Hello, I am working with
functorch Jacobian, I am using it to get the Jacobian of an intermediate layer
i in my model with respect to the input
z, however, I have two methods to access the i-th layer, the first is by just taking the model indexes, as follows:
i_layer_out = G.model[:i+1](z)
and the second is using recurssion:
inp = z for j in range(i+1): inp = G.model[j](inp) i_layer_out = inp
The output of the model slightly varies depending on the approach, but the Jacobians are different, so I was wondering what is the best method to get the output of the intermediate layer, thanks for any comment.