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
- [Bug]: V1 rejection sampler reads draft_probs with the target's vocab width when the draft head is narrower (out-of-bounds reads, stray token ids)
- [Bug]: thinking_token_budget forces a second </think> when the natural </think> is inside accepted drafts (spec decode, V1 runner)
- [ROCm][Perf][MiniMax-M3] Defer the FFN all-reduce into the next layer's norm
- [KV Connector][NIXL] Transfer DeepSeek-V4.1 packed MLA rows whole
- [Bugfix] Preserve empty tensor metadata in packed IPC
- [Bugfix] Reject dynamic speculative decoding with ngram_gpu
- [Perf][MoE] Batch per-expert token count GPU sync into single transfer
- [Bugfix][Frontend] Honor default penalties in transcription and translation requests
- [Bugfix][Spec Decode] Apply min_tokens at the V1 bonus-token position
- [Perf][Distributed] Skip movedim+reshape for dim=0 in CudaCommunicator reduce_scatter
- Docs
- Python not yet supported