LearningRateFinder creates errors for schedulers in val stage #20355
Description
Bug description
I have a lightning module which logs the metrics val_loss, and a scheduler that monitors it
def get_plateau_scheduler(self, optimizer):
plateau_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min',...)
scheduler = {'scheduler': plateau_scheduler, 'interval': 'epoch'
'monitor': 'val_loss'} # <<<<<
return scheduler
class MyModel(ParentModule): # It's a lightning modsule
def configure_optimizers(self):
ret = super().configure_optimizers() # I get the optimizer Here
ret['lr_scheduler'] = self.get_plateau_scheduler(ret['optimizer'])
return retI also have a list of callback on of them is LearningRateFinder
I run a fit trainer = L.Trainer(logger=logger, callbacks=callbacks, **trainer_args). When the Lr Finder is in the list I get
ReduceLROnPlateau conditioned on metric val_loss which is not available. Available metrics are: ['lr-AdamW', 'train_loss', 'train_loss_step', 'time/backward', 'time/train_batch', 'train_loss_epoch', 'time/train_epoch']. Condition can be set using `monitor` key in lr scheduler dict
When I remove the LR finder, training seems to work well.
What version are you seeing the problem on?
v2.4
How to reproduce the bug
No response
Error messages and logs
# Error messages and logs here please
Environment
Current environment
#- PyTorch Lightning Version (e.g., 2.4.0):
#- PyTorch Version (e.g., 2.4):
#- Python version (e.g., 3.12):
#- OS (e.g., Linux):
#- CUDA/cuDNN version:
#- GPU models and configuration:
#- How you installed Lightning(`conda`, `pip`, source):
More info
No response