How to implement conv:clone in pytorch

Hi,
I am reproducing a torch model to a pytorch model. but i have difficulties in how to share the parameters and the gradParameters in two layers.
Here is torch scipt,

.
Thanks for any help.

you can reuse the same layer again and again. you dont need to clone a separate layer for a separate output.

For example, this works fine:

m = nn.Conv2d(...)

output1 = m(input1)
output2 = m(input2)

(output1 + output2).sum().backward()
1 Like