Modifying architecture of the models provided by the library

Hey man, did you find a solution to this? I want to load the pretrained model weights and then add a layer after the encoder of the T5 model for translation task. I’m not sure how to edit the architecture of the given models.
Pls share if you did!