I currently log my validation loss values to mlflow using the following:
def validation_step(self, batch: Tuple[List[str], List[str]], *args) -> None:
loss = self.common_step(batch)
self.log(VALID_LOSS, loss, on_step=True, on_epoch=True, prog_bar=True
I can see in a few answers that logged_metric
was suggested in a few answers. However, considering that I am asking interested only in VALID_LOSS
at epoch end and not the stepwise one. Is there a way to access this from the trainer?