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
- [Fix] [AMD] quark: Fix the EP>1 crash in the Quark MXFP4 W4A16 MoE path and restrict it to gfx1151
- [Bug] Attention-DP ranks are indistinguishable to routers and dashboards: `KvMetrics.data_parallel_rank` is always 0 and Prometheus series omit `dp_rank`
- [Perf] deterministic: Launch a fixed number of Triton decode blocks for large split budgets
- [Fix] [AMD] deterministic: Run block-FP8 linears through Triton on ROCm
- [Fix] deterministic: Batch-invariant Kimi Delta Attention
- [Fix] [AMD] deterministic: Run FP8 MoE experts through the Triton runner on ROCm
- [Fix] metrics: Correct MFU/MBU bandwidth overcounting and GraniteMoE AutoTokenizer fallback
- [Perf] [NVIDIA] glm53-flash: Add opt-in KDA projection overlap on SM90
- [Perf] [AMD] DFlash: FP8 KV and tuned launch for draft SWA verify
- [Bug] benchmark: --dataset-name random silently requires the ShareGPT download; fails on offline/restricted networks
- Docs
- Python not yet supported