Open
Description
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