MatTools
收藏资源简介:
MatTools是一个评估大型语言模型在材料科学工具应用中能力的基准数据集。它由两个互补的组件构成:一个材料模拟工具问答基准,包含69,225个问答对,用于评估LLM对材料科学工具的理解能力;另一个是实际工具使用基准,包含49个任务(138个子任务),要求LLM生成用于材料属性计算的Python代码。数据集旨在帮助开发更有效的AI系统,以解决材料科学和通用科学研究中的问题。
MatTools is a benchmark dataset for evaluating the capabilities of large language models (LLMs) in applications of materials science tools. It comprises two complementary components: one is a materials simulation tool question-answering benchmark containing 69,225 question-answer pairs, designed to assess LLMs' understanding of materials science tools; the other is a practical tool usage benchmark encompassing 49 tasks (138 subtasks), which requires LLMs to generate Python code for materials property calculations. This dataset aims to facilitate the development of more effective AI systems to solve problems in materials science and general scientific research.

- 1MatTools: Benchmarking Large Language Models for Materials Science Tools香港大学 · 2025年



