checkpoints/epoch_25/training_state.json

{"epoch": 25, "average_batch_loss": 10.911422100067139}