checkpoints/epoch_14/training_state.json

{"epoch": 14, "average_batch_loss": 8.572290210723876}