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
- [Feature]: nvfp4 KV cache on SM120 — flashinfer ships the kernels, vLLM isn't wired to them (working prototype, 245K ctx on a 5090)
- [Perf] Add per-phase cold-start startup span benchmark harness
- Make `load_weights` completely optional
- [Attention][MiniMax-M3] Add FlashInfer MSA backend for SM120/SM121
- [Bug]: XQA decode under FULL cudagraph capture silently corrupts attention output (fp8 and nvfp4 KV)
- [Bugfix] Propagate VLLM_MARLIN_INPUT_DTYPE into INC/AutoRound quant path
- [Doc] Correct gRPC and OpenTelemetry flag names in security and metrics docs
- Unexpected EngineCore death exits serving process 0, defeating Restart=on-failure
- [Bug]: LoRA on MLA kv_b_proj is silently never applied — decode bypasses it via absorption, prefill via a stale pre-wrap reference
- [Do not merge!] [Build] Migrate vendored DeepGEMM from pybind to TORCH_LIBRARY (abi3)
- Docs
- Python not yet supported