checkpoints/epoch_28/training_state.json

{"epoch": 28, "average_batch_loss": 11.444709224700928}