checkpoints/epoch_36/training_state.json

{"epoch": 36, "average_batch_loss": 11.218118057250976}