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
- Lightning perf pr
- [Bug]: compressed-tensors MXFP4 MoE (`CutlassExpertsMxfp4`) crashes with an illegal memory access on GB200 (SM100) on main
- [Rust Frontend] packed_top_k_logprobs on raw generate
- [Bugfix] Fix DeepSeek-V4 FP8 O-projection crash with a forced --linear-backend
- [Bugfix] Fix ExllamaLinearKernel wrong output for group sizes such as 16 and 96
- [Experimental][CPU][MRV2] Run Model Runner V2 on CPU without Triton (custom sampler class + triton to numba transpilation)
- [Bug]: Exception on a non-output TP rank is logged and the worker dequeues the next RPC, so ranks desync and collectives mispair (cross-node deadlock)
- WIP [RFC]: SupportsReload, a single post-load contract and reload verifier (follow-up to #59502)
- [RFC]: Peer-device expert tier: run a subset of routed experts on a second GPU in the same host (out-of-tree plugin first, then a small seam in RoutedExperts)
- [Bugfix][CPU] Handle strided tensors in GELU-tanh
- Docs
- Python not yet supported