MatmulCastFusionPass

Description

Fuses the MatMul/MatMulV2/BatchMatMul/BatchMatMulV2 and cast operators into the MatMul/MatMulV2/BatchMatMul/BatchMatMulV2 operator.

Before: After:

Restrictions

This fusion takes effect when the input data type of MatMul is float16 and the output data type of cast is float32.

Availability

Atlas 200/300/500 Inference Product

Atlas Training Series Product