checkpoints/epoch_24/training_state.json

{"epoch": 24, "average_batch_loss": 10.091862697601318}