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

v0.11.2

Windows build of release v0.11.2 for Python 3.12, Cuda 12.4, Pytorch 2.8.0

IMPORTANT: Starting from version v0.11.2, you need to compile Pytorch 2.8.0 from source due to Gloo distributed issue with Pytorch 2.8+

Wait for the Pytorch issue to be resolved, or build Pytorch from source following https://github.com/pytorch/pytorch#from-source

v0.11.0

Windows build of release v0.11.0 for Python 3.12, Cuda 12.4, Pytorch 2.7.1

libuv is disabled by default for vLLM cli / Openai api server.

You need to compile Pytorch 2.7.1 from source with libuv support. If you do not want to compile pytorch from source:

1. Install the 2.7.1 stable version with pip install torch==2.7.1+cu126 torchaudio==2.7.1+cu126 torchvision==0.22.1+cu126 --index-url https://download.pytorch.org/whl/cu126
2. Disable libuv with os.environ["USE_LIBUV"] = "0" before importing vLLM on your own code to avoid RuntimeError: use_libuv was requested but PyTorch was built without libuv support

v0.10.2

Windows build of release v0.10.2 for Python 3.12, Cuda 12.4, Pytorch 2.7.1

libuv is disabled by default for vLLM cli / Openai api server.

You need to compile Pytorch 2.7.1 from source with libuv support. If you do not want to compile pytorch from source:

1. Install the 2.7.1 stable version with pip install torch==2.7.1+cu126 torchaudio==2.7.1+cu126 torchvision==0.22.1+cu126 --index-url https://download.pytorch.org/whl/cu126
2. Disable libuv with os.environ["USE_LIBUV"] = "0" before importing vLLM on your own code to avoid RuntimeError: use_libuv was requested but PyTorch was built without libuv support

v0.10.1.1

Starting with this version, vLLM wheels will be released for Pytorch 2.7.1 instead of Pytorch 2.6.0

Windows build of release v0.10.1.1 for Python 3.12, Cuda 12.4, Pytorch 2.7.1

libuv is disabled by default for vLLM cli / Openai api server.

You need to compile Pytorch 2.7.1 from source with libuv support. If you do not want to compile pytorch from source:

1. Install the 2.7.1 stable version with pip install torch==2.7.1+cu126 torchaudio==2.7.1+cu126 torchvision==0.22.1+cu126 --index-url https://download.pytorch.org/whl/cu126
2. Disable libuv with os.environ["USE_LIBUV"] = "0" before importing vLLM on your own code to avoid RuntimeError: use_libuv was requested but PyTorch was built without libuv support

v0.10.0

Starting with this version, vLLM wheels will be released for Pytorch 2.7.1 instead of Pytorch 2.6.0

Windows build of release v0.10.0 for Python 3.12, Cuda 12.4, Pytorch 2.7.1

libuv is disabled by default for vLLM cli / Openai api server.

You need to compile Pytorch 2.7.1 from source with libuv support. If you do not want to compile pytorch from source:

1. Install the 2.7.1 stable version with pip install torch==2.7.1+cu126 torchvision==0.22.1+cu126 torchaudio==2.7.1+cu126 --index-url https://download.pytorch.org/whl/cu126
2. Disable libuv with os.environ["USE_LIBUV"] = "0" before importing vLLM on your own code to avoid RuntimeError: use_libuv was requested but PyTorch was built without libuv support

v0.9.2

Windows build of release v0.9.2 for Python 3.12, Cuda 12.4, Pytorch 2.6.0 Nightly with Gloo distributed backend + libuv, cuDSS, cuDNN and cuBLAS enabled.

libuv is disabled by default for vLLM cli / Openai api server.

You need to compile Pytorch 2.6.0 from source with libuv support. If you do not want to compile pytorch:

1. Install the 2.6.0 stable version with pip install torch==2.6.0+cu124 torchvision==0.21.0+cu124 torchaudio==2.6.0+cu124 --index-url https://download.pytorch.org/whl/cu124
2. Disable libuv with os.environ["USE_LIBUV"] = "0" before importing vLLM on your own code to avoid RuntimeError: use_libuv was requested but PyTorch was built without libuv support

v0.9.1

Windows build of release v0.9.1 for Python 3.12, Cuda 12.4, Pytorch 2.6.0 Nightly with Gloo distributed backend + libuv, cuDSS, cuDNN and cuBLAS enabled.

libuv is disabled by default for vLLM cli / Openai api server.

You need to compile Pytorch 2.6.0 from source with libuv support. If you do not want to compile pytorch:

1. Install the 2.6.0 stable version with pip install torch==2.6.0+cu124 torchvision==0.21.0+cu124 torchaudio==2.6.0+cu124 --index-url https://download.pytorch.org/whl/cu124
2. Disable libuv with os.environ["USE_LIBUV"] = "0" before importing vLLM on your own code to avoid RuntimeError: use_libuv was requested but PyTorch was built without libuv support

v0.9.0.1

Windows build of release v0.9.0.1 for Python 3.12, Cuda 12.4, Pytorch 2.6.0 Nightly with Gloo distributed backend + libuv, cuDSS, cuDNN and cuBLAS enabled.

Make sure to have installed the following torch version in your environment before installing the wheel:

pip install torch==2.6.0+cu124 torchvision==0.21.0+cu124 torchaudio==2.6.0+cu124 --index-url https://download.pytorch.org/whl/cu124

v0.8.5

Windows build of release v0.8.5.post1 for Python 3.12, Cuda 12.4, Pytorch 2.6.0 Nightly with Gloo distributed backend + libuv, cuDSS, cuDNN and cuBLAS enabled.

Make sure to have installed the following torch version in your environment before installing the wheel:

pip install torch==2.6.0+cu124 torchvision==0.21.0+cu124 torchaudio==2.6.0+cu124 --index-url https://download.pytorch.org/whl/cu124