Models without labels?

I've trained an autoencoder in keras/tensorflow, to function basically as a smart error-correction function. It seems promising, but although I can successfully convert the model using coremltools, I cannot import it... When I drop it in my Xcode project it complains that "Classifier models must provide class labels." Well, it's not a classifier model, so what can I do? Should I provide bogus labels, or is there an argument that allows coreml to ignore/forgo labels?