CMPhysBench
收藏资源简介:
CMPhysBench是一个用于评估大型语言模型在凝聚态物理领域问题解决能力的基准数据集。该数据集由超过520个精心策划的问题组成,涵盖了凝聚态物理的代表子领域和基础理论框架,如磁性、超导性、强关联系统等。为了保证对问题解决过程的深入理解,数据集专注于计算问题,要求LLMs独立生成完整的解决方案。此外,通过利用表达式的基于树的表示,我们引入了可扩展表达式编辑距离(SEED)评分,它提供了细粒度(非二进制)的部分学分,并提供了预测与真实情况之间相似性的更准确的评估。
CMPhysBench is a benchmark dataset for evaluating the problem-solving capabilities of large language models (LLMs) in the field of condensed matter physics. This dataset comprises over 520 carefully curated questions, covering representative subfields and foundational theoretical frameworks of condensed matter physics such as magnetism, superconductivity, strongly correlated systems, and more. To enable in-depth understanding of the problem-solving process, the dataset focuses on computational problems and requires LLMs to independently generate complete solutions. Furthermore, by leveraging tree-based representations of mathematical expressions, we introduce the Scalable Expression Edit Distance (SEED) score, which provides fine-grained (non-binary) partial credits and enables more accurate assessment of the similarity between predicted outputs and ground-truth solutions.

- 1通过中国科学院物理研究所 · 2025年



