The interface for rnn's LSTM updateOutput() seems slightly different from torch-rnn's updateOutput(). Is there a nice analog for changing code from using one to the other?
More directly, my code currently uses torch-rnn's LSTM implementation but I would like to switch to rnn's LSTM implementation because it seems to be able to converted to cudnn.LSTM more easily. I've seen this fork: https://github.com/ngimel/torch-rnn, but converting to cudnn doesn't seem to work.
The interface for rnn's LSTM updateOutput() seems slightly different from torch-rnn's updateOutput(). Is there a nice analog for changing code from using one to the other?
More directly, my code currently uses torch-rnn's LSTM implementation but I would like to switch to rnn's LSTM implementation because it seems to be able to converted to cudnn.LSTM more easily. I've seen this fork: https://github.com/ngimel/torch-rnn, but converting to cudnn doesn't seem to work.
Does anyone know any work arounds?