RuntimeError: No CUDA GPUs are available

#115
by liuganghuggingface - opened

I encountered the issue 'RuntimeError: No CUDA GPUs are available' when I deployed the model using zero-gpu. Any idea on how to fix it? Thank you!

image.png

ZeroGPU Explorers org

import spaces at starting of code.

Yes. I import spaces in the first line of the app.py. But it doesn't solve my question.
image.png

Here is my code: https://huggingface.co/spaces/liuganghuggingface/Polymer-Design-With-GraphDiT/tree/main

The issue seems to be related to the customized graph diffusion model in the graph_decoder/diffusion_model.py file, but I haven't identified a solution for it yet.

Update: Fortunately, I found the issue was caused by importing the torch_geometric package in this line. After commenting it out, everything works fine. I haven't looked deeper into the cause, but it seems these two packages are currently incompatible.

Sign up or log in to comment