD) as described above.
Outputs
output (heterogeneous) - T: The output values with the same shape as input tensor (the original size without coercion).
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftmax_13#
- class mlprodict.npy.xop_auto_import_.OnnxSoftmax_13(*args, **kwargs)#
Version
name: Softmax (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
The operator computes the normalized exponential values for the given input:
Softmax(input, axis) = Exp(input) / ReduceSum(Exp(input), axis=axis, keepdims=1)
The “axis” attribute indicates the dimension along which Softmax will be performed. The output tensor has the same shape and contains the Softmax values of the corresponding input.
Attributes
- axis:
Describes the dimension Softmax will be performed on. Negative
value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.
Inputs
input (heterogeneous) - T: The input tensor of rank >= axis.
Outputs
output (heterogeneous) - T: The output values with the same shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftplus#
- class mlprodict.npy.xop_auto_import_.OnnxSoftplus(*args, **kwargs)#
Version
name: Softplus (GitHub)
domain: main
since_version: 1
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Softplus takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the softplus function, y = ln(exp(x) + 1), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: 1D input tensor
Outputs
Y (heterogeneous) - T: 1D input tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftplus_1#
- class mlprodict.npy.xop_auto_import_.OnnxSoftplus_1(*args, **kwargs)#
Version
name: Softplus (GitHub)
domain: main
since_version: 1
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Softplus takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the softplus function, y = ln(exp(x) + 1), is applied to the tensor elementwise.
Inputs
X (heterogeneous) - T: 1D input tensor
Outputs
Y (heterogeneous) - T: 1D input tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftsign#
- class mlprodict.npy.xop_auto_import_.OnnxSoftsign(*args, **kwargs)#
Version
name: Softsign (GitHub)
domain: main
since_version: 1
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Calculates the softsign (x/(1+|x|)) of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The softsign (x/(1+|x|)) values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSoftsign_1#
- class mlprodict.npy.xop_auto_import_.OnnxSoftsign_1(*args, **kwargs)#
Version
name: Softsign (GitHub)
domain: main
since_version: 1
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Calculates the softsign (x/(1+|x|)) of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The softsign (x/(1+|x|)) values of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSpaceToDepth#
- class mlprodict.npy.xop_auto_import_.OnnxSpaceToDepth(*args, **kwargs)#
Version
name: SpaceToDepth (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
SpaceToDepth rearranges blocks of spatial data into depth. More specifically, this op outputs a copy of the input tensor where values from the height and width dimensions are moved to the depth dimension.
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C * blocksize * blocksize, H/blocksize, W/blocksize].
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSpaceToDepth_1#
- class mlprodict.npy.xop_auto_import_.OnnxSpaceToDepth_1(*args, **kwargs)#
Version
name: SpaceToDepth (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
SpaceToDepth rearranges blocks of spatial data into depth. More specifically, this op outputs a copy of the input tensor where values from the height and width dimensions are moved to the depth dimension.
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C * blocksize * blocksize, H/blocksize, W/blocksize].
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSpaceToDepth_13#
- class mlprodict.npy.xop_auto_import_.OnnxSpaceToDepth_13(*args, **kwargs)#
Version
name: SpaceToDepth (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
SpaceToDepth rearranges blocks of spatial data into depth. More specifically, this op outputs a copy of the input tensor where values from the height and width dimensions are moved to the depth dimension.
Attributes
blocksize (required): Blocks of [blocksize, blocksize] are moved.
Inputs
input (heterogeneous) - T: Input tensor of [N,C,H,W], where N is the batch axis, C is the channel or depth, H is the height and W is the width.
Outputs
output (heterogeneous) - T: Output tensor of [N, C * blocksize * blocksize, H/blocksize, W/blocksize].
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSplit#
- class mlprodict.npy.xop_auto_import_.OnnxSplit(*args, **kwargs)#
Version
name: Split (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Split a tensor into a list of tensors, along the specified ‘axis’. Either input ‘split’ or the attribute ‘num_outputs’ should be specified, but not both. If the attribute ‘num_outputs’ is specified, then the tensor is split into equal sized parts. If the tensor is not evenly splittable into num_outputs, the last chunk will be smaller. If the input ‘split’ is specified, it indicates the sizes of each output in the split.
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input). Default value is
0.num_outputs: Number of outputs to split parts of the tensor into. If the tensor is not evenly splittable the last chunk will be smaller.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: The tensor to split
split (optional, heterogeneous) - tensor(int64): Optional length of each output. Values should be >= 0.Sum of the values must be equal to the dim value at ‘axis’ specified.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSplitToSequence#
- class mlprodict.npy.xop_auto_import_.OnnxSplitToSequence(*args, **kwargs)#
Version
name: SplitToSequence (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Split a tensor into a sequence of tensors, along the specified ‘axis’. Lengths of the parts can be specified using argument ‘split’. ‘split’ must contain only positive numbers. ‘split’ is either a scalar (tensor of empty shape), or a 1-D tensor. If ‘split’ is a scalar, then ‘input’ will be split into equally sized chunks(if possible). Last chunk will be smaller if the ‘input’ size along the given axis ‘axis’ is not divisible by ‘split’. Otherwise, the tensor is split into ‘size(split)’ chunks, with lengths of the parts on ‘axis’ specified in ‘split’. In this scenario, the sum of entries in ‘split’ must be equal to the dimension size of input tensor on ‘axis’.
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1]. Default value is
0.keepdims: Keep the split dimension or not. Default 1, which means we keep split dimension. If input ‘split’ is specified, this attribute is ignored. Default value is
1.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: The tensor to split
split (optional, heterogeneous) - I: Length of each output. It can be either a scalar(tensor of empty shape), or a 1-D tensor. All values must be >= 0.
Outputs
output_sequence (heterogeneous) - S: One or more outputs forming a sequence of tensors after splitting
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all tensor types.
I in ( tensor(int32), tensor(int64) ): Constrain split size to integral tensor.
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain output types to all tensor types.
OnnxSplitToSequence_11#
- class mlprodict.npy.xop_auto_import_.OnnxSplitToSequence_11(*args, **kwargs)#
Version
name: SplitToSequence (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Split a tensor into a sequence of tensors, along the specified ‘axis’. Lengths of the parts can be specified using argument ‘split’. ‘split’ must contain only positive numbers. ‘split’ is either a scalar (tensor of empty shape), or a 1-D tensor. If ‘split’ is a scalar, then ‘input’ will be split into equally sized chunks(if possible). Last chunk will be smaller if the ‘input’ size along the given axis ‘axis’ is not divisible by ‘split’. Otherwise, the tensor is split into ‘size(split)’ chunks, with lengths of the parts on ‘axis’ specified in ‘split’. In this scenario, the sum of entries in ‘split’ must be equal to the dimension size of input tensor on ‘axis’.
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1]. Default value is
0.keepdims: Keep the split dimension or not. Default 1, which means we keep split dimension. If input ‘split’ is specified, this attribute is ignored. Default value is
1.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: The tensor to split
split (optional, heterogeneous) - I: Length of each output. It can be either a scalar(tensor of empty shape), or a 1-D tensor. All values must be >= 0.
Outputs
output_sequence (heterogeneous) - S: One or more outputs forming a sequence of tensors after splitting
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input types to all tensor types.
I in ( tensor(int32), tensor(int64) ): Constrain split size to integral tensor.
S in ( seq(tensor(bool)), seq(tensor(complex128)), seq(tensor(complex64)), seq(tensor(double)), seq(tensor(float)), seq(tensor(float16)), seq(tensor(int16)), seq(tensor(int32)), seq(tensor(int64)), seq(tensor(int8)), seq(tensor(string)), seq(tensor(uint16)), seq(tensor(uint32)), seq(tensor(uint64)), seq(tensor(uint8)) ): Constrain output types to all tensor types.
OnnxSplit_1#
- class mlprodict.npy.xop_auto_import_.OnnxSplit_1(*args, **kwargs)#
Version
name: Split (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Split a tensor into a list of tensors, along the specified ‘axis’. The lengths of the split can be specified using argument ‘axis’ or optional second input blob to the operator. Otherwise, the tensor is split to equal sized parts.
