AI-windows-whl

AI-windows-whl

Pre-compiled Python whl for Flash-attention, SageAttention, NATTEN, xFormer etc

Stars: 147

Visit
 screenshot

AI-windows-whl is a curated collection of pre-compiled Python wheels for difficult-to-install AI/ML libraries on Windows. It addresses the common pain point of building complex Python packages from source on Windows by providing direct links to pre-compiled `.whl` files for essential libraries like PyTorch, Flash Attention, xformers, SageAttention, NATTEN, Triton, bitsandbytes, and other packages. The goal is to save time for AI enthusiasts and developers on Windows, allowing them to focus on creating amazing things with AI.

README:

Windows AI Wheels

Logo

A curated collection of pre-compiled Python wheels for difficult-to-install AI/ML libraries on Windows.

Report a Broken Link · Request a New Wheel

Table of Contents
  1. About The Project
  2. Getting Started
  3. Available Wheels

About The Project

This repository was created to address a common pain point for AI enthusiasts and developers on the Windows platform: building complex Python packages from source. Libraries like flash-attention, xformers are essential for high-performance AI tasks but often lack official pre-built wheels for Windows, forcing users into a complicated and error-prone compilation process.

The goal here is to provide a centralized, up-to-date collection of direct links to pre-compiled .whl files for these libraries, primarily for the ComfyUI community and other PyTorch users on Windows. This saves you time and lets you focus on what's important: creating amazing things with AI.

(back to top)

Getting Started

Follow these simple steps to use the wheels from this repository.

Prerequisites

  1. Python for Windows: Ensure you have a compatible Python version installed (PyTorch currently supports Python 3.9 - 3.12 on Windows). You can get it from the official Python website.

Installation

To install a wheel, use pip with the direct URL to the .whl file. Make sure to enclose the URL in quotes.

# Example of installing a specific flash-attention wheel
pip install "https://huggingface.co/lldacing/flash-attention-windows-wheel/blob/main/flash_attn-2.7.4.post1+cu128torch2.7.0cxx11abiFALSE-cp312-cp312-win_amd64.whl"

[!TIP] Find the package you need in the Available Wheels section below, find the row that matches your environment (Python, PyTorch, CUDA version), and copy the link for the pip install command.

(back to top)

Available Wheels

Here is the list of tracked packages.

PyTorch

The foundation of everything. Install this first from the official source.

For convenience, here are direct installation commands for specific versions on Linux/WSL with an NVIDIA GPU. For other configurations (CPU, macOS, ROCm), please use the official install page.

Stable Version (2.8.0)

This is the recommended version for most users.

CUDA Version Pip Install Command
CUDA 12.9 pip install torch torchvision --index-url https://download.pytorch.org/whl/cu129
CUDA 12.8 pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128
CUDA 12.6 pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126
CPU only pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu

Previous Stable Version (2.7.1)

CUDA Version Pip Install Command
CUDA 12.8 pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu128
CUDA 12.6 pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu126
CUDA 11.8 pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu118
CPU only pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cpu

Nightly Versions

Use these for access to the latest features, but expect potential instability.

PyTorch 2.9 (Nightly)

CUDA Version Pip Install Command
CUDA 12.9 pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu130
CUDA 12.8 pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128
CUDA 12.6 pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu126

Torchaudio

Package Version PyTorch Ver CUDA Ver Download Link
2.8.0 2.9.0 12.8 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

Flash Attention

High-performance attention implementation.

