Code for CoSENT: Consistent Sentence Embedding via Similarity Ranking
- main.py: main file for CoSENT.
- datasets: dictionary contains all datasets used in our paper.
- loss.py: CoSENT loss.
- arguments.py: configurations for training and evaluation.
- utils.py: tools for reading datasets.
- SentEval.py and senteval: codes for SentEval tasks, senteval is copied from SentEval: evaluation toolkit for sentence embeddings
The implementation of CoSENT is tested under Python 3.9.13, with the following packages installed:
pytorch==1.12.0tqdm==4.64.0transformers==4.23.1sentence-transformers==2.2.2datasets==2.6.1