refactor TurbomindModelConfig

[lvhan028]

No description provided.

[irexyc]

Shoud we assert _group_size ==128?

[lvhan028]

Yes.

# Compatible to awq models that are quantized by lmdeploy (<=v0.3.0)
    if not group_size:
        group_size = 128

    if engine_config.model_format in ['awq', 'gptq']:
        assert group_size == 128, \
            f'model format is "{engine_config.model_format}" ' \
            f'but group_size is {group_size}. Currently, only 128 ' \
            'is supported'

[irexyc]

test /mnt/140/InternLM/internlm2_5-7b-chat convert, load from workspace and load from hf.