How to Combine mlmodel files?

What goal are you trying to achieve here?

It's easy to guess.

For example, you have two models, one predicting cats' breeds and the other dogs' breeds. Rather than submitting a picture to both and comparing scores, if you can combine both models into one, it's better.

I'm unsure if there's any way to do that outside of retraining the whole model. But then, the amount of data would drop the precision down.

Would be an interesting feature for CreateML/CoreML (CoreML calls already take in an array of models, it would be nice if there's an option to return only the best guess)