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 supported61 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Bug][Spec Decode] MTP draft from a draft-only checkpoint runs with an uninitialised embedding under pipeline parallelism (no error; acceptance collapses)
- [Bugfix][Core] Reserve only the blocks a spec decode step adds when admitting async KV loads
- [Attention] Run SM90 FP8-KV attention on FA4 native FP8
- [KV Connector] Recreate NIXL agents across sleep without UCX overrides
- [Model] Dory: per-request loop count via SamplingParams.extra_args
- [Spec Decode][DeepSeek-V4.1] Support DSpark with pipeline parallelism
- [Bugfix][Parser] Fix multi-token reasoning-to-tool handoff with repeated thought blocks
- [Bug]: Responses API: `reasoning.summary` is accepted and echoed, but reasoning items always come back with `summary: []`
- [Perf][GDN] Slice contiguous spec/non-spec token blocks in mixed GDN batches
- [Perf][GDN] Index mixed-batch spec rows once per step; slice when they lead
- Docs
- Python not yet supported