I have some legacy code that finetunes a network. I need to implement a ReduceLROnPlateau. As a first step, I\'m trying to feed in a constant learning rate as a tensor. Here is