SESAME-BERT

Code for the paper "SesameBERT: Attention for Anywhere". The code is based on google-research/bert .

Requirements

For running the code follwing dependies are required :

1. google-research/bert
2. pip install numpy, tensorflow

Datasets

All these datasets can be found in https://github.com/nyu-mll/GLUE-baselines

Getting Started

GLUE BENCHMARK 19th worldwide (~2020)

sh sesamebert.sh
sh bert.sh

HANS Evaluation Set

python evaluate_heur_output.py ./sesamebert_output/ADV/ADV.tsv 1>./sesamebert_output/ADV/eval_hans.txt

Cite

@article{Su2019SesameBERT,
  title={SesameBERT},
  author={Ta-Chun Su and Hsiang-Chih Cheng},
  conference = {IEEE DSAA 2020},
  year={2019}
}