neural_compressor.adaptor.tf_utils.graph_rewriter.int8.meta_op_optimizer
Meta OP Graph Rewriter.
Module Contents
Classes
Fuse the pattern like Dequantize + MetaOp + Quantize into MetaOp(set its type to int8). |
- class neural_compressor.adaptor.tf_utils.graph_rewriter.int8.meta_op_optimizer.MetaInfoChangingMemOpOptimizer(model)[source]
Fuse the pattern like Dequantize + MetaOp + Quantize into MetaOp(set its type to int8).
With such changes, the Quantize and Dequantize OP will removed for better performance.