MultiAPI

MultiAPI是由加州大学戴维斯分校IFM实验室开发的一个大型多模态指令-函数基准数据集，包含235个可执行的API函数和2,038个提示。该数据集经过严格的人工精细化处理，确保在大型语言模型评估中的鲁棒性和相关性。MultiAPI旨在扩展大型语言模型在多模态环境中的熟练度，通过提供一个独特的平台来评估工具增强的大型语言模型处理多模态任务的能力。数据集的应用领域包括增强大型语言模型的任务处理能力，特别是在多模态问题解决技能方面，以期在理想化的未来实现人工通用智能。

MultiAPI is a large-scale multimodal instruction-function benchmark dataset developed by the IFM Lab at the University of California, Davis. It comprises 235 executable API functions and 2,038 prompts. This dataset has undergone rigorous manual refinement to guarantee robustness and relevance for evaluations of large language models (LLMs). MultiAPI aims to expand the multimodal proficiency of large language models, providing a unique platform to evaluate the capabilities of tool-augmented large language models when handling multimodal tasks. The dataset’s application scenarios include enhancing the task-processing capabilities of large language models, particularly their multimodal problem-solving skills, with the long-term goal of advancing toward artificial general intelligence (AGI) in an idealized future.

搜集汇总

背景与挑战

背景概述

MultiAPI是一个用于评估大型语言模型多模态能力的数据集，包含JSON、JSONL和Python文件，提供函数调用和提示数据。当前版本为预览版，可能存在错误，最终版本将在审查后发布。