Package Version PyTorch Ver Python Ver CUDA Ver CXX11 ABI Download Link
2.8.3 2.9.0 3.12 12.8 Link
2.8.3 2.8.0 3.12 12.8 Link
2.8.2 2.9.0 3.12 12.8 Link
2.8.2 2.8.0 3.10 12.8 Link
2.8.2 2.8.0 3.11 12.8 Link
2.8.2 2.8.0 3.12 12.8 Link
2.8.2 2.7.0 3.10 12.8 Link
2.8.2 2.7.0 3.11 12.8 Link
2.8.2 2.7.0 3.12 12.8 Link
2.8.1 2.8.0 3.12 12.8 Link
2.8.0.post2 2.8.0 3.12 12.8 Link
2.7.4.post1 2.8.0 3.10 12.8 Link
2.7.4.post1 2.8.0 3.12 12.8 Link
2.7.4.post1 2.7.0 3.10 12.8 Link
2.7.4.post1 2.7.0 3.11 12.8 Link
2.7.4.post1 2.7.0 3.12 12.8 Link
2.7.4 2.8.0 3.10 12.8 Link
2.7.4 2.8.0 3.11 12.8 Link
2.7.4 2.8.0 3.12 12.8 Link
2.7.4 2.7.0 3.10 12.8 Link
2.7.4 2.7.0 3.11 12.8 Link
2.7.4 2.7.0 3.12 12.8 Link
2.7.4 2.6.0 3.10 12.6 Link
2.7.4 2.6.0 3.11 12.6 Link
2.7.4 2.6.0 3.12 12.6 Link
2.7.4 2.6.0 3.10 12.4 Link
2.7.4 2.6.0 3.11 12.4 Link
2.7.4 2.6.0 3.12 12.4 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

xformers

Another library for memory-efficient attention and other optimizations.

[!NOTE] PyTorch provides official pre-built wheels for xformers. You can often install it with pip install xformers if you installed PyTorch correctly. If that fails, find your matching wheel at the index link above.

ABI3 version, any Python 3.9-3.12

Package Version PyTorch Ver CUDA Ver Download Link
0.0.32.post2 2.8.0 12.8 Link
0.0.32.post2 2.8.0 12.9 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

SageAttention

Package Version PyTorch Ver Python Ver CUDA Ver Download Link
2.1.1 2.5.1 3.9 12.4 Link
2.1.1 2.5.1 3.10 12.4 Link
2.1.1 2.5.1 3.11 12.4 Link
2.1.1 2.5.1 3.12 12.4 Link
2.1.1 2.6.0 3.9 12.6 Link
2.1.1 2.6.0 3.10 12.6 Link
2.1.1 2.6.0 3.11 12.6 Link
2.1.1 2.6.0 3.12 12.6 Link
2.1.1 2.6.0 3.12 12.6 Link
2.1.1 2.6.0 3.13 12.6 Link
2.1.1 2.7.0 3.10 12.8 Link
2.1.1 2.8.0 3.12 12.8 Link

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

SageAttention 2.2 (SageAttention2++)

[!NOTE] Only supports CUDA >= 12.8, therefore PyTorch >= 2.7.

Package Version PyTorch Ver Python Ver CUDA Ver Download Link
2.2.0.post2 2.5.1 >3.9 12.4 Link
2.2.0.post2 2.6.0 >3.9 12.6 Link
2.2.0.post2 2.7.1 >3.9 12.8 Link
2.2.0.post2 2.8.0 >3.9 12.8 Link
2.2.0.post2 2.9.0 >3.9 12.8 Link
2.2.0 2.7.1 3.9 12.8 Link
2.2.0 2.7.1 3.10 12.8 Link
2.2.0 2.7.1 3.11 12.8 Link
2.2.0 2.7.1 3.12 12.8 Link
2.2.0 2.7.1 3.13 12.8 Link
2.2.0 2.8.0 3.9 12.8 Link
2.2.0 2.8.0 3.10 12.8 Link
2.2.0 2.8.0 3.11 12.8 Link
2.2.0 2.8.0 3.12 12.8 Link
2.2.0 2.8.0 3.13 12.8 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

SpargeAttn

Package Version PyTorch Ver CUDA Ver Download Link
0.1.0.post1 2.7.1 12.8 Link
0.1.0.post1 2.8.0 12.8 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

