Right now, the decoders aren't actually doing anything except passing the encoded data through embedding, dropout, and linear output layers. We need to actually use the recurrent unit, which might (definitely does?) mean implementing the forward_step() method from the old version.
Right now, the decoders aren't actually doing anything except passing the encoded data through embedding, dropout, and linear output layers. We need to actually use the recurrent unit, which might (definitely does?) mean implementing the
forward_step()
method from the old version.