def forward(self, inputs, mask, hidden, lengths):
sorted_lengths, sort, unsort = sort_by_lengths(lengths)
embedded, mask = self.embeddings(inputs)
output_encoder, hidden = self.encoder(sort(embedded), hidden=None, mask=sort(mask), lengths=sorted_lengths)
representations, attentions = self.attention(output_encoder, mask=sort(mask), lengths=sorted_lengths)
output = self.hidden2out(unsort(representations))
return output, hidden
def forward(self, inputs, mask, hidden, lengths):
sorted_lengths, sort, unsort = sort_by_lengths(lengths)
embedded, mask = self.embeddings(inputs)
output, hidden = self.encoder(sort(embedded),
hidden,
sort(mask),
lengths=sorted_lengths)
output = self.dropout(unsort(output))
decoded = self.decoder(
output.view(output.size(0) * output.size(1), output.size(2)))
return decoded.view(output.size(0), output.size(1),
decoded.size(1)), hidden
def forward(self, inputs, mask, hidden, lengths):
sorted_lengths, sort, unsort = sort_by_lengths(lengths)
batch_size = len(inputs)
embedded, mask = self.embeddings(inputs)
output_encoder, (ht, ct) = self.encoder(sort(embedded),
hidden=None,
mask=sort(mask),
lengths=sorted_lengths)
if self.attention:
representations, attentions = self.attention(
output_encoder, mask=sort(mask), lengths=sorted_lengths)
else:
representations = ht[-2:].transpose(0, 1).contiguous().view(
batch_size, -1)
output = self.hidden2out(unsort(representations))
return output, hidden