Attributes
axis: Which axis to split on
split: length of each output
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: The tensor to split
split (optional, heterogeneous) - T: Optional list of output lengths (see also arg ‘split’)
Outputs
Between 1 and 2147483647 outputs.
outputs… (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
OnnxSplit_11#
- class mlprodict.npy.xop_auto_import_.OnnxSplit_11(*args, **kwargs)#
Version
name: Split (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Split a tensor into a list of tensors, along the specified ‘axis’. Lengths of the parts can be specified using argument ‘split’. Otherwise, the tensor is split to equal sized parts.
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input). Default value is
0.split: length of each output. Values should be >= 0.
Inputs
input (heterogeneous) - T: The tensor to split
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSplit_13#
- class mlprodict.npy.xop_auto_import_.OnnxSplit_13(*args, **kwargs)#
Version
name: Split (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Split a tensor into a list of tensors, along the specified ‘axis’. Lengths of the parts can be specified using input ‘split’. Otherwise, the tensor is split to equal sized parts.
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input). Default value is
0.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: The tensor to split
split (optional, heterogeneous) - tensor(int64): Optional length of each output. Values should be >= 0.Sum of the values must be equal to the dim value at ‘axis’ specified.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSplit_18#
- class mlprodict.npy.xop_auto_import_.OnnxSplit_18(*args, **kwargs)#
Version
name: Split (GitHub)
domain: main
since_version: 18
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 18.
Summary
Split a tensor into a list of tensors, along the specified ‘axis’. Either input ‘split’ or the attribute ‘num_outputs’ should be specified, but not both. If the attribute ‘num_outputs’ is specified, then the tensor is split into equal sized parts. If the tensor is not evenly splittable into num_outputs, the last chunk will be smaller. If the input ‘split’ is specified, it indicates the sizes of each output in the split.
Attributes
axis: Which axis to split on. A negative value means counting dimensions from the back. Accepted range is [-rank, rank-1] where r = rank(input). Default value is
0.num_outputs: Number of outputs to split parts of the tensor into. If the tensor is not evenly splittable the last chunk will be smaller.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: The tensor to split
split (optional, heterogeneous) - tensor(int64): Optional length of each output. Values should be >= 0.Sum of the values must be equal to the dim value at ‘axis’ specified.
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSplit_2#
- class mlprodict.npy.xop_auto_import_.OnnxSplit_2(*args, **kwargs)#
Version
name: Split (GitHub)
domain: main
since_version: 2
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 2.
Summary
Split a tensor into a list of tensors, along the specified ‘axis’. Lengths of the parts can be specified using argument ‘split’. Otherwise, the tensor is split to equal sized parts.
Attributes
axis: Which axis to split on. Default value is
0.split: length of each output
Inputs
input (heterogeneous) - T: The tensor to split
Outputs
Between 1 and 2147483647 outputs.
outputs (variadic, heterogeneous) - T: One or more outputs forming list of tensors after splitting
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSqrt#
- class mlprodict.npy.xop_auto_import_.OnnxSqrt(*args, **kwargs)#
Version
name: Sqrt (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Square root takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the square root is, y = x^0.5, is applied to the tensor elementwise. If x is negative, then it will return NaN.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSqrt_1#
- class mlprodict.npy.xop_auto_import_.OnnxSqrt_1(*args, **kwargs)#
Version
name: Sqrt (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Square root takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the square root is, y = x^0.5, is applied to the tensor elementwise. If x is negative, then it will return NaN.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSqrt_13#
- class mlprodict.npy.xop_auto_import_.OnnxSqrt_13(*args, **kwargs)#
Version
name: Sqrt (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Square root takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the square root is, y = x^0.5, is applied to the tensor elementwise. If x is negative, then it will return NaN.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSqrt_6#
- class mlprodict.npy.xop_auto_import_.OnnxSqrt_6(*args, **kwargs)#
Version
name: Sqrt (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Square root takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the square root is, y = x^0.5, is applied to the tensor elementwise. If x is negative, then it will return NaN.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSqueeze#
- class mlprodict.npy.xop_auto_import_.OnnxSqueeze(*args, **kwargs)#
Version
name: Squeeze (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Remove single-dimensional entries from the shape of a tensor. Takes an input axes with a list of axes to squeeze. If axes is not provided, all the single dimensions will be removed from the shape. If an axis is selected with shape entry not equal to one, an error is raised.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: Tensors with at least max(dims) dimensions.
axes (optional, heterogeneous) - tensor(int64): List of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
Outputs
squeezed (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSqueeze_1#
- class mlprodict.npy.xop_auto_import_.OnnxSqueeze_1(*args, **kwargs)#
Version
name: Squeeze (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Remove single-dimensional entries from the shape of a tensor. Takes a parameter axes with a list of axes to squeeze. If axes is not provided, all the single dimensions will be removed from the shape. If an axis is selected with shape entry not equal to one, an error is raised.
Attributes
axes: List of non-negative integers, indicate the dimensions to squeeze.
Inputs
data (heterogeneous) - T: Tensors with at least max(dims) dimensions.
Outputs
squeezed (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSqueeze_11#
- class mlprodict.npy.xop_auto_import_.OnnxSqueeze_11(*args, **kwargs)#
Version
name: Squeeze (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Remove single-dimensional entries from the shape of a tensor. Takes a parameter axes with a list of axes to squeeze. If axes is not provided, all the single dimensions will be removed from the shape. If an axis is selected with shape entry not equal to one, an error is raised.
Attributes
axes: List of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
Inputs
data (heterogeneous) - T: Tensors with at least max(dims) dimensions.
Outputs
squeezed (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxSqueeze_13#
- class mlprodict.npy.xop_auto_import_.OnnxSqueeze_13(*args, **kwargs)#
Version
name: Squeeze (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Remove single-dimensional entries from the shape of a tensor. Takes an input axes with a list of axes to squeeze. If axes is not provided, all the single dimensions will be removed from the shape. If an axis is selected with shape entry not equal to one, an error is raised.
Inputs
Between 1 and 2 inputs.
data (heterogeneous) - T: Tensors with at least max(dims) dimensions.
axes (optional, heterogeneous) - tensor(int64): List of integers indicating the dimensions to squeeze. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(data).
Outputs
squeezed (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxStringNormalizer#
- class mlprodict.npy.xop_auto_import_.OnnxStringNormalizer(*args, **kwargs)#
Version
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
StringNormalization performs string operations for basic cleaning. This operator has only one input (denoted by X) and only one output (denoted by Y). This operator first examines the elements in the X, and removes elements specified in “stopwords” attribute. After removing stop words, the intermediate result can be further lowercased, uppercased, or just returned depending the “case_change_action” attribute. This operator only accepts [C]- and [1, C]-tensor. If all elements in X are dropped, the output will be the empty value of string tensor with shape [1] if input shape is [C] and shape [1, 1] if input shape is [1, C].
Attributes
case_change_action: string enum that cases output to be lowercased/uppercases/unchanged. Valid values are “LOWER”, “UPPER”, “NONE”. Default is “NONE” Default value is
'NONE'.is_case_sensitive: Boolean. Whether the identification of stop words in X is case- sensitive. Default is false Default value is
0.locale: Environment dependent string that denotes the locale according to which output strings needs to be upper/lowercased.Default en_US or platform specific equivalent as decided by the implementation.
stopwords: List of stop words. If not set, no word would be removed from X.
Inputs
X (heterogeneous) - tensor(string): UTF-8 strings to normalize
Outputs
Y (heterogeneous) - tensor(string): UTF-8 Normalized strings
OnnxStringNormalizer_10#
- class mlprodict.npy.xop_auto_import_.OnnxStringNormalizer_10(*args, **kwargs)#
Version
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
StringNormalization performs string operations for basic cleaning. This operator has only one input (denoted by X) and only one output (denoted by Y). This operator first examines the elements in the X, and removes elements specified in “stopwords” attribute. After removing stop words, the intermediate result can be further lowercased, uppercased, or just returned depending the “case_change_action” attribute. This operator only accepts [C]- and [1, C]-tensor. If all elements in X are dropped, the output will be the empty value of string tensor with shape [1] if input shape is [C] and shape [1, 1] if input shape is [1, C].
