Thank you, @Etienne38 ! Looks great!
The only issue is everything is meant for DyNET! And as a learner, I'm already overwhelmed by the number of frameworks! Tensorflow, Theano, Torch now OpenNMT, DyNet! So far OpenNMT seems like the best bet so i might stick to that. Because I'm only interested in the inference part of NMT and speeding that up.