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
- [Spec Decode] Expose placeholder text for synthetic acceptance
- [Performance]: MRv2 token-logprob kernel reads the vocabulary twice and runs one program per row
- [Perf][MRv2] Read the vocabulary once in the token-logprob kernel, and split rows across programs
- [Perf][Spec Decode] Fused vocab-parallel argmax for LogitsProcessor.get_top_tokens()
- [Bugfix][Sampler] Stabilize standalone Triton top-p normalization
- [Perf][GDN] Reuse prefill lengths and fuse FlashInfer gate conversion
- docs(nixl): document metrics aggregation semantics across TP ranks (#41230)
- [Security] Fall back to eager when encoder CUDA graph replay exceeds capture
- [Core][NIXL] Transfer KV between different P/D KV cache group plans
- [Bug]: GLM-5.3-Flash NVFP4 on B300: CUDA illegal access persists with MTP disabled; blocking/eager reports persistent_topk
- Docs
- Python not yet supported