Implementing SR-NMT in OpenNMT-{py,tf}

Hi all,

I am implementing SR-NMT, which has been proposed in Deep Neural Machine Translation with Weakly-Recurrent Units, for both OpenNMT-py and OpenNMT-tf.

I’m investing more time now into the pytorch implementation, which is the one that is working better.
I would like to know whether you are interested in eventually merging them in the master branches, and if yes, I will ask your help with some problems I have using Tensorflow. I have just started to use it.

Thanks in advance for your answers.


Do you have future plans with this architecture? I don’t see a reason to use it when attention or cnn-architectures are producing better results. Do you want to extract more properties with it in a multiple encoder-architecture?