LR Schedulers compatibility

[Coen van der Grinten]

Hi all,

When implementing other LR Schedulers besides the ReduceLROnPlateau LR scheduler, the following exception is raised:

lightning_lite.utilities.exceptions.MisconfigurationException: The provided lr scheduler <lr_scheduler>doesn't follow PyTorch's LRScheduler API. You should override theLightningModule.lr_scheduler_step hook with your own logic if you are using a custom LR scheduler.

However, this issue does not appear when implementing the ReduceLROnPlateau given by the original template. I've tried several different schedulers where this issue persists. This issue seems to be due to the lightning module, however lightning is already the latest version.

Any idea how to circumvent this issue?

[Coen van der Grinten]

This issue is related not related to the Lightning-Hydra template.

[Ziyang Chen]

@coenvdgrinten Hi, I met the same issue here. I am using pytorch (2.0.0) and pytorch-lightning (1.6.4). Do you have any solution for it?

[Coen van der Grinten]

Hi there, sorry for the late response.
So far i've not been able to resolve this. If is find a solution, i will post it here.

[Ziyang Chen]

@coenvdgrinten Hi, I think I just find a solution for it. If you update both pytorch and pytorch-lightning to above 2.0, the issue will disappear.