[Feature] Deepseek-r1 / v3 awq deploy

### Motivation

Hello,

I am currently working with the DeepSeek-R1 and DeepSeek-V3 models and have encountered challenges due to their large size. I am exploring the possibility of using Activation-aware Weight Quantization (AWQ) to quantize the DeepSeek-R1 model, with the goal of deploying it efficiently using TurboMind.

I noticed that DeepSeek-V2 has already been successfully integrated with TurboMind, as referenced in [PR #3218](https://github.com/InternLM/lmdeploy/pull/3218). Given this precedent, I would like to inquire if similar support can be extended to the DeepSeek-R1 model.

Now I quanted the deepseek-r1 model by the [script](https://huggingface.co/cognitivecomputations/DeepSeek-V3-AWQ/discussions/3).

Specifically, I am interested in:
If TurboMind can be adapted to support the quantized DeepSeek-R1 model, similar to the DeepSeek-V2 implementation.

Any guidance or resources that could assist in achieving this integration.

### Related resources

_No response_

### Additional context

_No response_