TbeBatchMatMulQuantFusionPass

Description

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

Pattern 1:

Pattern 2:

Constraints

  • BatchMatMul can be MatMul, MatMulV2, BatchMatMul, and BatchMatMulV2.
  • Dynamic shapes are not supported.
  • The Elemwise node must be FastGeluV2.

Applicable Products

Atlas 200I/500 A2 inference products

Atlas inference products

Atlas training products