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
- Fix fp8 KV cache scale loading in MiMoV2MTP
- [Benchmark] NVIDIA Confidential Computing CPU<->GPU bridge microbench
- [Bug]: InternLM2 tool parser (streaming) drops all content after `<|action_start|>` when no `<|plugin|>` follows
- [Bug]: `TensorizerConfig(lora_dir=...)` leaves `tensorizer_dir=None` and breaks serialization round trips
- [Bugfix][Tool Parser] Fix xLAM streaming args: deep nesting and braces in strings
- [RFC]: Bounded prefix reuse for target-token prefill scoring
- [Bug][XPU][MRV2]: engine dies at startup on the first PIECEWISE graph capture with TP=2 + MTP on a GDN/Mamba-hybrid model (oneCCL collective inside capture; regression from #56531; #58415 and local [10]/[11] patches do NOT fix it (see Update 2, 2026-10-10))
- [Perf][ROCm][Quantization] gfx942 MXFP8 linear for decode via AITER asm A16W8 GEMM
- [Attention] FlashInfer SM120 route for the sparse indexer's paged MQA logits
- [Bug]: vllm_compile_cache.py is rewritten in place, so concurrent cold starts sharing a cache can read or write a partial file
- Docs
- Python not yet supported