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 supported50 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Bugfix][Frontend] Handle logprobs across streaming continuations
- [Bug]: Responses API executes only the final built-in tool call in a parallel tool-call turn
- [Bugfix][Multimodal] Do not expand prompt updates that are already applied
- [Bugfix][Model Runner V2] Separate Mamba aligned-metadata and state-copy contexts
- [Bug] Kimi-K3 TP8xPP2 + DSpark spec-on: mid-decode target-forward NaN detonates MLA decode under concurrency (fp8 latent-cache bytes carry the corruption)
- [Bugfix][Reasoning] End kimi_k3 reasoning at a response opener with no think markers
- [Bug]: Mamba-hybrid models (granite-4.0-h) give a different, garbled greedy answer every time for a ONE-token prompt under full CUDA graphs
- [Bug]: with a `--reasoning-parser` configured, structured outputs on `/v1/completions` are not enforced, with no warnings
- [Bug]: `POST /reset_prefix_cache` returns success while the CPU offload tier keeps serving the same blocks
- [Bug]: /tokenize returns 400 "cannot pickle ValidatorIterator" for assistant `content: null` + `tool_calls` (DeepSeek-V4.1), while chat completions accepts the same messages
- Docs
- Python not yet supported