ms_hackathon_23_agent_train_dev

ModelScope Agent探索让LLM作为中枢，来集成调用多种ModelScope的众多AI模型能力，通过API插件的方式集成多种AI小模型。本数据集为插件的训练数据集，包括3600条的训练集和对应的验证集，测试集。模搭社区Hackathon2023上海，赛道二-赛题3，Agent相关训练和验证相关数据集。

The ModelScope Agent initiative explores leveraging Large Language Models (LLMs) as the central hub to integrate and invoke the capabilities of a vast array of AI models hosted on ModelScope, by integrating multiple small-scale AI models via API plugins. This dataset serves as the training dataset for the plugins, consisting of 3600 training samples along with their corresponding validation and test sets. It is the Agent-related training and validation dataset for Track 2, Problem 3 of the 2023 Shanghai ModelScope Community Hackathon.

搜集汇总

背景与挑战

背景概述

该数据集为ModelScope Agent的插件训练提供支持，包含3600个训练样本及验证和测试集，涵盖通用SFT数据和四类插件。其目的是通过大语言模型整合多模态AI能力，以增强大模型的文本生成并弥补多模态生成短板。数据集以JSON格式存储，每个样本包含id和conversations字段，用于模拟系统、用户和助手之间的交互。