Attributes
case_change_action: string enum that cases output to be lowercased/uppercases/unchanged. Valid values are “LOWER”, “UPPER”, “NONE”. Default is “NONE” Default value is
'NONE'.is_case_sensitive: Boolean. Whether the identification of stop words in X is case- sensitive. Default is false Default value is
0.locale: Environment dependent string that denotes the locale according to which output strings needs to be upper/lowercased.Default en_US or platform specific equivalent as decided by the implementation.
stopwords: List of stop words. If not set, no word would be removed from X.
Inputs
X (heterogeneous) - tensor(string): UTF-8 strings to normalize
Outputs
Y (heterogeneous) - tensor(string): UTF-8 Normalized strings
OnnxSub#
- class mlprodict.npy.xop_auto_import_.OnnxSub(*args, **kwargs)#
Version
name: Sub (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary subtraction (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxSub_1#
- class mlprodict.npy.xop_auto_import_.OnnxSub_1(*args, **kwargs)#
Version
name: Sub (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Performs element-wise binary subtraction (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.consumed_inputs: legacy optimization attribute.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSub_13#
- class mlprodict.npy.xop_auto_import_.OnnxSub_13(*args, **kwargs)#
Version
name: Sub (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Performs element-wise binary subtraction (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxSub_14#
- class mlprodict.npy.xop_auto_import_.OnnxSub_14(*args, **kwargs)#
Version
name: Sub (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Performs element-wise binary subtraction (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
(Opset 14 change): Extend supported types to include uint8, int8, uint16, and int16.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all numeric tensors.
OnnxSub_6#
- class mlprodict.npy.xop_auto_import_.OnnxSub_6(*args, **kwargs)#
Version
name: Sub (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Performs element-wise binary subtraction (with limited broadcast support).
If necessary the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. When broadcasting is specified, the second tensor can either be of element size 1 (including a scalar tensor and any tensor with rank equal to or smaller than the first tensor), or having its shape as a contiguous subset of the first tensor’s shape. The starting of the mutually equal shape is specified by the argument “axis”, and if it is not set, suffix matching is assumed. 1-dim expansion doesn’t work yet.
For example, the following tensor shapes are supported (with broadcast=1):
shape(A) = (2, 3, 4, 5), shape(B) = (,), i.e. B is a scalar tensor shape(A) = (2, 3, 4, 5), shape(B) = (1, 1), i.e. B is an 1-element tensor shape(A) = (2, 3, 4, 5), shape(B) = (5,) shape(A) = (2, 3, 4, 5), shape(B) = (4, 5) shape(A) = (2, 3, 4, 5), shape(B) = (3, 4), with axis=1 shape(A) = (2, 3, 4, 5), shape(B) = (2), with axis=0
Attribute broadcast=1 needs to be passed to enable broadcasting.
Attributes
axis: If set, defines the broadcast dimensions. See doc for details.
broadcast: Pass 1 to enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: First operand, should share the type with the second operand.
B (heterogeneous) - T: Second operand. With broadcasting can be of smaller size than A. If broadcasting is disabled it should be of the same size.
Outputs
C (heterogeneous) - T: Result, has same dimensions and type as A
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxSub_7#
- class mlprodict.npy.xop_auto_import_.OnnxSub_7(*args, **kwargs)#
Version
name: Sub (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Performs element-wise binary subtraction (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First operand.
B (heterogeneous) - T: Second operand.
Outputs
C (heterogeneous) - T: Result, has same element type as two inputs
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int32), tensor(int64), tensor(uint32), tensor(uint64) ): Constrain input and output types to high-precision numeric tensors.
OnnxSum#
- class mlprodict.npy.xop_auto_import_.OnnxSum(*args, **kwargs)#
Version
name: Sum (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise sum of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for sum.
Outputs
sum (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSum_1#
- class mlprodict.npy.xop_auto_import_.OnnxSum_1(*args, **kwargs)#
Version
name: Sum (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Element-wise sum of each of the input tensors. All inputs and outputs must have the same shape and data type.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Sum.
Outputs
sum (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSum_13#
- class mlprodict.npy.xop_auto_import_.OnnxSum_13(*args, **kwargs)#
Version
name: Sum (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Element-wise sum of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for sum.
Outputs
sum (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSum_6#
- class mlprodict.npy.xop_auto_import_.OnnxSum_6(*args, **kwargs)#
Version
name: Sum (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Element-wise sum of each of the input tensors. All inputs and outputs must have the same shape and data type.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for Sum.
Outputs
sum (heterogeneous) - T: Output tensor. Same dimension as inputs.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxSum_8#
- class mlprodict.npy.xop_auto_import_.OnnxSum_8(*args, **kwargs)#
Version
name: Sum (GitHub)
domain: main
since_version: 8
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 8.
Summary
Element-wise sum of each of the input tensors (with Numpy-style broadcasting support). All inputs and outputs must have the same data type. This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
Between 1 and 2147483647 inputs.
data_0 (variadic, heterogeneous) - T: List of tensors for sum.
Outputs
sum (heterogeneous) - T: Output tensor.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTan#
- class mlprodict.npy.xop_auto_import_.OnnxTan(*args, **kwargs)#
Version
name: Tan (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the tangent of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The tangent of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTan_7#
- class mlprodict.npy.xop_auto_import_.OnnxTan_7(*args, **kwargs)#
Version
name: Tan (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Calculates the tangent of the given input tensor, element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The tangent of the input tensor computed element-wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTanh#
- class mlprodict.npy.xop_auto_import_.OnnxTanh(*args, **kwargs)#
Version
name: Tanh (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculates the hyperbolic tangent of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic tangent values of the input tensor computed element- wise
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTanh_1#
- class mlprodict.npy.xop_auto_import_.OnnxTanh_1(*args, **kwargs)#
Version
name: Tanh (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: False
This version of the operator has been available since version 1.
Summary
Calculates the hyperbolic tangent of the given input tensor element-wise.
Attributes
consumed_inputs: legacy optimization attribute.
Inputs
input (heterogeneous) - T: 1-D input tensor
Outputs
output (heterogeneous) - T: The hyperbolic tangent values of the input tensor computed element- wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTanh_13#
- class mlprodict.npy.xop_auto_import_.OnnxTanh_13(*args, **kwargs)#
Version
name: Tanh (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Calculates the hyperbolic tangent of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic tangent values of the input tensor computed element- wise
Type Constraints
T in ( tensor(bfloat16), tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTanh_6#
- class mlprodict.npy.xop_auto_import_.OnnxTanh_6(*args, **kwargs)#
Version
name: Tanh (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Calculates the hyperbolic tangent of the given input tensor element-wise.
Inputs
input (heterogeneous) - T: Input tensor
Outputs
output (heterogeneous) - T: The hyperbolic tangent values of the input tensor computed element- wise
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTfIdfVectorizer#
- class mlprodict.npy.xop_auto_import_.OnnxTfIdfVectorizer(*args, **kwargs)#
Version
name: TfIdfVectorizer (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
This transform extracts n-grams from the input sequence and save them as a vector. Input can be either a 1-D or 2-D tensor. For 1-D input, output is the n-gram representation of that input. For 2-D input, the output is also a 2-D tensor whose i-th row is the n-gram representation of the i-th input row. More specifically, if input shape is [C], the corresponding output shape would be [max(ngram_indexes) + 1]. If input shape is [N, C], this operator produces a [N, max(ngram_indexes) + 1]-tensor.
In contrast to standard n-gram extraction, here, the indexes of extracting an n-gram from the original sequence are not necessarily consecutive numbers. The discontinuity between indexes are controlled by the number of skips. If the number of skips is 2, we should skip two tokens when scanning through the original sequence. Let’s consider an example. Assume that input sequence is [94, 17, 36, 12, 28] and the number of skips is 2. The associated 2-grams are [94, 12] and [17, 28] respectively indexed by [0, 3] and [1, 4]. If the number of skips becomes 0, the 2-grams generated are [94, 17], [17, 36], [36, 12], [12, 28] indexed by [0, 1], [1, 2], [2, 3], [3, 4], respectively.
