[https://github.com/intel/neural-compressor/blob/master/docs/source/validated_model_list.md/#pytorch-models-with-torch-201cpu-in-woq-mode](url) shows the accuracy of `int4` compared to `fp32`. Is there any data about 4-bit floating point numbers (e.g. `nf4`, `fp4`, etc.) and their performance data? THANKS