Is there any possibility in pytorch where I can split a model from its input to a special layer? Fos this purpose, model.children is not the case as I didn’t care about the functions in between which are not objects.
I’m not completely understanding the issue.
Would you like to forward the input to an additional “special layer” in a model while the default forward pass will still be executed?
If so, you could create a custom module and adapt the
forward method accordingly to your use case.