checkpoints/epoch_34/training_state.json

{"epoch": 34, "average_batch_loss": 9.400419807434082}