causal-rewards/ultrafeedback-binarized-preferences-cleaned-neutral
收藏Hugging Face2025-04-16 更新2025-08-09 收录
下载链接：
https://hf-mirror.com/datasets/causal-rewards/ultrafeedback-binarized-preferences-cleaned-neutral
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了一系列的特征字段，包括源文本、提示文本、选中的内容及其角色、选中内容的评分和模型信息，以及被拒绝的内容及其角色、拒绝内容的评分和模型信息，还有一个表示中立态度的布尔字段。数据集被划分为训练集，大小为284,945,388字节，共有60,917个示例。提供了默认配置，包括训练数据的路径。
The dataset includes a series of feature fields such as source text, prompt text, chosen content and its role, rating and model information of the chosen content, content and role that have been rejected, rating and model information of the rejected content, and a boolean field indicating a neutral attitude. The dataset is split into a training set, which is 284,945,388 bytes in size and contains 60,917 examples. A default configuration is provided, including the path to the training data.
提供机构：
causal-rewards
