ScienceQA

该数据集名为ScienceQA基准，包含了21,000个多模态选择题，覆盖了3门学科、26个主题、127个类别以及379项技能，展现了丰富的领域多样性。此外，该数据集被用于评估LLaMA-SciTune在回答多模态推理问题上的性能。其规模达到21,000个问题，任务类型为多模态推理问答。

This dataset, named the ScienceQA benchmark, contains 21,000 multimodal multiple-choice questions, covering 3 disciplines, 26 topics, 127 categories and 379 skills, and boasts rich domain diversity. Additionally, this dataset is utilized to evaluate the performance of LLaMA-SciTune on multimodal reasoning question answering tasks. It comprises 21,000 questions in total, with the task type classified as multimodal reasoning question answering.

搜集汇总

背景与挑战

背景概述

ScienceQA是一个大规模多模态科学问答数据集，包含21,208个多选问题，覆盖多个科学领域，并提供详细的讲解和解释。数据集通过链式思维（CoT）提升语言模型的推理能力，适用于多步推理任务的研究。