Created by: wojtuss
The cpu_quantize_placement_pass
is a pass that is run after all graph optimization passes and before quantization passes to apply quantization strategy by setting the use_quantizer
attribute to true
for operators which are to be quantized.
This pass is a part of INT8v2 approach with the core implementation submitted in PR https://github.com/PaddlePaddle/Paddle/pull/15987.
test=develop