amazon/CodePrefBench

CodePreference数据集是一个用于代码偏好学习的资源，包含任务ID、指令提示、两个代码响应和首选响应的标识。数据集适用于代码相关任务，大小为1K到10K之间。

The CodePreference dataset includes the following fields: task_id for uniquely identifying tasks, instruction for code writing prompts, two code response options (choices), and the preference choice (gt_choice), where 0 or 1 indicates the preferred option. This dataset is used to evaluate the performance of code generation models, providing example code for various model evaluations.