The output vector (denoted by Y) stores the count of each n-gram; Y[ngram_indexes[i]] indicates the times that the i-th n-gram is found. The attribute ngram_indexes is used to determine the mapping between index i and the corresponding n-gram’s output coordinate. If pool_int64s is [94, 17, 17, 36], ngram_indexes is [1, 0], ngram_counts=[0, 0], then the Y[0] (first element in Y) and Y[1] (second element in Y) are the counts of [17, 36] and [94, 17], respectively. An n-gram which cannot be found in pool_strings/pool_int64s should be ignored and has no effect on the output. Note that we may consider all skips up to S when generating the n-grams.
The examples used above are true if mode is “TF”. If mode is “IDF”, all the counts larger than 1 would be truncated to 1 and the i-th element in weights would be used to scale (by multiplication) the count of the i-th n-gram in pool. If mode is “TFIDF”, this operator first computes the counts of all n-grams and then scale them by the associated values in the weights attribute.
Only one of pool_strings and pool_int64s can be set. If pool_int64s is set, the input should be an integer tensor. If pool_strings is set, the input must be a string tensor.
Attributes
max_gram_length (required): Maximum n-gram length. If this value is 3, 3-grams will be used to generate the output.
max_skip_count (required): Maximum number of items (integers/strings) to be skipped when constructing an n-gram from X. If max_skip_count=1, min_gram_length=2, max_gram_length=3, this operator may generate 2-grams with skip_count=0 and skip_count=1, and 3-grams with skip_count=0 and skip_count=1
min_gram_length (required): Minimum n-gram length. If this value is 2 and max_gram_length is 3, output may contain counts of 2-grams and 3-grams.
mode (required): The weighting criteria. It can be one of “TF” (term frequency), “IDF” (inverse document frequency), and “TFIDF” (the combination of TF and IDF)
ngram_counts (required): The starting indexes of 1-grams, 2-grams, and so on in pool. It is useful when determining the boundary between two consecutive collections of n-grams. For example, if ngram_counts is [0, 17, 36], the first index (zero-based) of 1-gram/2-gram/3-gram in pool are 0/17/36. This format is essentially identical to CSR (or CSC) sparse matrix format, and we choose to use this due to its popularity.
ngram_indexes (required): list of int64s (type: AttributeProto::INTS). This list is parallel to the specified ‘pool_*’ attribute. The i-th element in ngram_indexes indicate the coordinate of the i-th n-gram in the output tensor.
pool_int64s: List of int64 n-grams learned from the training set. Either this or pool_strings attributes must be present but not both. It’s an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
pool_strings: List of strings n-grams learned from the training set. Either this or pool_int64s attributes must be present but not both. It’s an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
weights: list of floats. This attribute stores the weight of each n-gram in pool. The i-th element in weights is the weight of the i-th n-gram in pool. Its length equals to the size of ngram_indexes. By default, weights is an all-one tensor.This attribute is used when mode is “IDF” or “TFIDF” to scale the associated word counts.
Inputs
X (heterogeneous) - T: Input for n-gram extraction
Outputs
Y (heterogeneous) - T1: Ngram results
Type Constraints
T in ( tensor(int32), tensor(int64), tensor(string) ): Input is ether string UTF-8 or int32/int64
T1 in ( tensor(float) ): 1-D tensor of floats
OnnxTfIdfVectorizer_9#
- class mlprodict.npy.xop_auto_import_.OnnxTfIdfVectorizer_9(*args, **kwargs)#
Version
name: TfIdfVectorizer (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
This transform extracts n-grams from the input sequence and save them as a vector. Input can be either a 1-D or 2-D tensor. For 1-D input, output is the n-gram representation of that input. For 2-D input, the output is also a 2-D tensor whose i-th row is the n-gram representation of the i-th input row. More specifically, if input shape is [C], the corresponding output shape would be [max(ngram_indexes) + 1]. If input shape is [N, C], this operator produces a [N, max(ngram_indexes) + 1]-tensor.
In contrast to standard n-gram extraction, here, the indexes of extracting an n-gram from the original sequence are not necessarily consecutive numbers. The discontinuity between indexes are controlled by the number of skips. If the number of skips is 2, we should skip two tokens when scanning through the original sequence. Let’s consider an example. Assume that input sequence is [94, 17, 36, 12, 28] and the number of skips is 2. The associated 2-grams are [94, 12] and [17, 28] respectively indexed by [0, 3] and [1, 4]. If the number of skips becomes 0, the 2-grams generated are [94, 17], [17, 36], [36, 12], [12, 28] indexed by [0, 1], [1, 2], [2, 3], [3, 4], respectively.
The output vector (denoted by Y) stores the count of each n-gram; Y[ngram_indexes[i]] indicates the times that the i-th n-gram is found. The attribute ngram_indexes is used to determine the mapping between index i and the corresponding n-gram’s output coordinate. If pool_int64s is [94, 17, 17, 36], ngram_indexes is [1, 0], ngram_counts=[0, 0], then the Y[0] (first element in Y) and Y[1] (second element in Y) are the counts of [17, 36] and [94, 17], respectively. An n-gram which cannot be found in pool_strings/pool_int64s should be ignored and has no effect on the output. Note that we may consider all skips up to S when generating the n-grams.
The examples used above are true if mode is “TF”. If mode is “IDF”, all the counts larger than 1 would be truncated to 1 and the i-th element in weights would be used to scale (by multiplication) the count of the i-th n-gram in pool. If mode is “TFIDF”, this operator first computes the counts of all n-grams and then scale them by the associated values in the weights attribute.
Only one of pool_strings and pool_int64s can be set. If pool_int64s is set, the input should be an integer tensor. If pool_strings is set, the input must be a string tensor.
Attributes
max_gram_length (required): Maximum n-gram length. If this value is 3, 3-grams will be used to generate the output.
max_skip_count (required): Maximum number of items (integers/strings) to be skipped when constructing an n-gram from X. If max_skip_count=1, min_gram_length=2, max_gram_length=3, this operator may generate 2-grams with skip_count=0 and skip_count=1, and 3-grams with skip_count=0 and skip_count=1
min_gram_length (required): Minimum n-gram length. If this value is 2 and max_gram_length is 3, output may contain counts of 2-grams and 3-grams.
mode (required): The weighting criteria. It can be one of “TF” (term frequency), “IDF” (inverse document frequency), and “TFIDF” (the combination of TF and IDF)
ngram_counts (required): The starting indexes of 1-grams, 2-grams, and so on in pool. It is useful when determining the boundary between two consecutive collections of n-grams. For example, if ngram_counts is [0, 17, 36], the first index (zero-based) of 1-gram/2-gram/3-gram in pool are 0/17/36. This format is essentially identical to CSR (or CSC) sparse matrix format, and we choose to use this due to its popularity.
ngram_indexes (required): list of int64s (type: AttributeProto::INTS). This list is parallel to the specified ‘pool_*’ attribute. The i-th element in ngram_indexes indicate the coordinate of the i-th n-gram in the output tensor.
pool_int64s: List of int64 n-grams learned from the training set. Either this or pool_strings attributes must be present but not both. It’s an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
pool_strings: List of strings n-grams learned from the training set. Either this or pool_int64s attributes must be present but not both. It’s an 1-D tensor starting with the collections of all 1-grams and ending with the collections of n-grams. The i-th element in pool stores the n-gram that should be mapped to coordinate ngram_indexes[i] in the output vector.
weights: list of floats. This attribute stores the weight of each n-gram in pool. The i-th element in weights is the weight of the i-th n-gram in pool. Its length equals to the size of ngram_indexes. By default, weights is an all-one tensor.This attribute is used when mode is “IDF” or “TFIDF” to scale the associated word counts.
