Post

Hello,

I’m trying to use your application, but I’m running into a problem related to PyTorch and the GPU backend.

I have a PC with an Intel Core Ultra 5 processor and an Intel Arc 130V GPU (16GB). I tried installing the AMD version, but I noticed that the code still uses some torch.cuda calls (for example torch.cuda.get_device_capability()), which causes an error because CUDA is only available for NVIDIA GPUs.

I was wondering if there is any possible solution to:

• run the application without GPU acceleration and send the computations directly to the CPU;
• use another compatible backend (such as Intel XPU/OpenVINO if possible);
• or modify the code to use an equivalent backend instead of CUDA for AMD/Intel GPUs.

I also noticed another issue during the Pinokio installation. The script was trying to use {{input.event[0]}}, which caused an ENOENT error because the path could not be found. To bypass this issue, I replaced it directly with the GitHub repository path:
https://github.com/deepbeepmeep/Wan2GP

After that, the installation continued, but the CUDA-related issue remained.

I was also wondering if it would be possible to have a truly AMD-compatible version (without CUDA dependencies), because even the AMD version seems to still contain parts of the code designed for NVIDIA.

I’m happy to test any experimental changes or solutions if needed.