sequence to sequence about char-rnn

[karpathy]

yes. You'd have to change the code quite substantially of course, but this gives you an idea of what's involved at least, and how to put things together.

@imaglib, I'm involved in this same task. Still developing, may be we could collaborate? Reach me at [email protected]

[ffmpbgrnn]

Hi guys, I am implementing encoder-decoder with GRU using Torch as well. It seems in the original paper the decoder is conditioned on the encoder output. Does anyone have experience on this? Will it affect the performance a lot if I just eliminate the encoder context?

[karpathy]

There are no plans currently to allow char-rnn to perform sequence-to-sequence prediction. Closing the issue.