dataset_info:
- config_name: AR
features:- name: prompt
dtype: string - name: SemanticCategory
dtype: string - name: BehaviorID
dtype: string
splits: - name: test
num_bytes: 31033
num_examples: 159
download_size: 17592
dataset_size: 31033
- name: prompt
- config_name: default
features:- name: prompt
dtype: string - name: SemanticCategory
dtype: string - name: BehaviorID
dtype: string
splits: - name: test
num_bytes: 23271
num_examples: 159
download_size: 15315
dataset_size: 23271
configs:
- name: prompt
- config_name: AR
data_files:- split: test
path: AR/test-*
- split: test
- config_name: default
data_files:- split: test
path: data/test-*
- split: test
About
This dataset is taken from the behavior prompts proposed in the test split of HarmBench corresponding to samples with no context.
Citation
If you find HarmBench useful in your research, please consider citing the paper.
title={HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal},
author={Mantas Mazeika and Long Phan and Xuwang Yin and Andy Zou and Zifan Wang and Norman Mu and Elham Sakhaee and Nathaniel Li and Steven Basart and Bo Li and David Forsyth and Dan Hendrycks},
year={2024},
eprint={2402.04249},
archivePrefix={arXiv},
primaryClass={cs.LG}
}