Using One Model for Training Another Model

Hi everyone. I am using already-trained pytorch models ‘ae_model’ and ‘classifier’ to train another pytorch model. However, the resulting loss function (BCELoss) is not updating the weights of the new model because the prediction grad is None. The requires_grad of all intermediate and final outputs is True. How can I mitigate this issue? Thank you for your help.
Should a better solution consider creating a Master class that combines these models altogether?