Example to demonstrate how to work with fairly complex models using Pytorch subclassing API

[SM1991CODES]

Hi,
This is a great framework. However, the examples are fairly monolithic in architecture.
They are either Sequential or have a single module representing the NN.
Is there some example that shows how to deal with networks that are made up of several blocks, each of which are modules themselves?

Example:

class Net(nn.Module):

def init(self,....):
self.enc = Encoder()
self.dec = Decoder()
self.head1 = ClassificationHead()
self.head2 = RegressionHead()

def forward():
...

class Encoder(nn.Module):
def init(self,...):
...

def forward(self, x):
...

class Decoder(nn.Module):
def init(self,...):
...

def forward(self, x):
...

This would be really useful to try out architectures for tasks like semantic segmentation and object detection.
Please help