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 supported62 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Bugfix][NIXL] Stop reading from a remote engine whose READ has stalled
- [Bug]: FlashInfer metadata builder races under async scheduling for non-speculative models since #57214 (CUDA illegal memory access / GPU hang)
- [Perf][Model] Qwen4Exp QSA DCP: opt-in sharded selector
- [HiSparse] GLM-5.3-Flash support
- [Bug]: min_tokens keeps EOS masked on the bonus token under speculative decoding (V1 runner), changes greedy output
- [Bug]: CPU backend: C++ rejection_random_sample kernel accepts -1 placeholder drafts, so structured-output requests can sample from an unmasked row
- [Perf][Attention] Cache FlashAttention KV cache transpose+split views
- [Core] Expose prefix reuse policy in KV cache group metadata
- [Bug]: CPU without Triton: slot-mapping fallback has a stale signature after #57169, V1 inference fails on the first forward step
- [Perf][Distributed] Skip movedim+reshape for dim=0 in CudaCommunicator all_gather
- Docs
- Python not yet supported