V100RequantFusionPass
Description
Optimizes the quantization nodes in the inference scenario.
Insert the RequantHostCpuOpV2 operator into the input of AscendDequant based on the following structures.
Scenario 1
Scenario 2
Scenario 3
Restrictions
If there are multiple AscendDequant operators, the scale values of all AscendDequant operators must be the same.
Availability
Parent topic: Graph Fusion Patterns