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] mem_cache: Read KV pages from a peer engine over NVLink
- [Fix] dsv4/topk_v2: Move the cluster kernels' top-k buffer to dynamic shared memory
- [Perf] attention: Optimize FA3/FA4 tree attention
- [Fix] [Ascend] mHC: Use PyTorch fallback on NPU
- [Bug] ModelOpt MIXED_PRECISION on Llama / Mistral 3: fused qkv_proj and gate_up_proj are built unquantized (the model classes declare no packed_modules_mapping)
- [Fix] Llama, Llava: Declare packed_modules_mapping so ModelOpt MIXED_PRECISION resolves qkv_proj and gate_up_proj
- [Feat] scheduler: Expose scheduler idle state in server info
- [Bug] FP8 checkpoint whose lm_head and embed_tokens are stored in FP8 but listed in modules_to_not_convert: SGLang loads the raw FP8 codes as BF16 values
- [Fix] VocabParallelEmbedding: Refuse a float8 checkpoint weight in an unquantized embedding or lm_head
- [Test] attention: Cover Triton merge state with an unsplit oracle
- Docs
- Python not yet supported