[Feature] Integration of TurboMind AWQ and GPTQ

[zhyncs]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

The AWQ and GPTQ of TurboMind should be among the best-performing open-source implementations currently available. We plan to integrate them into SGLang, and once the integration is complete, we can consider removing SGLang's dependency on vLLM's AWQ and GPTQ kernel.

During development, we can initially install the wheel https://github.com/InternLM/turbomind/releases/tag/v0.0.1 manually for verification and later add the TurboMind repo as a dependency in sgl-kernel.

ref
https://github.com/InternLM/turbomind

Related resources

No response

[Chen-0210]

Hi, I’d like to work on this issue.