prediction_key

The key from the model Run's outputs to use as the eval prediction. If not provided, will use the only output key or raise an error if there are multiple.