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
- [Performance]: FlashInfer NVFP4 KV cache causal prefill is ~1.7-1.8x slower than FP8 on SM120
- [Rust Frontend] Propagate W3C trace headers to engine-core requests
- [kv_offload] Session Aware Eviction Policy
- [Bugfix] Stop the Granite reasoning marker leaking into streamed reasoning
- [Bug]: EAGLE embed_tokens sharing decision is made per-rank without cross-rank agreement — TP ranks build different drafters (rank0 keeps drafter embeddings, rank1 ties target's); acceptance collapses to ~0.45/draft
- [Bug]: Model Runner V2 + Gemma-4: memory profiler reports ~0.9 GiB more "available KV" than MRv1 → no runtime headroom → CUDA OOM under saturating load (no spec involved); with EAGLE it crashes earlier at cudagraph capture (stable-ABI aten::empty)
- [Doc]: Update Quickstart documentation with Apple Silicon / Metal execution requirements and compatibility note
- [Feature] Qwen3-Next (GDN): mamba_cache_mode="all" prefix caching with speculative decoding (MTP) on V1
- [Bugfix][Core][Spec Decode] Skip spec-decode block reservation on the external-KV-load step (P/D)
- [7/N][warmup][DSv4] Migrate router GEMM kernels
- Docs
- Python not yet supported