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
Add a CodeTriage badge to vllm
Help out
- Issues
- [2/N][Feat][Perf] Add new warmup infrastructure for JITs. Add predicate filtering for JIT warmup, and migrate Inkling FA4
- [Bug]: [Parser] kimi_k2 streaming tool-call args skip schema type coercion applied in non-streaming (silent type mismatch)
- [Bug]: [Performance] AsyncLLM `add_request` runs multimodal `process_inputs` (HF processor) synchronously on the event-loop thread, serializing per-request image preprocessing and starving the GPU under concurrent multi-image requests
- [Bugfix][Parser] kimi_k2: route streaming through arg_converter so schema type coercion applies
- [Bug]: logprob_token_ids: [] silently bypasses validation and is discarded, on both Chat Completions and Completions
- [Feature][MoE]W4A16 cold-expert CPU offload via in-graph HIP gather (experimental)
- [Bugfix][Kernel] Fixes issue #49290: MRotaryEmbedding Triton kernel hardcodes Neox pairing for GPT-J style models
- [Loader] Support prefetch in multi-thread safetensors loader
- [XPU][LoRA] Fix fused MoE LoRA shrink BLOCK_SIZE_N for small ranks
- [Frontend] Add --max-waiting-queue-length to bound the waiting queue
- Docs
- Python not yet supported