Running the integrated GraphGym

[nilskre]

Two questions regarding running the integrated GraphGym:

1. Is there any documentation, on how the virtual environment for GraphGym needs to be set up / which dependencies are necessary?
2. The docs state, that running on the CPU is possible by adding device: cpu to the *.yaml file. Is this file meant? After adding it there it seems to work (Log: GPU available: True (cuda), used: False). But running a single experiment still fails with RuntimeError: CUDA unknown error - this may be due to an incorrectly set up environment, e.g. changing env variable CUDA_VISIBLE_DEVICES after program start. Setting the available devices to be zero.

Thanks in advance!

[rusty1s]

1. Necessary dependencies can be installed via pip install torch-geometric[graphgym].
2. Let me take a look.

[rusty1s]

I have set accelerator: cpu in the yml file, and it runs fine for me. I can confirm that the data is still on the CPU. Can you share some more information where it crashes for you exactly?

[nilskre]

Thanks a lot for this fast and helpful response!
I can now confirm, that with accelerator: cpu in the yml file everything works as expected. Thanks!
Could it be the case, that the docs about this are outdated? They state Run GraphGym with CPU backend: GraphGym supports CPU backend as well – you only need to add the line device: cpu to the *.yaml file. (this is not working).

[rusty1s]

Yes, indeed. Will fix.

[rusty1s]

Fixed in #7274.