prem-research/guardrail_samples

Prem Studio Guardrail Datasets包含两个密切相关的安全/护栏数据集，用于在Prem Studio中训练小型安全模型，风格类似于Llama Guard。这两个数据集分别用于检测用户消息和代理/助手消息中的不安全内容。每个数据集都遵循相同的模式，包括系统提示、用户消息以及最后的助手消息，后者是安全评估的金标准。数据集包含14个类别的安全政策。

Prem Studio Guardrail Datasets contain two closely related safety/guardrail datasets used in Prem Studio to train small safety models in the style of Llama Guard. These datasets are for detecting unsafe content in user messages and agent/assistant messages respectively. Each dataset follows the same pattern, including a system prompt, a user message, and a final assistant message that serves as the gold standard for safety assessment. The datasets include a 14-category safety policy.