Nunchaku

Package Version PyTorch Ver Python Ver Download Link
1.0.0 2.5 3.10 Link
1.0.0 2.5 3.11 Link
1.0.0 2.5 3.12 Link
1.0.0 2.6 3.10 Link
1.0.0 2.6 3.11 Link
1.0.0 2.6 3.12 Link
1.0.0 2.6 3.13 Link
1.0.0 2.7 3.10 Link
1.0.0 2.7 3.11 Link
1.0.0 2.7 3.12 Link
1.0.0 2.7 3.13 Link
1.0.0 2.8 3.10 Link
1.0.0 2.8 3.11 Link
1.0.0 2.8 3.12 Link
1.0.0 2.8 3.13 Link
1.0.0 2.9 3.10 Link
1.0.0 2.9 3.11 Link
1.0.0 2.9 3.12 Link
1.0.0 2.9 3.13 Link
0.3.2 2.5 3.10 Link
0.3.2 2.5 3.11 Link
0.3.2 2.5 3.12 Link
0.3.2 2.6 3.10 Link
0.3.2 2.6 3.11 Link
0.3.2 2.6 3.12 Link
0.3.2 2.7 3.10 Link
0.3.2 2.7 3.11 Link
0.3.2 2.7 3.12 Link
0.3.2 2.8 3.10 Link
0.3.2 2.8 3.11 Link
0.3.2 2.8 3.12 Link
0.3.2 2.9 3.12 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

NATTEN

Neighborhood Attention Transformer.

Package Version PyTorch Ver Python Ver CUDA Ver Download Link
0.17.5 2.6.0 3.10 12.6 Link
0.17.5 2.6.0 3.11 12.6 Link
0.17.5 2.6.0 3.12 12.6 Link
0.17.5 2.7.0 3.10 12.8 Link
0.17.5 2.7.0 3.11 12.8 Link
0.17.5 2.7.0 3.12 12.8 Link
0.17.3 2.4.0 3.10 12.4 Link
0.17.3 2.4.0 3.11 12.4 Link
0.17.3 2.4.0 3.12 12.4 Link
0.17.3 2.4.1 3.10 12.4 Link
0.17.3 2.4.1 3.11 12.4 Link
0.17.3 2.4.1 3.12 12.4 Link
0.17.3 2.5.0 3.10 12.4 Link
0.17.3 2.5.0 3.11 12.4 Link
0.17.3 2.5.0 3.12 12.4 Link
0.17.3 2.5.1 3.10 12.4 Link
0.17.3 2.5.1 3.11 12.4 Link
0.17.3 2.5.1 3.12 12.4 Link

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

Triton (Windows Fork)

Triton is a language and compiler for writing highly efficient custom deep-learning primitives. Not officially supported on Windows, but a fork provides pre-built wheels.

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

bitsandbytes

A lightweight wrapper around CUDA custom functions, particularly for 8-bit optimizers, matrix multiplication (LLM.int8()), and quantization functions.

▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲▼▲

RadialAttention for ComfyUI

(back to top)

▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀

Accessing Data Programmatically (wheels.json)

All wheel information in this repository is managed in the wheels.json file, which serves as the single source of truth. The tables in this README are automatically generated from this file.

This provides a stable, structured JSON endpoint for any external tool or application that needs to access this data without parsing Markdown.

How to Use

You can access the raw JSON file directly via the following URL:

https://raw.githubusercontent.com/wildminder/AI-windows-whl/main/wheels.json

Example using curl:

curl -L -o wheels.json https://raw.githubusercontent.com/wildminder/AI-windows-whl/main/wheels.json

The file contains a list of packages, each with its metadata and an array of wheels, where each wheel object contains version details and a direct download url.

(back to top)

▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀

Contributing

Contributions are what make the open source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

If you have found a new pre-built wheel or a reliable source, please fork the repo and create a pull request, or simply open an issue with the link.

(back to top)

Acknowledgments

This repository is simply a collection of links. Huge thanks to the individuals and groups who do the hard work of building and hosting these wheels for the community:

For Tasks:

Click tags to check more tools for each tasks

For Jobs:

Alternative AI tools for AI-windows-whl

Similar Open Source Tools

For similar tasks

For similar jobs