com.microsoft - QOrderedLayerNormalization#

QOrderedLayerNormalization - 1 (com.microsoft)#

Version

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

Summary

QOrderedLayerNormalization

Attributes

  • axis: The first normalization dimension: normalization will be performed along dimensions axis : rank(inputs). Default value is ?.

  • epsilon: The epsilon value to use to avoid division by zero. Default value is ?.

  • order_X: cublasLt order of input X. Default is ROW MAJOR. See the schema of QuantizeWithOrder for order definition. Default value is ?.

  • order_Y: cublasLt order of matrix Y, must be same as order_X. Default is ROW MAJOR. Default value is ?.

Inputs

  • X (heterogeneous) - Q: Input data tensor from the previous layer.

  • scale_X (heterogeneous) - S: scale of the quantized X

  • scale (heterogeneous) - F: Scale tensor, i.e., gamma vector.

  • B (optional, heterogeneous) - F: Bias tensor.

  • scale_Y (heterogeneous) - S: scale of the quantized X

Outputs

  • Y (heterogeneous) - Q: Output data tensor.

Examples