Set callbacks to be based on validation loss (not training loss)

[turmeric-blend]

This is my current setup for a run, EarlyStoppingCallback and ModelCheckpointCallback seems to be based on training loss. How to switch for callbacks to be based on validation loss?

losses = {'MeanReturn': MeanReturns(),
      'CumulativeReturn': CumulativeReturn(),
      'SharpeRatio': SharpeRatio(), 
      'SortinoRatio': SortinoRatio(),}

run = Run(model,
          losses['SharpeRatio'],
          dataloader_train,
          val_dataloaders={'train': dataloader_train,
                           'valid': dataloader_valid},
          metrics = {'MeanReturn': losses['MeanReturn'], 
                     'CumulativeReturn': losses['CumulativeReturn'], 
                 'SharpeRatio': losses['SharpeRatio'], 
                 'SortinoRatio': losses['SortinoRatio']},
          optimizer=optimizer,
          callbacks=[EarlyStoppingCallback(dataloader_name='valid', 
                               metric_name='loss',
                       patience=patience), 
                 ModelCheckpointCallback(folder_path=saved_model_folder, 
                         dataloader_name='valid', 
                         metric_name='loss'), 
             TensorBoardCallback(log_dir=tensorboard_path, 
                     log_benchmarks=True)],
      device=device)

[turmeric-blend]

I realised I read tensorboard wrongly