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
- [Bug]: NIXL: `remove_remote_agent` does not release UCX endpoints, so each `engine_ttl` eviction + re-handshake leaks until handshakes fail
- [Frontend] Accept the model path that /v1/models publishes
- [Bugfix][Core] Release the profile run's allocator cache before a pinned KV cache
- [Bugfix][Core] Unblock scheduling after a failed async KV load
- [Core] Release the CUDA graph pool on sleep instead of backing it up
- [Bugfix][Frontend] Count prompt-opened Inkling reasoning tokens
- [Kernel] Add opt-in LiteTopK FP32 decode with producer histograms
- [DeepSeek-V4-P3] deterministic MoE DeepEP
- [Bug] DeepSeek-V4-Flash tool-call format failures associated with prefix-cache reuse; worst for short cached head + long uncached suffix; `cache_salt` reduces failures but is confounded with hit shape (v0.27.1, fp8 KV, MTP)
- [Bugfix] Bound prompt_embeds decode budget per request
- Docs
- Python not yet supported