[FEATURE] Support BitBLAS Backend for QuantLinear

[LeiWang1999]

Hello everyone, we have recently published our library that supports mixed-precision BLAS operations on GPUs. This is the pull request to integrate it as an inference kernel in quantized linear with AutoGPTQ.

Hope this can help us to explore something fascinating.

The repo for BitBLAS is: https://github.com/microsoft/BitBLAS

And our benchmark results:

• End2End Integration with Quantize Inference Kernel for AutoGPTQ and vLLM.

[LeiWang1999]

CC @PanQiWei @qwopqwop200

[qwopqwop200]

Currently, this project is mainly managed by fxmarty.
@fxmarty

[LeiWang1999]

Thanks @Qubitium , Please cc @fxmarty