Inputs
X (heterogeneous) - T: Input for n-gram extraction
Outputs
Y (heterogeneous) - T1: Ngram results
Type Constraints
T in ( tensor(int32), tensor(int64), tensor(string) ): Input is ether string UTF-8 or int32/int64
T1 in ( tensor(float) ): 1-D tensor of floats
OnnxThresholdedRelu#
- class mlprodict.npy.xop_auto_import_.OnnxThresholdedRelu(*args, **kwargs)#
Version
name: ThresholdedRelu (GitHub)
domain: main
since_version: 10
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
ThresholdedRelu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = x for x > alpha, y = 0 otherwise, is applied to the tensor elementwise.
Attributes
alpha: Threshold value Default value is
1.0.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxThresholdedRelu_10#
- class mlprodict.npy.xop_auto_import_.OnnxThresholdedRelu_10(*args, **kwargs)#
Version
name: ThresholdedRelu (GitHub)
domain: main
since_version: 10
function: True
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
ThresholdedRelu takes one input data (Tensor<T>) and produces one output data (Tensor<T>) where the rectified linear function, y = x for x > alpha, y = 0 otherwise, is applied to the tensor elementwise.
Attributes
alpha: Threshold value Default value is
1.0.
Inputs
X (heterogeneous) - T: Input tensor
Outputs
Y (heterogeneous) - T: Output tensor
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
OnnxTile#
- class mlprodict.npy.xop_auto_import_.OnnxTile(*args, **kwargs)#
Version
name: Tile (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Constructs a tensor by tiling a given tensor. This is the same as function tile in Numpy, but no broadcast. For example A = [[1, 2], [3, 4]], B = [1, 2], tile(A, B) = [[1, 2, 1, 2], [3, 4, 3, 4]]
Inputs
input (heterogeneous) - T: Input tensor of any shape.
repeats (heterogeneous) - T1: 1D int64 tensor of the same length as input’s dimension number, includes numbers of repeated copies along input’s dimensions.
Outputs
output (heterogeneous) - T: Output tensor of the same dimensions and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
T1 in ( tensor(int64) ): Constrain repeat’s type to int64 tensors.
OnnxTile_1#
- class mlprodict.npy.xop_auto_import_.OnnxTile_1(*args, **kwargs)#
Version
name: Tile (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Repeat the elements of a tensor along an axis.
Inputs
input (heterogeneous) - T: Input tensor of any shape.
tiles (heterogeneous) - T: Number of repeated copies to make of the input tensor.
axis (heterogeneous) - T: Axis along which to repeat.
Outputs
output (heterogeneous) - T: Output tensor of same shape and type as input.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input types to float tensors.
T1 in ( tensor(int64) ): Constrain tiles and axis’s type to int64 tensors.
OnnxTile_13#
- class mlprodict.npy.xop_auto_import_.OnnxTile_13(*args, **kwargs)#
Version
name: Tile (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Constructs a tensor by tiling a given tensor. This is the same as function tile in Numpy, but no broadcast. For example A = [[1, 2], [3, 4]], B = [1, 2], tile(A, B) = [[1, 2, 1, 2], [3, 4, 3, 4]]
Inputs
input (heterogeneous) - T: Input tensor of any shape.
repeats (heterogeneous) - T1: 1D int64 tensor of the same length as input’s dimension number, includes numbers of repeated copies along input’s dimensions.
Outputs
output (heterogeneous) - T: Output tensor of the same dimensions and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
T1 in ( tensor(int64) ): Constrain repeat’s type to int64 tensors.
OnnxTile_6#
- class mlprodict.npy.xop_auto_import_.OnnxTile_6(*args, **kwargs)#
Version
name: Tile (GitHub)
domain: main
since_version: 6
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 6.
Summary
Constructs a tensor by tiling a given tensor. This is the same as function tile in Numpy, but no broadcast. For example A = [[1, 2], [3, 4]], B = [1, 2], tile(A, B) = [[1, 2, 1, 2], [3, 4, 3, 4]]
Inputs
input (heterogeneous) - T: Input tensor of any shape.
repeats (heterogeneous) - T1: 1D int64 tensor of the same length as input’s dimension number, includes numbers of repeated copies along input’s dimensions.
Outputs
output (heterogeneous) - T: Output tensor of the same dimensions and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
T1 in ( tensor(int64) ): Constrain repeat’s type to int64 tensors.
OnnxTopK#
- class mlprodict.npy.xop_auto_import_.OnnxTopK(*args, **kwargs)#
Version
name: TopK (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Retrieve the top-K largest or smallest elements along a specified axis. Given an input tensor of shape [a_1, a_2, …, a_n, r] and integer argument k, return two outputs:
- -Value tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n]
which contains the values of the top k elements along the specified axis
- -Index tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] which
contains the indices of the top k elements (original indices from the input tensor).
If “largest” is 1 (the default value) then the k largest elements are returned. If “sorted” is 1 (the default value) then the resulting k elements will be sorted. If “sorted” is 0, order of returned ‘Values’ and ‘Indices’ are undefined.
- Given two equivalent values, this operator uses the indices along the axis as
a tiebreaker. That is, the element with the lower index will appear first.
Attributes
axis: Dimension on which to do the sort. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.largest: Whether to return the top-K largest or smallest elements. Default value is
1.sorted: Whether to return the elements in sorted order. Default value is
1.
Inputs
X (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_n, r]
K (heterogeneous) - tensor(int64): A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
Outputs
Values (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing top K values from the input tensor
Indices (heterogeneous) - I: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing the corresponding input tensor indices for the top K values.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxTopK_1#
- class mlprodict.npy.xop_auto_import_.OnnxTopK_1(*args, **kwargs)#
Version
name: TopK (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Retrieve the top-K elements along a specified axis. Given an input tensor of shape [a_1, a_2, …, a_n, r] and integer argument k, return two outputs:
- -Value tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n]
which contains the values of the top k elements along the specified axis
- -Index tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] which
contains the indices of the top k elements (original indices from the input tensor).
- Given two equivalent values, this operator uses the indices along the axis as
a tiebreaker. That is, the element with the lower index will appear first.
Attributes
axis: Dimension on which to do the sort. Default value is
-1.k (required): Number of top elements to retrieve
Inputs
X (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_n, r]
Outputs
Values (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing top K values from the input tensor
Indices (heterogeneous) - I: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing the corresponding input tensor indices for the top K values.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxTopK_10#
- class mlprodict.npy.xop_auto_import_.OnnxTopK_10(*args, **kwargs)#
Version
name: TopK (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 10.
Summary
Retrieve the top-K elements along a specified axis. Given an input tensor of shape [a_1, a_2, …, a_n, r] and integer argument k, return two outputs:
- -Value tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n]
which contains the values of the top k elements along the specified axis
- -Index tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] which
contains the indices of the top k elements (original indices from the input tensor).
- Given two equivalent values, this operator uses the indices along the axis as
a tiebreaker. That is, the element with the lower index will appear first.
Attributes
axis: Dimension on which to do the sort. Default value is
-1.
Inputs
X (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_n, r]
K (heterogeneous) - tensor(int64): A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
Outputs
Values (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing top K values from the input tensor
Indices (heterogeneous) - I: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing the corresponding input tensor indices for the top K values.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16) ): Constrain input and output types to float tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxTopK_11#
- class mlprodict.npy.xop_auto_import_.OnnxTopK_11(*args, **kwargs)#
Version
name: TopK (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Retrieve the top-K largest or smallest elements along a specified axis. Given an input tensor of shape [a_1, a_2, …, a_n, r] and integer argument k, return two outputs:
- -Value tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n]
which contains the values of the top k elements along the specified axis
- -Index tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] which
contains the indices of the top k elements (original indices from the input tensor).
If “largest” is 1 (the default value) then the k largest elements are returned. If “sorted” is 1 (the default value) then the resulting k elements will be sorted. If “sorted” is 0, order of returned ‘Values’ and ‘Indices’ are undefined.
- Given two equivalent values, this operator uses the indices along the axis as
a tiebreaker. That is, the element with the lower index will appear first.
Attributes
axis: Dimension on which to do the sort. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input). Default value is
-1.largest: Whether to return the top-K largest or smallest elements. Default value is
1.sorted: Whether to return the elements in sorted order. Default value is
1.
