ATMOSSCI-BENCH10

该数据集是一个基准数据集，由67个问题模板构建而成，旨在评估不同大型语言模型（LLM）在处理大气科学问题时的表现，特别关注多位有效数字的准确性。数据集包含了有效数字变化（低、标准、高）的版本，以分析数值精度对LLM性能的影响。规模上，从模板生成了10个问题测试集，任务是对大型语言模型在大气科学问题上的性能进行评估。

This dataset is a benchmark dataset constructed from 67 question templates, designed to evaluate the performance of various Large Language Models (LLMs) when handling atmospheric science problems, with a particular focus on the accuracy of multi-digit significant figures. The dataset includes versions with varying levels of significant figures (low, standard, high) to analyze the impact of numerical precision on LLM performance. In terms of scale, 10 question test sets are generated from these templates, which are used to assess the performance of LLMs on atmospheric science-related tasks.