Fixed FP8 fake quantization to use fp32 amax scaling; Added support for FP8 per-channel quantization#381

realAsma merged 2 commits intomainNVIDIA/TensorRT-Model-Optimizer:mainfrom

asma/fp8_fakequant_fixNVIDIA/TensorRT-Model-Optimizer:asma/fp8_fakequant_fixCopy head branch name to clipboard