com.microsoft - QEmbedLayerNormalization#

QEmbedLayerNormalization - 1#

Version

This version of the operator has been available since version 1 of domain com.microsoft.

Summary

Attributes

  • epsilon - FLOAT : The epsilon value to use to avoid division by zero.

Inputs

  • input_ids (heterogeneous) - T1:

  • segment_ids (optional, heterogeneous) - T1:

  • word_embedding_quant (heterogeneous) - T2:

  • position_embedding_quant (heterogeneous) - T2:

  • segment_embedding (optional, heterogeneous) - T2:

  • gamma_quant (heterogeneous) - T2:

  • beta_quant (heterogeneous) - T2:

  • mask (optional, heterogeneous) - T1:

  • word_embedding_scale (heterogeneous) - T:

  • position_embedding_scale (heterogeneous) - T:

  • segment_embedding_scale (optional, heterogeneous) - T:

  • gamma_scale (heterogeneous) - T:

  • beta_scale (heterogeneous) - T:

  • word_embedding_zero_point (heterogeneous) - T2:

  • position_embedding_zero_point (heterogeneous) - T2:

  • segment_embedding_zero_point (optional, heterogeneous) - T2:

  • gamma_zero_point (heterogeneous) - T2:

  • beta_zero_point (heterogeneous) - T2:

Outputs

  • layernorm_out (heterogeneous) - T:

  • mask_index_out (heterogeneous) - T1:

Type Constraints

  • T1 in ( tensor(int32) ): Constrain mask index to integer types

  • T2 in ( tensor(int8), tensor(uint8) ): Constrain input and output types to int8 tensors.

  • T in ( tensor(float) ): Constrain input and output types to float32 tensors.

Examples