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
- [Feat] tracing: Record detokenizer time on root spans
- [Feat] logging: Add flag to redact request tokens from logs
- [Fix] deepseek_v4: route unquantized BF16/F16/F32 checkpoints off the MXFP4 expert path
- [Fix] MiMo-V2: Normalize images with the checkpoint's image_mean/image_std
- [Perf] [AMD] GLM5: Skip redundant -inf pre-fill of KPool MQA-logits
- [Fix] qwen2_moe: Replicate the shared expert under DWDP
- [Perf] [NVIDIA] attention: Add opt-in table-gated CUDA verify for Frozen-KV MTP
- [Fix] muse: recover channel headers after unframed prose
- [Bug] `update_weights_from_disk` after `release_memory_occupation` is accepted and crashes the scheduler (segfault in `weight_loader`)
- [Bug] With Speculative decoding + abort, completion can return more tokens than `max_tokens`
- Docs
- Python not yet supported