Using two neural network modules to optimize only one

2

Hi,

I am not sure if I understand your question.
It seems that you have two network: subnetwork1 and subnetwork2, subnetwork2 is pre-trained, and what you want looks like as below:

data -> subnetwork1 -> output1 -> subnetwork2 -> loss

And the params of pre-trained subnetwork2 do not update.
I think you can set requires_grad=False to params of subnetwork2, this thread may help you.

1 Like