GeoEval
收藏资源简介:
GeoEval是由中国科学院自动化研究所开发的综合性几何问题解决评估数据集,包含2000个主要问题集、750个反向推理问题集、2000个增强问题集和300个难题集。该数据集涵盖平面几何、立体几何和解析几何等多种类型,支持文本和图形双输入,旨在全面评估大型语言模型和多模态模型在解决几何数学问题上的能力。数据集通过不同难度级别的复杂度评级,为模型性能的精细分析提供了坚实框架。
GeoEval is a comprehensive geometric problem-solving evaluation dataset developed by the Institute of Automation, Chinese Academy of Sciences. It comprises four subsets: 2000 main problem sets, 750 inverse reasoning problem sets, 2000 augmented problem sets, and 300 challenging problem sets. Covering multiple geometry categories including plane geometry, solid geometry and analytic geometry, the dataset supports dual inputs of text and graphics. It is designed to comprehensively evaluate the problem-solving capabilities of large language models (LLMs) and multimodal models for geometric mathematical problems. By adopting complexity ratings across different difficulty levels, it provides a robust framework for fine-grained analysis of model performance.

- 1GeoEval: Benchmark for Evaluating LLMs and Multi-Modal Models on Geometry Problem-Solving中国科学院自动化研究所 · 2024年



