Attention-based encoder-decoder model for machine translation

Change the hard-coded paths to data in nmt.py then either run,

./run.sh 

or

THEANO_FLAGS=device=gpu,floatX=float32 python train_nmt.py 

Based on Kyunghyun Cho's tutorial code with some small modifications.