checkpoints/epoch_23/training_state.json

{"epoch": 23, "average_batch_loss": 13.02716760635376}