step-by-step solution
The following standardized installation process is recommended for the specific environment of the Windows platform:
- Pre-configuration::
- Install Visual Studio 2022 and check the "Desktop Development with C++" component.
- Ensure CUDA 12.6+ environment via official NVIDIA drivers (requires RTX 20-series graphics card or higher)
- Precision Dependent Installation::
- Create a Python 3.11 virtual environment using conda:
conda create -n nunchaku python=3.11 - Installation by pip specified version:
pip install torch==2.6.0 torchvision==0.16.0 --index-url https://download.pytorch.org/whl/cu121
- Create a Python 3.11 virtual environment using conda:
- Alternative Installation Programs::
- Use the precompiled wheel directly:
pip install https://huggingface.co/mit-han-lab/nunchaku/resolve/main/nunchaku-0.1.4...win_amd64.whl - or set it when compiling from source
set CL=/std:c++17Solving MSVC compatibility problems
- Use the precompiled wheel directly:
Frequently Asked Questions (FAQs):
- If you encounter MSB8036 error, you need to install Windows 10 SDK (10.0.19041.0)
- pass (a bill or inspection etc)nvidia-smiWhen verifying the CUDA version, it is important to note the correspondence between the driver version and the CUDA Toolkit version
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 " How to optimize the installation success of Nunchaku on Windows systems?
