Wenhua Cheng

Currently, we need to support each model individually, and the code lacks cohesion and uniformity.

Focus on WA quantization of MX_FP first refer to https://github.com/intel/neural-compressor/blob/master/neural_compressor/strategy/bayesian.py https://github.com/pytorch/ao/tree/main/torchao/quantization/prototype/mixed_precision

https://github.com/pytorch/ao/pull/870