NASK-PIB/harmful_prompts_sample
收藏Hugging Face2025-02-05 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/NASK-PIB/harmful_prompts_sample
下载链接
链接失效反馈官方服务：
资源简介：
Harmful Prompts Sample数据集是一个通过Bielik-11B-v2模型生成的red-teaming查询子集，经过人工审核和注释。该数据集包含category、subcategory_activity、style和style_detail等列，用于生成符合特定标准的有害提示。此外，数据集还包含了LLM预测的类别、提示是否安全、风格是否正确转移以及安全时的类别标签。
The Harmful Prompts Sample dataset is a subset of red-teaming queries generated using the Bielik-11B-v2 model and has been manually reviewed and annotated. It includes columns such as category, subcategory_activity, style, and style_detail for generating harmful prompts that meet specific criteria. Additionally, the dataset contains the LLM predicted category, whether the prompt is safe, whether the style has been correctly transferred, and the category label when it is unsafe.
提供机构：
NASK-PIB
