/nmt

Neural Machine Translation with various deep learning models!

Primary LanguageJupyter Notebook

Neural Machine Translation (NMT)

A repository to explore different techniques for neural machine translation or Sequence-to-sequence models using Pytorch.
All models are written from scratch in order to be easily understood and integrateable in any future work!

Stats

I only ran all models for 10 epochs. All LSTM models are 2 layered.

Model BLEU (test set) Time (seconds/epoch) GTX 1050 Time (seconds/epoch) RTX 3070
Simple LSTM model 0.152 100 28
Bi-LSTM model 0.208 141 36
Attention with Bi-LSTM model 0.267 269 63
Transformer 0.363 75 16
Transformer with GloVe 0.370 49 12

Notes:

  1. All notebooks are heavily commented.
  2. Torchtext comes with Multi30K dataset. For some reasons (I dont remember), I did not use it.
  3. There is a notebook lstm-exploration.ipynb which explores how LSTM cell works in Pytorch.
  4. Feel free to raise an issue or send a pull request ⚡️.