sglang
https://github.com/sgl-project/sglang
Python
SGLang is a high-performance serving framework for large language models and multimodal models.
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported1 Subscribers
Add a CodeTriage badge to sglang
Help out
- Issues
- [Perf] [AMD] mla: Fuse the split combine, W_vc and the output gate for Kimi-K3 decode on gfx950
- [Perf] [AMD] kimi-k3: TP8 LL column-parallel latent tail fused with the next attention read on gfx950
- [Perf] [AMD] kimi-k3: Fuse the attention output's TP8 all-reduce with the FFN-side read on gfx950
- [Perf] [AMD] kimi-k3: Fused router top-k + MoE route sort and zero-copy MoE output for decode on gfx950
- [Perf] [AMD] kimi-k3: Column-parallel MoE front and route-sort bank partials for decode on gfx950
- [Perf] [AMD] mla: One-launch q_a / kv_a norms and q_b projection for Kimi-K3 decode on gfx950
- [Refactor] memory-pool: Describe DeepSeek V4 device pools with buffer info
- [CI] Docker: reuse Hugging Face kernel download cache
- [Bug] NGRAM speculative decoding corrupts radix-cached KV with the overlap scheduler when requests end on a stop token (0.5.22)
- [Bug] NVFP4 Marlin fallback (SM80-SM90) garbles every output when the block scales are small (vllm-project/vllm#34577 not ported)
- Docs
- Python not yet supported