vllm
https://github.com/vllm-project/vllm
Python
A high-throughput and memory-efficient inference and serving engine for LLMs
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 supported35 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- Preserve DeepSeek sigmoid grouped routing metadata
- Account scheduled spec slots on empty output rows
- [Bugfix][Frontend] Return tool_calls finish reason from derendered tool calls
- [Bug]: XPU: INC dequantizes int4 MoE experts to bf16 (OOM), which presents as an empty_cache livelock
- [Performance]: How to set allow_hybrid_mode in DeepEP V2
- [Bug]: vllm 校验arguments合法性
- [Perf] SM120 PCIe serving stack: SP/async-TP enablement, FlashInfer spec-decode FULL cudagraphs, and PCIe-safe multi-GPU comms
- Support attention-HMA PP prefill in NixlPushConnector
- [Bug]: flashinfer_b12x MoE broken after PR #43328: "num_tokens exceeds max_num_tokens" with dp_size>1 and OOM in B12xMoEWrapper buffer allocation
- [Bug] MiniMax-M3 tool-call parser buffers arguments — no incremental streaming
- Docs
- Python not yet supported