Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

First SOTA ssm based model, vllm currently supports it but there is some parallel work in vllm to optimise it aswell

• Simplify Jamba state management vllm-project/vllm#7428
• [Kernel/Model] Migrate mamba_ssm and causal_conv1d kernels to vLLM vllm-project/vllm#7651

https://huggingface.co/collections/ai21labs/jamba-15-66c44befa474a917fcf55251

Related resources

vllm implementation
vllm-project/vllm#4115