Inputs
X (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_n, r]
K (heterogeneous) - tensor(int64): A 1-D tensor containing a single positive value corresponding to the number of top elements to retrieve
Outputs
Values (heterogeneous) - T: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing top K values from the input tensor
Indices (heterogeneous) - I: Tensor of shape [a_1, a_2, …, a_{axis-1}, k, a_{axis+1}, … a_n] containing the corresponding input tensor indices for the top K values.
Type Constraints
T in ( tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to numeric tensors.
I in ( tensor(int64) ): Constrain index tensor to int64
OnnxTranspose#
- class mlprodict.npy.xop_auto_import_.OnnxTranspose(*args, **kwargs)#
Version
name: Transpose (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Transpose the input tensor similar to numpy.transpose. For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), the output shape will be (2, 1, 3).
Attributes
perm: A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
transposed (heterogeneous) - T: Transposed output.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxTranspose_1#
- class mlprodict.npy.xop_auto_import_.OnnxTranspose_1(*args, **kwargs)#
Version
name: Transpose (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Transpose the input tensor similar to numpy.transpose. For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), the output shape will be (2, 1, 3).
Attributes
perm: A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
transposed (heterogeneous) - T: Transposed output.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxTranspose_13#
- class mlprodict.npy.xop_auto_import_.OnnxTranspose_13(*args, **kwargs)#
Version
name: Transpose (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Transpose the input tensor similar to numpy.transpose. For example, when perm=(1, 0, 2), given an input tensor of shape (1, 2, 3), the output shape will be (2, 1, 3).
Attributes
perm: A list of integers. By default, reverse the dimensions, otherwise permute the axes according to the values given.
Inputs
data (heterogeneous) - T: An input tensor.
Outputs
transposed (heterogeneous) - T: Transposed output.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxTrilu#
- class mlprodict.npy.xop_auto_import_.OnnxTrilu(*args, **kwargs)#
Version
name: Trilu (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Given a 2-D matrix or batches of 2-D matrices, returns the upper or lower triangular part of the tensor(s). The attribute “upper” determines whether the upper or lower part is retained. If set to true, the upper triangular matrix is retained. Lower triangular matrix is retained otherwise. Default value for the “upper” attribute is true. Trilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists of the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal. All other elements in the matrix are set to zero. If k = 0, the triangular part on and above/below the main diagonal is retained. If upper is set to true, a positive k retains the upper triangular matrix excluding the main diagonal and (k-1) diagonals above it. A negative k value retains the main diagonal and |k| diagonals below it. If upper is set to false, a positive k retains the lower triangular matrix including the main diagonal and k diagonals above it. A negative k value excludes the main diagonal and (|k|-1) diagonals below it.
Attributes
upper: Boolean. Indicates whether upper or lower part of matrix is retained. Default is true. Default value is
1.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: Input tensor of rank 2 or higher.
k (optional, heterogeneous) - tensor(int64): A 0-D tensor containing a single value corresponding to the number diagonals above or below the main diagonal to exclude or include. Default value is 0 if it’s not specified.
Outputs
output (heterogeneous) - T: Output tensor of the same type and shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxTrilu_14#
- class mlprodict.npy.xop_auto_import_.OnnxTrilu_14(*args, **kwargs)#
Version
name: Trilu (GitHub)
domain: main
since_version: 14
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 14.
Summary
Given a 2-D matrix or batches of 2-D matrices, returns the upper or lower triangular part of the tensor(s). The attribute “upper” determines whether the upper or lower part is retained. If set to true, the upper triangular matrix is retained. Lower triangular matrix is retained otherwise. Default value for the “upper” attribute is true. Trilu takes one input tensor of shape [*, N, M], where * is zero or more batch dimensions. The upper triangular part consists of the elements on and above the given diagonal (k). The lower triangular part consists of elements on and below the diagonal. All other elements in the matrix are set to zero. If k = 0, the triangular part on and above/below the main diagonal is retained. If upper is set to true, a positive k retains the upper triangular matrix excluding the main diagonal and (k-1) diagonals above it. A negative k value retains the main diagonal and |k| diagonals below it. If upper is set to false, a positive k retains the lower triangular matrix including the main diagonal and k diagonals above it. A negative k value excludes the main diagonal and (|k|-1) diagonals below it.
Attributes
upper: Boolean. Indicates whether upper or lower part of matrix is retained. Default is true. Default value is
1.
Inputs
Between 1 and 2 inputs.
input (heterogeneous) - T: Input tensor of rank 2 or higher.
k (optional, heterogeneous) - tensor(int64): A 0-D tensor containing a single value corresponding to the number diagonals above or below the main diagonal to exclude or include. Default value is 0 if it’s not specified.
Outputs
output (heterogeneous) - T: Output tensor of the same type and shape as the input tensor.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxUnique#
- class mlprodict.npy.xop_auto_import_.OnnxUnique(*args, **kwargs)#
Version
name: Unique (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Find the unique elements of a tensor. When an optional attribute ‘axis’ is provided, unique subtensors sliced along the ‘axis’ are returned. Otherwise the input tensor is flattened and unique values of the flattened tensor are returned.
This operator returns the unique values or sliced unique subtensors of the input tensor and three optional outputs. The first output tensor ‘Y’ contains all unique values or subtensors of the input. The second optional output tensor ‘indices’ contains indices of ‘Y’ elements’ first occurance in ‘X’.. The third optional output tensor ‘inverse_indices’ contains, for elements of ‘X’, its corresponding indices in ‘Y’. “. The fourth optional output tensor ‘counts’ contains the count of each element of ‘Y’ in the input.
Outputs are either sorted in ascending order or optionally in the order of the first occurrence of the values in the input.
https://docs.scipy.org/doc/numpy/reference/generated/numpy.unique.html
- Example 1:
input_X = [2, 1, 1, 3, 4, 3] attribute_sorted = 0 attribute_axis = None output_Y = [2, 1, 3, 4] output_indices = [0, 1, 3, 4] output_inverse_indices = [0, 1, 1, 2, 3, 2] output_counts = [1, 2, 2, 1]
- Example 2:
input_X = [[1, 3], [2, 3]] attribute_sorted = 1 attribute_axis = None output_Y = [1, 2, 3] output_indices = [0, 2, 1] output_inverse_indices = [0, 2, 1, 2] output_counts = [1, 1, 2]
- Example 3:
input_X = [[1, 0, 0], [1, 0, 0], [2, 3, 4]] attribute_sorted = 1 attribute_axis = 0 output_Y = [[1, 0, 0], [2, 3, 4]] output_indices = [0, 2] output_inverse_indices = [0, 0, 1] output_counts = [2, 1]
- Example 4:
- input_x = [[[1., 1.], [0., 1.], [2., 1.], [0., 1.]],
[[1., 1.], [0., 1.], [2., 1.], [0., 1.]]]
attribute_sorted = 1 attribute_axis = 1
intermediate data are presented below for better understanding:
there are 4 subtensors sliced along axis 1 of input_x (shape = (2, 4, 2)): A: [[1, 1], [1, 1]],
[[0, 1], [0, 1]], [[2, 1], [2, 1]], [[0, 1], [0, 1]].
there are 3 unique subtensors: [[1, 1], [1, 1]], [[0, 1], [0, 1]], [[2, 1], [2, 1]].
sorted unique subtensors: B: [[0, 1], [0, 1]],
[[1, 1], [1, 1]], [[2, 1], [2, 1]].
output_Y is constructed from B: [[[0. 1.], [1. 1.], [2. 1.]],
[[0. 1.], [1. 1.], [2. 1.]]]
output_indices is to map from B to A: [1, 0, 2]
output_inverse_indices is to map from A to B: [1, 0, 2, 0]
output_counts = [2 1 1]
Attributes
axis: (Optional) The dimension to apply unique. If not specified, the unique elements of the flattened input are returned. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
sorted: (Optional) Whether to sort the unique elements in ascending order before returning as output. Must be one of 0, or 1 (default). Default value is
1.
Inputs
X (heterogeneous) - T: A N-D input tensor that is to be processed.
Outputs
Between 1 and 4 outputs.
