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 supported51 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [BugFix][ROCM] DFlash2 fix sliding-window prefix attention NaNs
- [Bugfix][Model] Gemma4 MM: fall back gracefully when torch.accelerator has no device (fixes TPU/JAX vision crash)
- [Test] Add a structured-outputs x tool-calling gate
- [Bugfix][Tool Parser] Fix Hermes streaming args (brace-match) + spec-decode x tool-calling gate
- [WIP][PCP] Dispatch CUDA graphs by rank-local token count
- [Bugfix] Make speculative-decode logprob test tie-aware
- [Bugfix] Handle excluded lm_head re-tying and unavailable cast weights
- [Bugfix] Don't route TOWER/CONNECTOR LoRA mappings to the language wrapper
- [Bug]: FULL cudagraph dummy runs set seq_lens > max_model_len, overrunning the block table (TRITON_ATTN illegal memory access; FLEX_ATTENTION shape error)
- [Spec Decode] Let the speculator build its own draft prefill attention metadata
- Docs
- Python not yet supported