DASCL: Dictionary-Assisted Supervised Contrastive Learning

Last Updated: November 29, 2022

Authors: Patrick Y. Wu, Richard Bonneau, Joshua A. Tucker, and Jonathan Nagler

This repository includes a PyTorch implementation of the dictionary-assisted supervised contrastive learning framework, described here. We will be updating this repository in the lead up to EMNLP 2022; please check back as we get closer to the conference.

Usage

To use the functions in this package, install the package first. To do this, navigate to the directory you would like to install the files in. Then,

git clone https://github.com/SMAPPNYU/DASCL
cd DASCL
pip install -r requirements.txt
python setup.py develop

We have implemented DASCL for BERT and RoBERTa. The functions are used in a very similar manner to BertForSequenceClassification and RobertaForSequenceClassification.