innodatalabs/rt3-wildguardmix
收藏Hugging Face2025-03-25 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/innodatalabs/rt3-wildguardmix
下载链接
链接失效反馈官方服务：
资源简介：
Wildjailbreak数据集是一个红队训练数据集，用于训练模型识别和响应不同的安全风险类别。数据集包含了角色、内容和预期标签等信息，用于判定消息内容是否安全，并根据风险类别给出相应的响应。数据集的样本包括系统消息、用户消息和助手消息，并带有预期安全标签。
The Wildjailbreak dataset is a red-teaming training dataset designed to train models to identify and respond to different safety risk categories. It contains information such as roles, content, and expected labels to determine whether the content of messages is safe and to provide appropriate responses based on risk categories. The dataset samples include system messages, user messages, and assistant messages, each with an expected safety label.
提供机构：
innodatalabs
