Ignore submodule while loading onto the GPU.

[konrad-karanowski]

Hello. I have a following code:

`class Lightning(LightningModule):

def __init__(self):
    super().__init__()
    
    self.net1 = Net()
    
    self.net2 = Net()`

I want parameters of self.net2 to be ignored while loading model onto the GPU during training. How can I achieve this?