Graphgym config cfg.train.auto_resume not working?

[nilskre]

Hello,

the setting is the training of a GraphGym model, especially starting training again from a saved checkpoint.

Expected behavior: By setting cfg.train.auto_resume to true the training process should check, if a saved model from before exists and continue training from this model/epoch.

Observed behavior: Training is again started from scratch (only the folder with the saved checkpoint is not deleted).

Can you reproduce this?
Thanks in advance!

[rusty1s]

I am sorry for the late reply. I think this flag has been accidentally removed in a re-factor, I am really sorry. Let me try to bring it back.

[nilskre]

No worries, thanks for the explanation and confirmation.