[New Model]: add unsloth/MiniMax-M2-GGUF support

[kaikuo]

The model to consider.

Can you add support for unsloth/MiniMax-M2-GGUF? Both UD-Q4-K-XL and Q4_K_M is preferred.

The closest model vllm already supports.

MiniMax/MiniMax-M2 has already supported
many GGUF models also supported.

What's your difficulty of supporting the model you want?

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[jeejeelee]

Have you tried the main branch, also cc @Isotr0py

[Isotr0py]

Our GGUF support is really relying on upstream transformers' experimental interoperability with GGUF (it only supports some old models and single checkpoint file: https://github.com/huggingface/transformers/blob/c01e711ee57e70af4fbf2d056411503ce49c80ec/src/transformers/integrations/ggml.py#L36-L291).

I think we should decouple from it to support more newer and larger (sharded) models 😅