In this repository we publish the code related to our work on Text Style Transfer using the recurrent paraphrase architecture proposed by Krishna[1]. We augmented this training pipeline to include an additional reward fine tuning step using the TRL library.
We publish the code in the form colab notebooks dedicated to the domains of data cleaning, training and evaluation.
[1] Krishna, K., Wieting, J., & Iyyer, M. (2020). Reformulating unsupervised style transfer as paraphrase generation. arXiv preprint arXiv:2010.05700.