checkpoints/epoch_1/training_state.json

{"epoch": 1, "average_batch_loss": 46.6366584777832}