checkpoints/epoch_26/training_state.json

{"epoch": 26, "average_batch_loss": 9.535414772033691}