support gguf_q4k_m / gguf_q4k_s

[rnwang04]

Description

1. Why the change?

https://github.com/analytics-zoo/nano/issues/1316#issuecomment-2076658639

2. User API changes

  model = AutoModelForCausalLM.from_pretrained(model_path,
                                               load_in_low_bit='gguf_q4k_m',
                                               optimize_model=True,
                                               torch_dtype=torch.float16,
                                               trust_remote_code=True,
                                               use_cache=True)

  model = AutoModelForCausalLM.from_pretrained(model_path,
                                               load_in_low_bit='gguf_q4k_s',
                                               optimize_model=True,
                                               torch_dtype=torch.float16,
                                               trust_remote_code=True,
                                               use_cache=True)

3. Summary of the change

• add q5k precision
• add gguf_q4k_m / gguf_q4k_s

4. How to test?

• [x] Local test
• [ ] Unit test
• [x] performance validation: https://github.com/analytics-zoo/nano/issues/1316#issuecomment-2078580860