How to ensure RegressionHead stays positive for prediction?

[nrchade]

I set up an AutoModel using an ImageInput and StructedDataInput with the output being RegressionHead(metrics]["mae"]). The best model solved and when I ran predict with the model, all of the values were negative numbers despite there being no negative numbers in the training target. Is there a way to ensure the model prediction is always greater than or equal to zero? Is this symptomatic of some other issue? i.e. need more training data