Unsupervised Paraphrasing using Transfer Learning

This is an implementation of a transfer learning approach based on unsupervised paraphrasing for Indic languages, which uses a T5 transformer-based architecture.

Usage

• model.py has pretrained model and tokenizer related code.
• data folder have all the code related data preprocessing.
• main.py has code related to training the model on task adaptation.
• ssmain.py has the code related to training self supervised model
• generate_data.py has the code related to generating pseudo labels.
• eval.py has the code related to generating evalutation metrics.

All the file are ran plainly with python without any arguments.