RuntimeError: split_with_sizes expects split_sizes to sum exactly to 9301 error when running code

[specialized boy]

I am facing this error when trying to run the codes, why is this happening?
RuntimeError: split_with_sizes expects split_sizes to sum exactly to 9301 (input tensor's size at dimension 0), but got split_sizes=[0, 7566, 638]

[Jean Kaddour]

I'm not sure; I haven't encountered this issue before. Looks like a torch/torch-geometric issue. Which versions are you using?

Also, am I understanding correctly from the screenshots that this happens late during training, i.e., at Epoch 388? If so, it's likely a training instability. Is this an experiment reproduction from the paper, or are you running it with different hparams/data/etc.?

[specialized boy]

I am using torch-geometric==2.2.0 and torch==1.13.1 versions.

Yes, Its the reproduction of the paper. I just git cloned the repo in Colab and started training.

[Jean Kaddour]

Could you please try with

torch==1.10.2
torch-cluster==1.5.9
torch-geometric==2.0.3
torch-scatter==2.0.9
torch-sparse==0.6.12
torch-spline-conv==1.2.1

With these versions, I recently reproduced the results