Windows build of release v0.24.0 for Python 3.12, CUDA 13 and Pytorch 2.11
Install the wheel with pip install vllm-0.24.0+cu132-cp312-cp312-win_amd64.whl --extra-index-url https://download.pytorch.org/whl/cu130
IMPORTANT
You need CUDA 13 to use the wheel.
-
The default wheel is built for Ampere (RTX 3090 / Ti) + ADA (RTX 4090) + Blackwell (RTX 5090 / 6000 PRO) support.
-
NCCL + Tensor / Pipeline parallelism for multi-gpu inference support on Windows.
Star the repo, give a like and share the announcement on LinkedIn or share on Reddit to help support the project.