how should the custom backward() be defined when i have a non linear activation function?
I think in the docw he gives an example where there the neural networks has no activation function.
https://pytorch.org/docs/master/notes/extending.html
how should the custom backward() be defined when i have a non linear activation function?
I think in the docw he gives an example where there the neural networks has no activation function.
https://pytorch.org/docs/master/notes/extending.html