module onnxrt.ops_cpu.op_layer_normalization#

Inheritance diagram of mlprodict.onnxrt.ops_cpu.op_layer_normalization

Short summary#

module mlprodict.onnxrt.ops_cpu.op_layer_normalization

Runtime operator.

source on GitHub



truncated documentation


LayerNormalization ================== This is layer normalization defined in ONNX as function. The overall computation …



truncated documentation




truncated documentation


Returns the list of arguments as well as the list of parameters with the default values (close to the signature). …


Returns the list of modified parameters.


Returns the list of optional arguments.


Returns the list of optional arguments.


Returns all parameters in a dictionary.



truncated documentation




Runtime operator.

source on GitHub

class mlprodict.onnxrt.ops_cpu.op_layer_normalization.LayerNormalization(onnx_node, desc=None, **options)#

Bases: OpRun

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.


  • 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 nameaxisi-1typeINT (INT)

  • epsilon: The epsilon value to use to avoid division by zero. Default value is nameepsilonf9.999999747378752e-06typeFLOAT (FLOAT)

  • stash_type: Type of Mean and InvStdDev. This also specifies stage one’s computation precision. Default value is namestashtypei1typeINT (INT)


Between 2 and 3 inputs.

  • X (heterogeneous)T: Tensor to be normalized.

  • Scale (heterogeneous)T: Scale tensor.

  • B (optional, heterogeneous)T: Bias tensor.


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 tensor(float16), tensor(float), tensor(double), tensor(bfloat16): Constrain input types and output Y type to float tensors.

  • U tensor(float), tensor(bfloat16): Type of Mean and InvStdDev tensors.


Onnx name: LayerNormalization

This version of the operator has been available since version 17.

Runtime implementation: LayerNormalization

__init__(onnx_node, desc=None, **options)#
_run(X, Scale, B=None, attributes=None, verbose=0, fLOG=None)#

Should be overwritten.

source on GitHub

mlprodict.onnxrt.ops_cpu.op_layer_normalization._layer_normalization(X, W, B, axis=-1, epsilon=1e-05)#