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 supported51 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Feature]: Expose CPU vs P2P attribution for multi-tier KV restores
- [RFC]: Optimization: Take over a partial-hit block instead of CoW copy when ref_cnt == 1
- [Rust Frontend] Add OpenAI Responses API endpoint
- [Kernel][ROCm] Partition the cached-context scan for short-query attention (spec-decode verify)
- [Kernel][ROCm] Flash-decoding sequence partitioning for the Triton paged decode path
- [Bugfix][Frontend] Accept thinking as enable_thinking alias
- [Bug][DSV4-Flash][DSpark] Startup dies in kernel warmup with "ragged spec decode batch: 7 tokens over 2 requests"
- [Bug][Frontend]: Anthropic system-turn handling (merge or in-place) breaks prefix caching for clients that append a system turn every turn; in-place re-role to user satisfies all constraints
- fix(spec_decode): thread spec_step_idx in llm_base_proposer for multi-layer MTP (#52688)
- Fix hybrid mamba running state block size seeding
- Docs
- Python not yet supported