checkpoints/epoch_9/training_state.json

{"epoch": 9, "average_batch_loss": 12.333997039794921}