Y (heterogeneous) - T: A tensor of the same type as ‘X’ containing all the unique values or subtensors sliced along a provided ‘axis’ in ‘X’, either sorted or maintained in the same order they occur in input ‘X’
indices (optional, heterogeneous) - tensor(int64): A 1-D INT64 tensor containing indices of ‘Y’ elements’ first occurance in ‘X’. When ‘axis’ is provided, it contains indices to subtensors in input ‘X’ on the ‘axis’. When ‘axis’ is not provided, it contains indices to values in the flattened input tensor.
inverse_indices (optional, heterogeneous) - tensor(int64): A 1-D INT64 tensor containing, for elements of ‘X’, its corresponding indices in ‘Y’. When ‘axis’ is provided, it contains indices to subtensors in output ‘Y’ on the ‘axis’. When ‘axis’ is not provided, it contains indices to values in output ‘Y’.
counts (optional, heterogeneous) - tensor(int64): A 1-D INT64 tensor containing the count of each element of ‘Y’ in input ‘X’
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input can be of any tensor type.
OnnxUnique_11#
- class mlprodict.npy.xop_auto_import_.OnnxUnique_11(*args, **kwargs)#
Version
name: Unique (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Find the unique elements of a tensor. When an optional attribute ‘axis’ is provided, unique subtensors sliced along the ‘axis’ are returned. Otherwise the input tensor is flattened and unique values of the flattened tensor are returned.
This operator returns the unique values or sliced unique subtensors of the input tensor and three optional outputs. The first output tensor ‘Y’ contains all unique values or subtensors of the input. The second optional output tensor ‘indices’ contains indices of ‘Y’ elements’ first occurance in ‘X’.. The third optional output tensor ‘inverse_indices’ contains, for elements of ‘X’, its corresponding indices in ‘Y’. “. The fourth optional output tensor ‘counts’ contains the count of each element of ‘Y’ in the input.
Outputs are either sorted in ascending order or optionally in the order of the first occurrence of the values in the input.
https://docs.scipy.org/doc/numpy/reference/generated/numpy.unique.html
- Example 1:
input_X = [2, 1, 1, 3, 4, 3] attribute_sorted = 0 attribute_axis = None output_Y = [2, 1, 3, 4] output_indices = [0, 1, 3, 4] output_inverse_indices = [0, 1, 1, 2, 3, 2] output_counts = [1, 2, 2, 1]
- Example 2:
input_X = [[1, 3], [2, 3]] attribute_sorted = 1 attribute_axis = None output_Y = [1, 2, 3] output_indices = [0, 2, 1] output_inverse_indices = [0, 2, 1, 2] output_counts = [1, 1, 2]
- Example 3:
input_X = [[1, 0, 0], [1, 0, 0], [2, 3, 4]] attribute_sorted = 1 attribute_axis = 0 output_Y = [[1, 0, 0], [2, 3, 4]] output_indices = [0, 2] output_inverse_indices = [0, 0, 1] output_counts = [2, 1]
- Example 4:
- input_x = [[[1., 1.], [0., 1.], [2., 1.], [0., 1.]],
[[1., 1.], [0., 1.], [2., 1.], [0., 1.]]]
attribute_sorted = 1 attribute_axis = 1
intermediate data are presented below for better understanding:
there are 4 subtensors sliced along axis 1 of input_x (shape = (2, 4, 2)): A: [[1, 1], [1, 1]],
[[0, 1], [0, 1]], [[2, 1], [2, 1]], [[0, 1], [0, 1]].
there are 3 unique subtensors: [[1, 1], [1, 1]], [[0, 1], [0, 1]], [[2, 1], [2, 1]].
sorted unique subtensors: B: [[0, 1], [0, 1]],
[[1, 1], [1, 1]], [[2, 1], [2, 1]].
output_Y is constructed from B: [[[0. 1.], [1. 1.], [2. 1.]],
[[0. 1.], [1. 1.], [2. 1.]]]
output_indices is to map from B to A: [1, 0, 2]
output_inverse_indices is to map from A to B: [1, 0, 2, 0]
output_counts = [2 1 1]
Attributes
axis: (Optional) The dimension to apply unique. If not specified, the unique elements of the flattened input are returned. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(input).
sorted: (Optional) Whether to sort the unique elements in ascending order before returning as output. Must be one of 0, or 1 (default). Default value is
1.
Inputs
X (heterogeneous) - T: A N-D input tensor that is to be processed.
Outputs
Between 1 and 4 outputs.
Y (heterogeneous) - T: A tensor of the same type as ‘X’ containing all the unique values or subtensors sliced along a provided ‘axis’ in ‘X’, either sorted or maintained in the same order they occur in input ‘X’
indices (optional, heterogeneous) - tensor(int64): A 1-D INT64 tensor containing indices of ‘Y’ elements’ first occurance in ‘X’. When ‘axis’ is provided, it contains indices to subtensors in input ‘X’ on the ‘axis’. When ‘axis’ is not provided, it contains indices to values in the flattened input tensor.
inverse_indices (optional, heterogeneous) - tensor(int64): A 1-D INT64 tensor containing, for elements of ‘X’, its corresponding indices in ‘Y’. When ‘axis’ is provided, it contains indices to subtensors in output ‘Y’ on the ‘axis’. When ‘axis’ is not provided, it contains indices to values in output ‘Y’.
counts (optional, heterogeneous) - tensor(int64): A 1-D INT64 tensor containing the count of each element of ‘Y’ in input ‘X’
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Input can be of any tensor type.
OnnxUnsqueeze#
- class mlprodict.npy.xop_auto_import_.OnnxUnsqueeze(*args, **kwargs)#
Version
name: Unsqueeze (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Insert single-dimensional entries to the shape of an input tensor (data). Takes one required input axes - which contains a list of dimension indices and this operator will insert a dimension of value 1 into the corresponding index of the output tensor (expanded).
- For example:
Given an input tensor (data) of shape [3, 4, 5], then Unsqueeze(data, axes=[0, 4]) outputs a tensor (expanded) containing same data as data but with shape [1, 3, 4, 5, 1].
The input axes should not contain any duplicate entries. It is an error if it contains duplicates. The rank of the output tensor (output_rank) is the rank of the input tensor (data) plus the number of values in axes. Each value in axes should be within the (inclusive) range [-output_rank , output_rank - 1]. The order of values in axes does not matter and can come in any order.
Inputs
data (heterogeneous) - T: Original tensor
axes (heterogeneous) - tensor(int64): List of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
Outputs
expanded (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxUnsqueeze_1#
- class mlprodict.npy.xop_auto_import_.OnnxUnsqueeze_1(*args, **kwargs)#
Version
name: Unsqueeze (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Insert single-dimensional entries to the shape of a tensor. Takes one required argument axes, a list of dimensions that will be inserted. Dimension indices in axes are as seen in the output tensor. For example:
Given a tensor such that tensor with shape [3, 4, 5], then Unsqueeze(tensor, axes=[0, 4]) has shape [1, 3, 4, 5, 1]
Attributes
axes (required): List of non-negative integers, indicate the dimensions to be inserted
Inputs
data (heterogeneous) - T: Original tensor
Outputs
expanded (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxUnsqueeze_11#
- class mlprodict.npy.xop_auto_import_.OnnxUnsqueeze_11(*args, **kwargs)#
Version
name: Unsqueeze (GitHub)
domain: main
since_version: 11
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 11.
Summary
Insert single-dimensional entries to the shape of an input tensor (data). Takes one required argument axes - which contains a list of dimension indices and this operator will insert a dimension of value 1 into the corresponding index of the output tensor (expanded).
- For example:
Given an input tensor (data) of shape [3, 4, 5], then Unsqueeze(data, axes=[0, 4]) outputs a tensor (expanded) containing same data as data but with shape [1, 3, 4, 5, 1].
The attribute axes should not contain any duplicate entries. It is an error if it contains duplicates. The rank of the output tensor (output_rank) is the rank of the input tensor (data) plus the number of values in axes. Each value in axes should be within the (inclusive) range [-output_rank , output_rank - 1]. The order of values in axes does not matter and can come in any order.
