Unused convolutional layers

[sotwi]

Hello @taasnim I am looking into your code because I'm trying to implement a similar system to evaluate my Text Generation project.

While looking into the code I ran into what looks like a mistake in lines 198 to 202 of the model.py code.

During initialization (and according to your paper) there should be 6 convolutional layers. However, during the forward pass you use the first 4 layers and then, in lines 198 to 202 instead of calling layers 5 and 6 you call layers 3 and 4 again. In practice, layers 5 and 6 are never really used.

[taasnim]

Hi @sotwi,
Thanks for figuring it out.
In my local repository, I updated the issue long ago but forgot to make a push here.
Anyway, I updated the code now.