I’m trying to make a custom end-to-end model implemented with Gluon and was wondering if it’s possible to use Skip Connections with a pure HybridSequential model.

I’ve created a Concat gluon wrapper HybridBlock for this purpose, but I don’t know how to use it within a HybridSequential properly.

My initial idea was to create a HybridSequential for both the encoder and decoder and the encoder sequential would connect with the decoder sequential model with concat.

```
class Decoder(HybridBlock):
def __init__(self):
self.model = HybridSequential()
self.model.add(
# Convolution blocks
)
def hybrid_forward(self, f, x):
return self.model(x)
class Encoder(HybridBlock):
def __init__(self, decoder_model):
self.model = HybridSequential()
self.model.add(
# Convolution Block
Concat(decoder_model[-1])
# etc...
)
def hybrid_forward(self, f, x):
return self.model(x)
class Model(HybridBlock):
def __init__(self):
self.model = HybridSequential()
decoder = Decoder()
self.model.add(
decoder,
Encoder(decoder.model)
)
def hybrid_forward(self, f, x):
return self.model(x)
```

But when I thought about how this would work during Feed Forward, it seemed like the Encoder would just re-forward through the decoder model blocks.

Any tips on how to move forward would be gladly appreciated.