xudongwu/PM10
收藏资源简介:
该数据集是一个用于比较两个文本响应在安全性和质量方面的数据集。每个示例包含一个提示(prompt)、两个可能的响应(response_0和response_1),以及标识每个响应是否安全的布尔值(is_response_0_safe和is_response_1_safe)。此外,数据集还提供了更好响应ID(better_response_id)和更安全响应ID(safer_response_id),用于指示哪个响应在整体质量或安全性上更优。数据集还包括选择(chosen)和拒绝(rejected)字段,可能用于强化学习或偏好建模任务。数据集分为训练集(8000个示例)和评估集(2000个示例),总大小约为15.5MB,适用于自然语言处理中的安全对齐、响应比较和模型训练。
This dataset is designed for comparing two text responses in terms of safety and quality. Each example includes a prompt, two possible responses (response_0 and response_1), and boolean flags indicating the safety of each response (is_response_0_safe and is_response_1_safe). Additionally, the dataset provides better_response_id and safer_response_id to indicate which response is superior in overall quality or safety. It also includes chosen and rejected fields, likely used for reinforcement learning or preference modeling tasks. The dataset is split into a training set (8,000 examples) and an evaluation set (2,000 examples), with a total size of approximately 15.5 MB, suitable for natural language processing applications such as safety alignment, response comparison, and model training.




