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
- Silent deterministic output corruption: first token id 0 ("!") then [PAD] (id 163839) loop, spec accept-length pinned at 1.00, zero errors — Kimi-K3.1-Pro (hybrid mamba + DFlash2 draft) on AMD MI300X/AITER, fp8 KV, v0.5.23.dev20261104 (e67b2062e2)
- [Perf] ASR: Lazily encode cumulative WAV prefixes on demand in split_audio_chunks
- [Fix] [AMD] moe: Stop declaring zero-filled A8W4 intermediate padding
- [Perf] [AMD] distributed: Run QuickReduce ahead of the custom all-reduce from a size threshold
- [Perf] [AMD] dspark: Accept on target vocab shards in verify
- [Perf] [AMD] dsa: Prefill rope-free bf16 KV pools with aiter's OPUS sparse kernel
- [Perf] [AMD] deepseek-v4.1: Run the closed mHC seams through Mega-mHC in prefill
- [Perf] [AMD] deepseek-v4.1: Run the DSpark draft's mHC seams through Mega-mHC
- [Feat] metrics: measure router-to-detokenizer delivery latency
- [Fix] attention: Refuse fa4 for KDA hybrids on consumer Blackwell
- Docs
- Python not yet supported