checkpoints/epoch_29/training_state.json

{"epoch": 29, "average_batch_loss": 8.996773643493652}