M3Exam
收藏资源简介:
M3Exam是一个专为评估大型语言模型(LLMs)的多语言、多模态和多层次能力而设计的新型基准数据集。该数据集包含来自多个国家的官方考试题目,涵盖9种不同语言,共计12,317个问题,其中约23%的问题需要处理图像才能成功解答。M3Exam通过其多语言性、多模态性和多层次结构,全面评估模型在不同语言、模态和教育水平上的表现。该数据集的应用领域主要集中在评估和追踪LLMs在多语言和多模态环境下的发展,以及理解模型智能在不同教育水平上的表现。
M3Exam is a novel benchmark dataset specifically designed to evaluate the multilingual, multimodal, and multi-level capabilities of large language models (LLMs). This dataset includes official examination questions from multiple countries, covering 9 distinct languages, with a total of 12,317 questions, approximately 23% of which require image processing to solve correctly. Through its multilingual, multimodal and multi-level structure, M3Exam comprehensively assesses model performance across different languages, modalities and educational levels. The primary application areas of this dataset focus on evaluating and tracking the advancement of LLMs in multilingual and multimodal settings, as well as gaining insights into model intelligence performance across varying educational levels.

- 1M3Exam: A Multilingual, Multimodal, Multilevel Benchmark for Examining Large Language Models阿里巴巴达摩院 · 2023年



