com.microsoft - GreedySearch#
GreedySearch - 1 (com.microsoft)#
Version
name: GreedySearch (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
Greedy Search for text generation.
Attributes
decoder (required): Decoder subgraph to execute in a loop. Default value is
?
.decoder_start_token_id: The id of the token that indicates decoding starts. Default value is
?
.encoder: The 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 decoder only like GPT-2; 1 for encoder decoder like Bart 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 2 and 7 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)
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) - I: 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) - I: 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)
attention_mask (optional, heterogeneous) - I: Custom attention mask. Shape is (batch_size, sequence_length)
Outputs
sequences (heterogeneous) - I: Word IDs of generated sequences. Shape is (batch_size, max_sequence_length)
Examples