Sequence to sequence model for neural machine translation

requirements:

• python 3.6
• pytorch 0.4.0
• nltk
• tqdm

What's included

• A vanilla seq2seq model implemented in PyTorch
• A keras version is stored in keras/. The code is copied from The Keras official tutorial

How to use

• Go to directory pytorch
• To train the model, simply run ./run.sh
• To test the model ./run_test.sh
• To see what config options you have python main.py --help

background

• this repo tries to implement the paper Sequence to Sequence Learning with Neural Networks, some details differ from the original paper, though.
• we are using data from http://www.manythings.org/anki/

更新

由于PyTorch版本更新了好多次，最近重新写了PyTorch的translation模型。

TODO

• add attention to the currently very basic model
• add beam search at inference

Bug report

If you find any bugs, please feel free to send an email to [email protected] , I will try to be responsive!