license: mit
dataset_info:
features:
- name: behavior
dtype: string - name: test_case
dtype: string - name: answer
dtype: string - name: behavior_id
dtype: string - name: functional_category
dtype: string - name: semantic_category
dtype: string
splits: - name: test
num_bytes: 23942571
num_examples: 9605
download_size: 8889803
dataset_size: 23942571
configs: - config_name: default
data_files:- split: test
path: data/test-*
task_categories:
- split: test
- text-generation
- question-answering
language: - en
tags: - harmful-data
- LLM
- LLM-Safety
- HarmBench
pretty_name: HarmGenData@HarmBench
size_categories: - 1K<n<10K
Harmful generations of large language models filtered from HarmBench
All the data here comes from HarmBench.
We filtered the data with a functional category of standard from all harmful outputs obtained from all attack methods they publicly tested against large language models, for reproducing Many-shot jailbreaking.
Reference:
- Mazeika, M., Phan, L., Yin, X., Zou, A., Wang, Z., Mu, N., ... & Hendrycks, D. (2024). Harmbench: A standardized evaluation framework for automated red teaming and robust refusal. arXiv preprint arXiv:2402.04249.
- Anil, C., Durmus, E., Rimsky, N., Sharma, M., Benton, J., Kundu, S., ... & Duvenaud, D. (2024, April). Many-shot jailbreaking. In The Thirty-eighth Annual Conference on Neural Information Processing Systems.