Bytenet Translation Save

A TensorFlow Implementation of Machine Translation In Neural Machine Translation in Linear Time

Project README

Bytenet Translation

A TensorFlow Implementation of Machine Translation In Neural Machine Translation in Linear Time

Requirements

  • numpy >= 1.11.1
  • TensorFlow >= 1.1 (Probably 1.0 should work as well.)
  • nltk >= 3.2.2 (only for calculating the bleu score)

Notes

  • This implementation is different from the paper in the following aspects.
    • I used the IWSLT 2016 de-en dataset, not the wmt 2014 de-en dataset, which is much bigger.
    • I applied a greedy decoder at the inference phase, not the beam search decoder.
    • I didn't implement Dynamic Unfolding.

Steps

Or if you'd like to use the pretrained model,

Results

After 15 epochs, I obtained the Bleu score 7.38, which is far from good. Maybe some part in the implementation is incorrect. Or maybe we need more data or a bigger model. Details are available in the results folder.

Open Source Agenda is not affiliated with "Bytenet Translation" Project. README Source: Kyubyong/bytenet_translation
Stars
60
Open Issues
3
Last Commit
4 years ago

Open Source Agenda Badge

Open Source Agenda Rating