checkpoints/epoch_10/training_state.json

{"epoch": 10, "average_batch_loss": 11.978851146697998}