MergeBench/safety_val

--- dataset_info: features: - name: prompt dtype: string - name: adversarial dtype: bool - name: response dtype: string - name: prompt_harm_label dtype: string - name: response_refusal_label dtype: string - name: response_harm_label dtype: string - name: subcategory dtype: string splits: - name: train num_bytes: 1364313 num_examples: 1000 download_size: 789561 dataset_size: 1364313 configs: - config_name: default data_files: - split: train path: data/train-* ---