Skip to content

josephch405/curriculum-nmt

Repository files navigation

curriculum-nmt

Setup

  1. Create conda environment: conda create --name curriculum_nmt python=3.7
  2. Install requirements in requirements.txt
  3. Run bash run_iwslt.sh download to download the IWSLT dataset
  4. Run bash run_iwslt.sh vocab to generate vocab files. This generates a iwslt_vocab.json and iwslt_word_freq.json

Usage

  1. Train the model locally on IWSLT with bash run_iwslt.sh train_local (with "none" ordering)

  2. Train the model with desired scoring and pacing functions locally on IWSLT e.g. bash run_iwslt.sh train_local rarity linear (with "rarity" ordering and "linear" pacing. see scoring.py and pacing.py for more options)

References

  1. Fine-Tuning by Curriculum Learning for Non-Autoregressive Neural Machine Translation arXiv
  2. On The Power of Curriculum Learning in Training Deep Networks arXiv code
  3. Competence-based Curriculum Learning for Neural Machine Translation arXiv
  4. Improving Neural Machine Translation Models with Monolingual Data arXiv

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published