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
- [Bug] DFlash and EAGLE-3 drafts build their layers without layer names, so a draft's quantization exclusions never match
- [Perf] rust/server: Avoid empty-header and batch-record allocations
- [Bug] TorchNativeLlamaForCausalLM / TorchNativePhi3ForCausalLM build plain torch.nn.Linear layers, so a quantized checkpoint's weights load without their scales
- [Perf] Qwen4-Exp: default --boundary-reduction to reduce-scatter under DP attention
- [Bug] SGLANG_ENABLE_NV_MEGA_RUNTIME_OPTIMIZATION: post-verify CUDA graph capture crashes on first sampled request (torch.compile JIT inside capture via flashinfer top_k)
- [Fix] io_struct: Fix `/generate` Http Native API behavior for the requests passing empty list element in `input_ids` parameter
- [Chore] ci: Update test est_time values
- [Feat] gemma4-diffusion: Support quantized checkpoints (FP8 and NVFP4)
- [Fix] dflash/eagle3: Build draft layers under their checkpoint names
- [Fix] ollama: Normalize Ollama options and propagate length done_reason in OllamaServing
- Docs
- Python not yet supported