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 supported60 Subscribers
Add a CodeTriage badge to vllm
Help out
- Issues
- [RFC]: [V1 Scheduler][KV Offload] Eliminate 1.0–1.8s Re-Queueing Stall for Completed Async KV Pulls via Front-of-Queue Re-Admission
- [CI][Bugfix] Ignore soft-failed jobs in `/ci retry` on a new commit
- [Beam] upd flashinfer api
- [Performance discussion] When do larger CUDA graph capture sizes help?
- [KV Offload] Warn when secondary-tier KV is tied to the parallel config
- [Performance]: Long-prefill decode starvation survives prefill/decode interleaving — head-to-head from two stacks on 2× GB10, plus a 4×98k-prefill OOM boundary
- [KV Offload] Add valkey secondary tier
- [Perf][DiffusionGemma] Canvas packing + canvas swap: up to 10x lower short-request latency under load
- [Bug]: float32 models: Triton attention, prefill attention and fused-MoE kernels compute `tl.dot` in TF32 (~1e-2 error relative to output RMS instead of ~1e-5)
- [Usage]: cu129-nightly
- Docs
- Python not yet supported