How does Learner.get_preds() output a Loss Tensor even for a Test Set

In a multiclass classification like Pets [Lesson 1] my understanding is that to calculate a loss function we would need the Actual label.

On a test set we don’t have Actual labels, but Learner.get_preds() still outputs a Loss Tensor filled with values. How are these calculated? What do these mean?

I looked at the docs and forums but couldn’t figure it out hence checking here.

Thanks for your time.