meituan/Libra-Bench
收藏Hugging Face2025-07-30 更新2025-08-09 收录
下载链接：
https://hf-mirror.com/datasets/meituan/Libra-Bench
下载链接
链接失效反馈官方服务：
资源简介：
Libra Bench是一个推理型奖励模型（RM）基准，包含3740个样本，由一系列挑战性的数学问题和先进的推理模型构建而成，旨在评估逐点判断正确性的准确性。数据集包含样本ID、数学问题、模型生成的响应、正确性标签、响应模型、问题来源和带有COT内容的完整响应等字段。
Libra Bench is a reasoning-oriented reward model (RM) benchmark consisting of 3,740 samples constructed from a diverse collection of challenging mathematical problems and advanced reasoning models, designed to evaluate the pointwise judging accuracy with respect to correctness. The dataset includes fields such as sample ID, mathematical problem, LLM-generated response, correctness label, response model, source of the problem, and full response with COT content.
提供机构：
meituan
