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] flashinfer: Add opt-in scoped caching for single-rank decode graphs
- [Feat] NPU: support shared-expert fusion across MoE A2A backends
- [Fix] process: Don't send SIGQUIT to PID 1 when the launcher dies during startup
- [Perf] [NVIDIA] diffusion/nomas: Fuse VAE QK normalization and MRoPE epilogue
- [Feat] scheduler: Add --disable-model-stop-tokens to disable model defaults stop tokens
- [Bug] Use-After-Free in SGLang UnifiedRadixCache Exposes Tenant KV Data
- [Bug] compressed-tensors: a shared expert kept unquantized in `ignore` is still fused into the quantized MoE (Qwen3-Next AWQ fails to load)
- [Bug] ModelOpt MIXED_PRECISION: layers with an unsupported quant_algo (FP8_PER_CHANNEL_PER_TOKEN) are built unquantized and load FP8 codes as BF16 values
- [Fix] speculative: Apply repetition penalty per draft position in EAGLE verify
- [Perf] Engine: freeze GC after startup, as the HTTP server does
- Docs
- Python not yet supported