v0.26.0

Windows build of release v0.26.0 for Python 3.12, CUDA 13 and Pytorch 2.11

Install the wheel with pip install vllm-0.26.0+cu132-cp312-cp312-win_amd64.whl --extra-index-url https://download.pytorch.org/whl/cu130

IMPORTANT

You need CUDA 13.

This wheel is built for Ampere (RTX 30XX) + ADA (RTX 40XX / PRO) + Blackwell (RTX 50XX / PRO).

• NCCL + Tensor / Pipeline parallelism for multi-gpu inference support on Windows.

• DeepSeek v4 DeepGEMM kernels inference support on Windows. Install with git clone -b deepgemm-nv-dev-windows --single-branch --recursive --shallow-submodules https://github.com/SystemPanic/DeepGEMM-windows.git & cd DeepGEMM-windows & pip install .

Star the repo, give a like and share the announcement on LinkedIn or share on Reddit to help support the project.

v0.25.0

Windows build of release v0.25.0 for Python 3.12, CUDA 13 and Pytorch 2.11

Install the wheel with pip install vllm-0.25.0+cu132-cp312-cp312-win_amd64.whl --extra-index-url https://download.pytorch.org/whl/cu130

IMPORTANT

You need CUDA 13.

This wheel is built for Ampere (RTX 30XX) + ADA (RTX 40XX / PRO) + Blackwell (RTX 50XX / PRO).

• NCCL + Tensor / Pipeline parallelism for multi-gpu inference support on Windows.

• DeepSeek v4 DeepGEMM kernels inference support on Windows. Install with git clone -b deepgemm-nv-dev-windows --single-branch --recursive --shallow-submodules https://github.com/SystemPanic/DeepGEMM-windows.git & cd DeepGEMM-windows & pip install .

Star the repo, give a like and share the announcement on LinkedIn or share on Reddit to help support the project.

v0.24.0

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.

v0.23.0

Windows build of release v0.23.0 for Python 3.12, CUDA 13 and Pytorch 2.11

Install the wheel with pip install vllm-0.23.0+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 (experimental).

Like and share the announcement on LinkedIn or Reddit to help support the project.

v0.22.1

Windows build of release v0.22.1 for Python 3.12, CUDA 13 and Pytorch 2.11

Install the wheel with pip install vllm-0.22.1+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 (experimental).

Like and share the announcement on LinkedIn or Reddit to help support the project.

v0.21.0

Windows build of release v0.21.0 for Python 3.12, CUDA 13 and Pytorch 2.11

Install the wheel with pip install vllm-0.21.0+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.

• This is the first release with NCCL + Tensor / Pipeline parallelism for multi-gpu inference support on Windows (experimental).

Like and share the announcement on LinkedIn or Reddit to help support the project.

v0.20.0

Windows build of release v0.20.0 for Python 3.12, CUDA 13 and Pytorch 2.11

Install the wheel with pip install vllm-0.20.0+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.

• This is the first release with NCCL + Tensor / Pipeline parallelism for multi-gpu inference support on Windows (experimental).

Like and share the announcement on LinkedIn or Reddit to help support the project.

v0.19.0

Windows build of release v0.19.0 for Python 3.12, Cuda 12.4 and Pytorch 2.11

IMPORTANT: Due to Gloo distributed issue with Pytorch 2.8 to Pytorch 2.10, Pytorch 2.11 is required.

Install the wheel with pip install vllm-0.19.0+cu124-cp312-cp312-win_amd64.whl --extra-index-url https://download.pytorch.org/whl/nightly/cu126

v0.17.0

Experimental build of release v0.17.0 for Python 3.12, Cuda 12 and Pytorch 2.11 nightly

IMPORTANT: Due to Gloo distributed issue with Pytorch 2.8 to Pytorch 2.10, Pytorch 2.11 (nightly) is required.

This vLLM build has been done with Pytorch 2.11.dev20260216+cu126, you can install it with pip install vllm-0.17.0+cu124-cp312-cp312-win_amd64.whl --extra-index-url https://download.pytorch.org/whl/nightly/cu126

v0.16.0

Experimental build of release v0.16.0 for Python 3.12, Cuda 12 and Pytorch 2.11 nightly

IMPORTANT: Due to Gloo distributed issue with Pytorch 2.8 to Pytorch 2.10, Pytorch 2.11 (nightly) is required.

This vLLM build has been done with Pytorch 2.11.dev20260216+cu126, you can install it with pip install vllm-0.16.0+cu124-cp312-cp312-win_amd64.whl --extra-index-url https://download.pytorch.org/whl/nightly/cu126