Doubt in seq-2-seq tutorial provided in the docs

#1

Hi,

In the seq-2-seq tutorial provided in the docs (http://pytorch.org/tutorials/intermediate/seq2seq_translation_tutorial.html#the-seq2seq-model), I am unable to understand the below part:

for i in range(self.n_layers):
        output, hidden = self.gru(output, hidden)

What is the use of n_layers here? Is it for stacked GRU’s? In that case, shouldn’t the hidden representation being sent in the next iteration be different?

Thanks

#2

Good point. It indeed seems a bit weird to me as well. It uses the same weights to do stacked RNN, which is unusual, but can still work for the simple dataset in the tutorial.