I am currently doing something like this:
brnn = nn.seqBRNN(vec_size,rnn_size)
brnn.forwardModule:maskZero()
brnn.backwardModule:maskZero()
I have extended seqBRNN to add the ability stack seqBRNN with a stack_size argument that adds modules in a loop. I also added a seqBRNN:maskZero() method on my local repo.
If this is something that you might want, should I send a merge request?
Thanks