checkpoints/epoch_17/training_state.json

{"epoch": 17, "average_batch_loss": 10.357013931274414}