Custom loss function depending on batch (Tensorflow, pytorch)

Question

This is a general Question. I know how to write a custom loss function for Tensorflow that works the following way. Lets say we have a custom loss function L(y_true,y_