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 supported61 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Attention] Conditional device/CPU query-lens mismatch support for FlashInfer XQA (SM120/121)
- [Bug][XPU] GDN fp16 overflow reproduces on Intel XPU (Qwen3.5-family): long-context needle 0/9, and the pending #54146/#62119 fixes do not reach the XPU path
- feat(kv-cache): expose block allocation generations
- [ROCm] Only run AITER import-time setup on ROCm
- [Feature][Spec Decode] Add GQA DSpark prefill context-KV-only loading
- [Kernel][Sampler] Replace Gaussian pivot with order statistic in combined top-k/top-p Triton kernel
- [Performance]: RoutedExperts.load_weights rebuilds and re-indexes the expert mapping on every call (~50 s on Qwen3.5-35B-A3B with runai_streamer)
- [Bug] Qwen3.8-Flash-Next on GB10 (v0.32 nightly): prompt content beyond ~indexer_budget (2048) tokens is invisible — model answers as if the prompt ended at 2K
- [ROCm][Perf] Fuse AITER split-K GEMM reduce into add + RMSNorm
- [Bug]: --attention-backend b12x cannot serve hybrid attention/mamba models on v0.32.0 — hybrid page alignment inflates the block size to 2176, B12X only supports (64, 128)
- Docs
- Python not yet supported