This repository contains source code for the research work described in our AAAI 2021 paper:
A context aware approach for generating natural language attack
Requirements
- Pytorch >= 0.4
- Tensorflow >= 1.0
- Numpy
- Python >= 3.6
- Tensorflow 2.1.0
- TensorflowHub
Dependencies
- Download counter-fitted-vectors from here, unzip it and place the txt file in the main directory.
- Download pretrained BERT for each dataset here, unzip it and place all the folders it in BERT/results/ directory.
- Download top 50 synonym file from here, unzip it and place the txt file in the main directory.
How to Run:
- Run the following command to get classification results.
bash bert.sh
- Run the following command to get inference results.
python run_attack_nli.py
The results will be available in results directory.