GAOKAO-Bench
收藏资源简介:
GAOKAO-Bench是由复旦大学计算机科学与技术学院创建的数据集,涵盖了2010至2022年间中国高考的所有科目题目,共计2811题。该数据集包含1781道客观题和1030道主观题,题型多样,包括单选、填空、改错、开放性问题等。数据集通过自动化脚本和人工标注将PDF格式的题目转换为JSON文件,数学公式则转换为LATEX格式。GAOKAO-Bench旨在为大型语言模型提供一个全面且贴近实际应用的评估基准,特别是在解决中国高考相关问题上的表现。
GAOKAO-Bench is a dataset created by the School of Computer Science and Technology at Fudan University, encompassing all subjects' questions from the Chinese Gaokao examination from 2010 to 2022, totaling 2811 questions. The dataset includes 1781 objective questions and 1030 subjective questions, featuring a variety of question types such as single-choice, fill-in-the-blanks, error correction, and open-ended questions. The questions in PDF format were converted to JSON files using automated scripts and manual annotation, with mathematical formulas converted to LATEX format. GAOKAO-Bench is designed to provide a comprehensive and practical assessment benchmark for large language models, particularly in their performance on addressing questions related to the Chinese Gaokao.

- 1Evaluating the Performance of Large Language Models on GAOKAO Benchmark复旦大学计算机科学与技术学院 · 2024年



