harryxi/reward-bench-by-subset

--- configs: - config_name: default data_files: - split: chat path: "chat.csv" - split: chat_hard path: "chat_hard.csv" - split: safety path: "safety.csv" - split: reasoning path: "reasoning.csv" --- basically just https://huggingface.co/datasets/allenai/reward-bench数据集配置： - 配置名称：默认（default） 数据文件： - 拆分集：chat，路径："chat.csv" - 拆分集：chat_hard，路径："chat_hard.csv" - 拆分集：safety，路径："safety.csv" - 拆分集：reasoning，路径："reasoning.csv" 本数据集本质上即为AllenAI发布的奖励基准（Reward Bench）数据集，其对应链接为https://huggingface.co/datasets/allenai/reward-bench