README.md

---
dataset_info:
- config_name: '250'
  features:
  - name: instruction
    dtype: string
  splits:
  - name: train
    num_bytes: 184018
    num_examples: 250
  download_size: 111927
  dataset_size: 184018
- config_name: '50'
  features:
  - name: instruction
    dtype: string
  splits:
  - name: train
    num_bytes: 35966
    num_examples: 50
  download_size: 27162
  dataset_size: 35966
configs:
- config_name: '250'
  data_files:
  - split: train
    path: 250/train-*
- config_name: '50'
  data_files:
  - split: train
    path: 50/train-*
---

This dataset is used for evaluaation in the paper [SafeChain](https://arxiv.org/abs/2502.12025).

Load the dataset with 250 (full) or 50 (small) split

```
from datasets import load_dataset

instructions = load_dataset('UWNSL/WildJailbreakEval', '250')['train']['instruction']

instructions = load_dataset('UWNSL/WildJailbreakEval', '50')['train']['instruction']
```