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 supported36 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Security] Fix P2P ZMQ socket exhaustion via unbounded peer sessions
- fix: prevent NIXL Decode worker crash from multi-prompt completions
- [Kernel] Hoist tensor-descriptor build out of unified-attention TD ti…
- [UX] Emit warning if some decode batches are not covered by CUDA graphs
- [RFC]: Unify functional P2P gating — one veto-only verdict for all P2P consumers (NCCL, custom allreduce, symm-mem)
- MooncakeConnector P/D: NVLink fallback transfer fails "Requested address not found" (fp8 MLA, v0.25.0)
- DeepSeek-V4 (deepseek_v4): fused topk_softplus_sqrt router rejects non-standard expert counts on CUDA (REAP 144-expert ckpts); torch fallback is XPU-gated
- deepseek_v4: decode runs unfused (breakable CUDA graph) — DeepseekV4ForCausalLM lacks @support_torch_compile; fullgraph capture blocked by inline deep_gemm/tilelang pybinds
- [KV Connector][NIXL] Support prefix caching with heterogeneous P/D logical block sizes
- [Bug]: MooncakeConnector registers all workers as dp_rank=0 under multi-port external DP load balancing
- Docs
- Python not yet supported