ANGO
收藏资源简介:
ANGO是一个专为中文领域设计的多项选择题评估基准,由中国的研究团队开发。该数据集包含71,149个问题,源自2008至2023年的中国公务员考试,涵盖34个省份的官方及模拟考试。ANGO首次提出关键点分类标准,每个问题对应多个关键点,增强评估结果的可解释性。数据集根据实际人类表现建立可量化的难度标准,将问题分为9个难度级别,为模型训练提供精确指导。此外,ANGO采用专门的抽样策略和新的评估框架,支持快速测试集迭代,减少数据泄露的影响。
ANGO is a multiple-choice evaluation benchmark specifically designed for the Chinese language domain, developed by a Chinese research team. This dataset contains 71,149 questions sourced from Chinese civil service examinations held between 2008 and 2023, covering official and mock examinations from 34 provinces. ANGO is the first to propose a key point classification standard, where each question corresponds to multiple key points, thereby enhancing the interpretability of evaluation results. Moreover, a quantifiable difficulty criterion established based on actual human performance is adopted to categorize questions into 9 difficulty levels, providing precise guidance for model training. Additionally, ANGO employs a specialized sampling strategy and a novel evaluation framework, which enables rapid test set iteration and mitigates the impact of data leakage.




