checkpoints/epoch_30/training_state.json

{"epoch": 30, "average_batch_loss": 13.126979503631592}