Description
Dataset Card for WMDP
The Weapons of Mass Destruction Proxy (WMDP) benchmark is a dataset of multiple-choice questions that serve as a proxy measurement of hazardous knowledge in biosecurity, cybersecurity, and chemical security. WMDP serves two roles: first, as an evaluation for hazardous knowledge in LLMs, and second, as a benchmark for unlearning methods to remove such hazardous knowledge.
See our paper, website, and GitHub for more details!
We implemented the WMDP evaluation… See the full description on the dataset page: https://huggingface.co/datasets/cais/wmdp.
Used by abliterated models
No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.
Metadata
Related datasets
Read further
Download the dataset
Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.
from datasets import load_dataset
ds = load_dataset("cais/wmdp") from huggingface_hub import snapshot_download
snapshot_download(repo_id="cais/wmdp", repo_type="dataset") git clone https://huggingface.co/datasets/cais/wmdp Files by folder
Showing the first 5,000 files. Full list on Hugging Face ↗
top level 3 files 294.1 KB
| .gitattributes | 2.3 KB | file | download |
| dataset.png | 287.1 KB | png | download |
| README.md | 4.6 KB | md | download |
wmdp-bio/ 1 file 258.1 KB
| test-00000-of-00001.parquet | 258.1 KB | parquet | download |
wmdp-chem/ 1 file 50.8 KB
| test-00000-of-00001.parquet | 50.8 KB | parquet | download |
wmdp-cyber/ 1 file 805.7 KB
| test-00000-of-00001.parquet | 805.7 KB | parquet | download |
No file list for this dataset yet.
This repository is private or no longer exists on Hugging Face.
Open on Hugging Face ↗No README stored for this dataset yet.