checkpoints/epoch_39/training_state.json

{"epoch": 39, "average_batch_loss": 11.265805740356445}