Best Practices

[istranical]

I'm struggling to find examples for setting up training + validation when using the Trainer with datamodule input.

It seems like trainer.fit(model, datamodule = ....) never performs validation, even if the validation step is defined in the model. Does that mean that I need to specify training and validation separately using trainer.fit and trainer.validate?

Specifying train_dataloader and val_dataloaders to trainer.fit() works just fine, but then I lose the benefits of using the datamodule in trainer.fit(..), including prepare_data() and setup() running automatically.

What's the best practice here? I feel like I'm missing something.