TbeBatchMatMulQuantFusionPass

Description

Performs UB fusion on BatchMatMul/GEMM and AscendDequant/AscendQuant/Elemwise in the following pattern subgraphs.

Mode 1:

Mode 2:

Restrictions

  • BatchMatMul can be MatMul, MatMulV2, BatchMatMul, and BatchMatMulV2.
  • The dynamic shape scenario is not supported.
  • The Elemwise node must be FastGeluV2.

Availability

Atlas 200/300/500 Inference Product

Atlas Training Series Product