How to change result of forward pass?

1

Hi fellow pytorchers,

I would like to change the result of a certain layer in the forward pass, for instance to choose the entries zeroed by ReLU. I know that using forward_hook is not an option, since it does not allow one to change the output of the layer. Is there any other way?

2

yes, directly just change the Tensor.
For example:

x = ReLU(input)
x[x.eq(0)] = 4