When a user instantiates a Model object with custom verifier models but only for a subset of the loaded base models, prediction will fail as it expects every base model to have a verifier.

The correct behavior should just skip the verification stage for base models that don't have an associated verifier model.