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 supported60 Subscribers
View all SubscribersAdd a CodeTriage badge to vllm
Help out
- Issues
- [Bug][ROCm] `ROCM_ATTN` returns wrong attention (and can fault the GPU) on RDNA for head size ≥ 96 with KV block size ≠ 16, and for MHA with head size 192/256
- [Bug]: DiffusionGemma prefills image tokens causally, although its config asks for bidirectional attention over them
- [Bugfix][Frontend] Accept Cohere v2 safety_mode OFF with the cmd3 format
- [Bugfix][Frontend] Honor X-Vllm-Priority in batch chat completions
- [ROCm][Kernel] Add split-K and shape-based tiles to Triton W4A16 on gfx950
- [Bugfix][Compile] Write vllm_compile_cache.py atomically
- [Perf][MRV2] Record FULL decode attention metadata builds in CUDA graphs
- [Performance]: Avoidable preemption at hybrid Attention/Mamba decode block transitions
- [Perf] [Sampling] - fix fallback and optimise fallback gather path for top-k in _topk_topp_kernel
- [Performance]: XPU block FP8: W8A16 instead of W8A8 took B70 TP4 single-stream decode from 35.27 to 57.87 tok/s
- Docs
- Python not yet supported