Pakaho-q1/window-wheel_store

q1 ComfyUI Installer — Pre-built Wheels 🚀

Welcome to the official repository for pre-built Windows (x64) wheels (.whl) designed to work seamlessly with the q1-comfyui_installer.

This repository resolves common installation and compilation issues for ComfyUI Windows users by providing pre-compiled binaries for advanced acceleration libraries — no CUDA toolkit or compiler required.

---

📋 Table of Contents

• Main Project Link
• Supported OS
• Compatibility Matrix
• Package Details
• Installation Guide
• Check Your Environment
• Troubleshooting

---

🔗 Main Project Link

• GitHub Repository: https://github.com/Pakaho-q1/q1-comfyui_installer

Refer to the GitHub repository for the full one-click ComfyUI installer script, complete instructions, and source code.

---

💻 Supported Operating System

• Windows 10 / 11 (64-bit) only (win_amd64)

⚠️ Linux and macOS are not supported by these wheels.

---

🛠️ Compatibility Matrix

Choose the wheel that matches your Python, PyTorch, and CUDA version.
(Full index available in wheels_index.json)

Python Version	PyTorch Version	CUDA Version
Python 3.11 (cp311)	PyTorch 2.9 / 2.10	CUDA 12.9 (cu128/cu129) & CUDA 13.0 (cu130)
Python 3.12 (cp312)	PyTorch 2.9 / 2.10	CUDA 12.9 (cu128/cu129) & CUDA 13.0 (cu130)
Python 3.13 (cp313)	PyTorch 2.9 / 2.10	CUDA 12.9 (cu128/cu129) & CUDA 13.0 (cu130)

---

📦 Package Details

This repository provides pre-built wheels for three performance-enhancing libraries:

1. FlashAttention-2 (flash_attn)

• Version: 2.8.3
• Description: Speeds up Transformer-based attention calculations and significantly reduces VRAM usage.
• CUDA Mapping:
  • CUDA 12.9 → use filename containing +cu128torch
  • CUDA 13.0 → use filename containing +cu130torch

2. Nunchaku (nunchaku)

• Version: 1.2.0 (for PyTorch 2.9) · 1.0.2 (for PyTorch 2.10)
• Description: An inference acceleration library optimized for Diffusion model speed.

3. SageAttention2++ (sageattention)

• Version: 2.2.0
• Description: An ultra-fast, memory-efficient attention library for image and video diffusion models.
• Special Note: Compiled as a shared cp39-abi3 wheel — one file supports Python 3.9 and higher (3.11, 3.12, 3.13).

---

📥 Installation Guide

You can install these wheels directly from Hugging Face using pip — no need to download the .whl files manually.

URL Structure

https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/<package_folder>/<filename.whl>

---

✅ Example 1: Python 3.12 + PyTorch 2.9 + CUDA 12.9

# Install FlashAttention-2
pip install "https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/flash_attn-2.8.3/flash_attn-2.8.3%2Bcu128torch2.9-cp312-cp312-win_amd64.whl"

# Install Nunchaku
pip install "https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/Nunchaku/nunchaku-1.2.0%2Btorch2.9-cp312-cp312-win_amd64.whl"

# Install SageAttention2++
pip install "https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/SageAttention2%2B%2B/sageattention-2.2.0%2Bcu128torch2.9.0andhigher.post4-cp39-abi3-win_amd64.whl"

---

✅ Example 2: Python 3.11 + PyTorch 2.10 + CUDA 13.0

# Install FlashAttention-2
pip install "https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/flash_attn-2.8.3/flash_attn-2.8.3%2Bcu130torch2.10-cp311-cp311-win_amd64.whl"

# Install Nunchaku
pip install "https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/Nunchaku/nunchaku-1.0.2%2Btorch2.10-cp311-cp311-win_amd64.whl"

# Install SageAttention2++
pip install "https://huggingface.co/Pakaho-q1/q1-comfyui_installer/resolve/main/SageAttention2%2B%2B/sageattention-2.2.0%2Bcu130torch2.9.0andhigher.post4-cp39-abi3-win_amd64.whl"

---

🔍 How to Check Your Environment

Open Terminal or Command Prompt inside your ComfyUI directory (or activate your virtual environment) and run:

# Check Python version
python --version

# Check PyTorch and CUDA version
python -c "import torch; print('Torch:', torch.__version__); print('CUDA:', torch.version.cuda)"

Use the output to select the correct wheel from the compatibility matrix above or from wheels_index.json.

---

🛟 Troubleshooting

Q: pip says "No matching distribution found"
A: Your Python, PyTorch, or CUDA version doesn't match the wheel filename. Re-check your environment with the commands above.

Q: The download URL returns a 404 error
A: The file may have been renamed or updated. Check wheels_index.json for the latest filenames.

Q: Can I use these on Linux or macOS?
A: No. These wheels are compiled for Windows x64 (win_amd64) only.

Q: Which SageAttention wheel should I pick?
A: There is only one file per CUDA version for SageAttention — it works with Python 3.9 and above automatically.

---

📄 License

This repository is licensed under the Apache License 2.0.