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
- [Core] Add a generation readiness check `/health_generate`
- [ROCm][Attention] Add an mxfp4_mla KV cache for rope-free sparse MLA (read/write path)
- [Performance][ROCm] KV connectors rule out ROCM_ATTN, so PD workers decode on ROCM_AITER_UNIFIED_ATTN (1.9–3.2x the per-token time for Qwen3 models on MI355X)
- [Bug]: `rank` for `logprob_token_ids` entries is their position in the request list, not the vocab rank
- [Bugfix][NIXL] Apply DCP config checks to NixlPullConnector
- [Bugfix][NIXL] Release partial remote descriptors when a handshake fails
- [Bugfix][NIXL] Drain handshakes before worker teardown
- [Bug]: Split top-p pipeline (_topp_sb_*) over-keeps tokens near the boundary, same stall class as #59804
- [Bug]: NIXL pull: one wedged connection repeatedly strands decode KV blocks (production log analysis)
- [Model Runner V2] Optional post-capture warmup of every decode batch size, logits row count and graph size
- Docs
- Python not yet supported