TensorFlow model.fit doesn't use EarlyStopping. Add a callback to stop training when val_loss stops improving. Saves compute.