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
- [Benchmark][Spec Decode] Build synthetic_acceptance_rates from a measured drafter
- [Bug]: Hybrid prefix caching: the Mamba "align" prompt-boundary state is freed mid-request and evicted before the request's attention blocks, so the next turn misses the whole prompt
- [Bugfix] Hybrid prefix caching: keep the Mamba prompt-boundary state until the request ends and evict it after its attention blocks
- [Performance]: Cold start on a fresh H100: -42.5% TTFT with overlapped weight staging and AOT reuse (Qwen3-8B-AWQ)
- [RFC]: use extensible token-aligned data format for aux output mooncake data
- [Bugfix] Stop draft-only spec decode queries under PP and clamp the EAGLE sample index
- [Rust Frontend] Support `--max-num-queued-reqs` and `--max-num-queued-tokens`
- [Bug]: Intel Arc Pro B70 XPU checkpoint-loading hang before readiness with Xe BCS faults on v0.31.0
- [Loader] Pinned host staging for the Run:ai streamer path
- [Reload] Keep captured-graph tensors valid across reload_weights after level-2 sleep (NVFP4 MoE + MLA)
- Docs
- Python not yet supported