Open DuHao10086 opened 4 years ago
I think its due to the graclus
call. I will investigate it, sorry for that :)
I have the exact same problem (with pytorch 1.7.0 and cuda 11.0).
If I change the clustering method to voxel_grid
the problem indeed does not occur.
❓ Questions & Help
Hi, first of all, thank you very much for building this comprehensive library. I'm currently working on a graph classification problem and I adapt your model in mnist_nn_conv.py for experiment. However, after a few epochs, the code is stuck and the gpu utilization keeps around 38%. The process can only be killed and not responding to keyboard interrupt ctrl+c. I have tested pytorch 1.5.0, 1.6.0 and on other gpus but same. Could you please provide any suggestions? Thank you very much!
`
And if I just use a simple GCN based model as follow, this problem won't occur during training. `