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
- [Perf][Gemma4] Add fused Triton kernels for QKV norm/RoPE, pre/post-FFN norms, PLE, and MoE routing
- [Bug]: EPLB counts load for the wrong expert when step_interval < window_size
- [Perf][Gemma4] Fuse MTP speculative decoding layers and sparse gather-GEMV LM head
- [Feature][KVConnector][NIXL] NixlPushConnector inverts the transfer initiator but hides no latency: the KV is still sent as one batch after prefill
- [Bug]: Sweep server readiness can succeed after its startup timeout
- [Bugfix][Benchmark] Reject expired sweep readiness probes
- [Bugfix][CPU] Avoid Triton JIT stall in _ranks_kernel on CPU
- [Bugfix] Return vocab rank instead of list position for logprob_token_ids
- [Bugfix] Reject `--convert embed/classify` with non-pooling runners at config time
- [KV Offload] Explain missing chunks when prepare_load fails
- Docs
- Python not yet supported