checkpoints/epoch_31/training_state.json

{"epoch": 31, "average_batch_loss": 9.387823123931884}