[Feature] a new model adapter to speed up many models inference performance on Intel CPU

[a3213105]

Hi guys，
We have developed an opensource solution to speedup LLM inference on CPUs （especially on XEON）.
This is our repo https://github.com/intel/xFasterTransformer .
Now we have supported many models, and more will be supported soon.
We see that FastChat is widely used, so we wanted to enable our solution in FastChat.
We will submit a new adapter to leverage our solution and so that end users can get better inference performance when their use CPUs.
Can I submit a pull request directly? Or do we need to do some process?

[merrymercy]

Sure. You can submit a PR directly.
Some examples are ExLlama integration (#2455) and vLLM worker (https://github.com/lm-sys/FastChat/blob/main/fastchat/serve/vllm_worker.py)

[a3213105]

Thanks， I will prepare the PR :)

Sure. You can submit a PR directly. Some examples are ExLlama integration (#2455) and vLLM worker (https://github.com/lm-sys/FastChat/blob/main/fastchat/serve/vllm_worker.py)

[a3213105]

Sure. You can submit a PR directly. Some examples are ExLlama integration (#2455) and vLLM worker (https://github.com/lm-sys/FastChat/blob/main/fastchat/serve/vllm_worker.py)

hi, I have committed a PR #2615 for this feature, thanks for your review.