[New Model]: Support Zyphra/Zamba2-7B

[mgoin]

The model to consider.

Announcement blog: https://www.zyphra.com/post/zamba2-7b

Base model: https://huggingface.co/Zyphra/Zamba2-7B
Instruct tuned: https://huggingface.co/Zyphra/Zamba2-7B-Instruct

The closest model vllm already supports.

Jamba, as it is a mixture of state-space and transformers blocks

Zamba2-7B-Instruct is a hybrid model composed of state-space (Mamba2) and transformer blocks.

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

Should be easy once Mamba2 support lands in #9292, however this use_shared_attention_lora case seems possibly complex

All of the HF-compatible modeling code can be found here: https://github.com/Zyphra/transformers_zamba2/tree/main/src/transformers/models/zamba2

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.

[yaronr]

+1

[scwall]

+1

[engchina]

+1

[github-actions]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!

[yury-tokpanov]

Hey. Yury here from Zyphra. We have an internal version that works, going to open a PR sometime soon.

[yury-tokpanov]

This is done now. Merged into 0.8.x version of vLLM