Requirements for Nunchaku's installation environment
To successfully deploy Nunchaku, the following are requiredTertiary environmental requirements::
1. Underlying dependencies
- Python 3.11+ (virtual environment recommended)
- PyTorch 2.6+ (2.7+ required for Blackwell GPUs)
- CUDA 12.2+ (Linux) / 12.6+ (Windows)
2. Compilation tools
- Linux: gcc/g++ 11+ (installable via Conda)
- Windows: Visual Studio with C++ components
3. Python package dependencies
- Necessary packages: ninja, wheel, diffusers, transformers, etc.
- Optional packages: opencv-python (image processing), gradio (demo interface)
Special attention:If using ComfyUI integration, you need to additionally install torchvision==0.21 and make sure the CUDA version matches PyTorch. It is recommended to refer to the official documentationdocs/setup_windows.mdAddressing specific environmental issues.
This answer comes from the articleNunchaku: an inference tool for efficiently running FLUX.1 and SANA 4-bit quantization modelsThe
May not be reproduced without permission:AI productivity tools " What system requirements do I need to fulfill to install Nunchaku?
