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
Add a CodeTriage badge to vllm
Help out
- Issues
- [Model] Use circular buffer for DeepSeek V4 C128 state
- [Metrics] Add priority-aware labels and metrics for priority scheduling
- [Refactor][Core] Use torch's device-agnostic streams and events in the V2 model runner
- Add MoE tuning config for RTX PRO 6000 Blackwell (E=512,N=512,fp8_w8a8)
- docs: clarify NIXL KV connector metrics aggregation semantics
- [Bug]: `--api-key` is bypassed by `/invocations`, unauthenticated inference on a server that requires a key
- [Bug]: Mistral's "always adjust_request" grammar path rebuilds the tool parser and compiles regex over the full vocab on the main event loop, on every request — even tool_choice="none"
- [Bug]: Qwen3 parser turns tool-call markup the model quotes into real tool calls (names outside request.tools, and duplicates of offered ones)
- [Rocm][Bugfix][Quantization] Auto-enable EP for Qwen3.8-Flash-Next-FP8 when FP8 MoE TP would fail block_n alignment
- [Kernel][ROCm] kv split decoding kernel for RDNA3/4
- Docs
- Python not yet supported