It comes to a situation that needs to train a model progressively, and I want to get a model with a small size, but just as the example in the following, my temporary model
