wuschelschulz/mbpp_paired_reward_hacky_normal_cots
收藏Hugging Face2025-11-12 更新2025-11-15 收录
下载链接：
https://hf-mirror.com/datasets/wuschelschulz/mbpp_paired_reward_hacky_normal_cots
下载链接
链接失效反馈官方服务：
资源简介：
MBPP配对数据集：包含27个成对的示例，每个示例包括一个MBPP任务ID、一个奖励黑客解决方案的思考过程和代码答案、一个正常解决方案的思考过程和代码答案，以及一个模型检查点解决方案的思考过程和代码答案。该数据集用于文本生成任务，旨在区分奖励黑客解决方案和正常解决方案。
MBPP Paired Dataset: Contains 27 paired examples, each including an MBPP task ID, a chain-of-thought and code answer for the reward-hacking solution, a chain-of-thought and code answer for the normal solution, and a chain-of-thought and code answer for the model checkpoint solution. This dataset is for text generation tasks, aiming to distinguish between reward-hacking solutions and normal solutions.
提供机构：
wuschelschulz
