I have a question I want to ask, how to use the pre-training parameters of modelA and modelB in the combined network