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] NVFP4 Marlin fallback (SM80-SM90) garbles every output when the block scales are small (vllm-project/vllm#34577 not ported)
- [Bug] Marlin MoE (AWQ W4A16) on SM86 intermittently returns wrong expert outputs, even for a single request; the Triton WNA16 kernel is exact
- [Bug] Client-controlled RID prefix allows cross-request abort through disconnect cleanup
- [Fix] CI: Preserve shared HF cache during incomplete retries
- [Bug] Data-plane API key accepted on mesh HA management routes in sgl-model-gateway (control-plane authorization bypass)
- [CI] rerun-test: Validate suites before dispatch
- [Bug] Client-supplied bootstrap address in Decode worker requests enables SSRF and blocks shared request processing
- [Fix] internvl: Load GPT-OSS backbone weights through GptOssForCausalLM.load_weights
- [Feature] Allow DWDP with pipeline parallelism (`--pp-size > 1`)
- [Fix] [AMD] moe: Keep the fake EP dispatch's zeroed correction bias on the aiter path
- Docs
- Python not yet supported