checkpoints/epoch_27/training_state.json

{"epoch": 27, "average_batch_loss": 10.83188690185547}