Attributes
axes (required): List of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
Inputs
data (heterogeneous) - T: Original tensor
Outputs
expanded (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxUnsqueeze_13#
- class mlprodict.npy.xop_auto_import_.OnnxUnsqueeze_13(*args, **kwargs)#
Version
name: Unsqueeze (GitHub)
domain: main
since_version: 13
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 13.
Summary
Insert single-dimensional entries to the shape of an input tensor (data). Takes one required input axes - which contains a list of dimension indices and this operator will insert a dimension of value 1 into the corresponding index of the output tensor (expanded).
- For example:
Given an input tensor (data) of shape [3, 4, 5], then Unsqueeze(data, axes=[0, 4]) outputs a tensor (expanded) containing same data as data but with shape [1, 3, 4, 5, 1].
The input axes should not contain any duplicate entries. It is an error if it contains duplicates. The rank of the output tensor (output_rank) is the rank of the input tensor (data) plus the number of values in axes. Each value in axes should be within the (inclusive) range [-output_rank , output_rank - 1]. The order of values in axes does not matter and can come in any order.
Inputs
data (heterogeneous) - T: Original tensor
axes (heterogeneous) - tensor(int64): List of integers indicating the dimensions to be inserted. Negative value means counting dimensions from the back. Accepted range is [-r, r-1] where r = rank(expanded).
Outputs
expanded (heterogeneous) - T: Reshaped tensor with same data as input.
Type Constraints
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxUpsample#
- class mlprodict.npy.xop_auto_import_.OnnxUpsample(*args, **kwargs)#
Version
name: Upsample (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been deprecated since version 10.
Summary
Upsample the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * scale).
Attributes
mode: Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc) Default value is
'nearest'.
Inputs
X (heterogeneous) - T: N-D tensor
scales (heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of ‘scales’ should be the same as the rank of input ‘X’.
Outputs
Y (heterogeneous) - T: N-D tensor after resizing
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
OnnxUpsample_10#
- class mlprodict.npy.xop_auto_import_.OnnxUpsample_10(*args, **kwargs)#
Version
name: Upsample (GitHub)
domain: main
since_version: 10
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been deprecated since version 10.
Summary
Upsample the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * scale).
Attributes
mode: Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc) Default value is
'nearest'.
Inputs
X (heterogeneous) - T: N-D tensor
scales (heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of ‘scales’ should be the same as the rank of input ‘X’.
Outputs
Y (heterogeneous) - T: N-D tensor after resizing
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
OnnxUpsample_7#
- class mlprodict.npy.xop_auto_import_.OnnxUpsample_7(*args, **kwargs)#
Version
name: Upsample (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Upsample the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * scale).
Attributes
mode: Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc) Default value is
'nearest'.scales (required): The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of ‘scales’ should be the same as the rank of input ‘X’.
Inputs
X (heterogeneous) - T: N-D tensor
Outputs
Y (heterogeneous) - T: N-D tensor after resizing
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxUpsample_9#
- class mlprodict.npy.xop_auto_import_.OnnxUpsample_9(*args, **kwargs)#
Version
name: Upsample (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Upsample the input tensor. Each dimension value of the output tensor is:
output_dimension = floor(input_dimension * scale).
Attributes
mode: Two interpolation modes: nearest (default), and linear (including bilinear, trilinear, etc) Default value is
'nearest'.
Inputs
X (heterogeneous) - T: N-D tensor
scales (heterogeneous) - tensor(float): The scale array along each dimension. It takes value greater than or equal to 1. The number of elements of ‘scales’ should be the same as the rank of input ‘X’.
Outputs
Y (heterogeneous) - T: N-D tensor after resizing
Type Constraints
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input ‘X’ and output ‘Y’ to all tensor types.
OnnxWhere#
- class mlprodict.npy.xop_auto_import_.OnnxWhere(*args, **kwargs)#
Version
name: Where (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Return elements, either from X or Y, depending on condition. Where behaves like [numpy.where](https://docs.scipy.org/doc/numpy/reference/generated/numpy.where.html) with three parameters.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
History - Version 16 adds bfloat16 to the types allowed (for the second and third parameter).
Inputs
condition (heterogeneous) - B: When True (nonzero), yield X, otherwise yield Y
X (heterogeneous) - T: values selected at indices where condition is True
Y (heterogeneous) - T: values selected at indices where condition is False
Outputs
output (heterogeneous) - T: Tensor of shape equal to the broadcasted shape of condition, X, and Y.
Type Constraints
B in ( tensor(bool) ): Constrain to boolean tensors.
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types (including bfloat).
OnnxWhere_16#
- class mlprodict.npy.xop_auto_import_.OnnxWhere_16(*args, **kwargs)#
Version
name: Where (GitHub)
domain: main
since_version: 16
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 16.
Summary
Return elements, either from X or Y, depending on condition. Where behaves like [numpy.where](https://docs.scipy.org/doc/numpy/reference/generated/numpy.where.html) with three parameters.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
History - Version 16 adds bfloat16 to the types allowed (for the second and third parameter).
Inputs
condition (heterogeneous) - B: When True (nonzero), yield X, otherwise yield Y
X (heterogeneous) - T: values selected at indices where condition is True
Y (heterogeneous) - T: values selected at indices where condition is False
Outputs
output (heterogeneous) - T: Tensor of shape equal to the broadcasted shape of condition, X, and Y.
Type Constraints
B in ( tensor(bool) ): Constrain to boolean tensors.
T in ( tensor(bfloat16), tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types (including bfloat).
OnnxWhere_9#
- class mlprodict.npy.xop_auto_import_.OnnxWhere_9(*args, **kwargs)#
Version
name: Where (GitHub)
domain: main
since_version: 9
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 9.
Summary
Return elements, either from X or Y, depending on condition. Where behaves like [numpy.where](https://docs.scipy.org/doc/numpy/reference/generated/numpy.where.html) with three parameters.
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
condition (heterogeneous) - B: When True (nonzero), yield X, otherwise yield Y
X (heterogeneous) - T: values selected at indices where condition is True
Y (heterogeneous) - T: values selected at indices where condition is False
Outputs
output (heterogeneous) - T: Tensor of shape equal to the broadcasted shape of condition, X, and Y.
Type Constraints
B in ( tensor(bool) ): Constrain to boolean tensors.
T in ( tensor(bool), tensor(complex128), tensor(complex64), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(string), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8) ): Constrain input and output types to all tensor types.
OnnxXor#
- class mlprodict.npy.xop_auto_import_.OnnxXor(*args, **kwargs)#
Version
name: Xor (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the xor logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxXor_1#
- class mlprodict.npy.xop_auto_import_.OnnxXor_1(*args, **kwargs)#
Version
name: Xor (GitHub)
domain: main
since_version: 1
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 1.
Summary
Returns the tensor resulted from performing the xor logical operation elementwise on the input tensors A and B.
If broadcasting is enabled, the right-hand-side argument will be broadcasted to match the shape of left-hand-side argument. See the doc of Add for a detailed description of the broadcasting rules.
Attributes
axis: If set, defines the broadcast dimensions.
broadcast: Enable broadcasting Default value is
0.
Inputs
A (heterogeneous) - T: Left input tensor for the logical operator.
B (heterogeneous) - T: Right input tensor for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.
OnnxXor_7#
- class mlprodict.npy.xop_auto_import_.OnnxXor_7(*args, **kwargs)#
Version
name: Xor (GitHub)
domain: main
since_version: 7
function: False
support_level: SupportType.COMMON
shape inference: True
This version of the operator has been available since version 7.
Summary
Returns the tensor resulted from performing the xor logical operation elementwise on the input tensors A and B (with Numpy-style broadcasting support).
This operator supports multidirectional (i.e., Numpy-style) broadcasting; for more details please check Broadcasting in ONNX.
Inputs
A (heterogeneous) - T: First input operand for the logical operator.
B (heterogeneous) - T: Second input operand for the logical operator.
Outputs
C (heterogeneous) - T1: Result tensor.
Type Constraints
T in ( tensor(bool) ): Constrain input to boolean tensor.
T1 in ( tensor(bool) ): Constrain output to boolean tensor.