I have defined a function to check whether my model’s weights are changing or not. All is well until I try to resume training from a checkpoint, where both the model and the optimizer state_dicts are saved/loaded. After loading from a checkpoint, my model’s weights show that they are no longer changing. I can re-instanciate my optimizer (defeating the purpose of saving its previous state), and the model’s weights will begin changing again. I have double-checked that the optimizer’s state matches before and after loading (all keys and values within optimizer.dict[‘state’] are the same). Does anyone have any suggestions to solve this issue?