ModelScopeGPT - A multimodal intelligent assistant developed by Alibaba DAMO Academy.

## Overview of ModelScopeGPT ModelScopeGPT is an intelligent assistant developed by Alibaba DAMO Academy. It provides multimodal capabilities, including poetry generation, painting, video creation, and voice playback. It integrates a knowledge base search engine and connects over 1,000 public AI models, supporting multi-turn conversations and API calls. It is specifically designed for the ModelScope community to assist users with model-related queries and creative content generation. ## Key Features of ModelScopeGPT The key features of ModelScopeGPT include: - **Multimodal capabilities**: Supports poetry generation, painting, video creation, and voice playback. - **Model integration**: Connects over 1,000 public AI models across natural language processing (NLP), computer vision (CV), audio processing, and multimodal tasks. - **Knowledge base search**: Integrates a knowledge base search engine to handle model-related questions in the ModelScope community. - **Conversation and API support**: Supports multi-turn conversations and can generate correct API call parameters based on previous interactions, demonstrating generalization to unseen APIs. - **Community support**: Specifically designed for the ModelScope community, enhancing user interaction with AI models. ## Accessing ModelScopeGPT Users can access ModelScopeGPT through the ModelScope platform at the official page: [ModelScopeGPT](https://www.modelscope.cn/studios/damo/ModelScopeGPT). It allows users to ask questions or request content generation. In its first month of release, it processed over 170,000 requests and attracted 40,000 user visits. ## Functions of ModelScopeGPT The main functions of ModelScopeGPT include: - **Creative content generation**: Generates poetry, paintings, videos, and voice content using AI models. - **Community Q&A**: Answers user questions about model usage and knowledge through its integrated knowledge base search engine. - **API interaction**: Supports API calls to perform tasks, such as renewing instances in cloud services, which is particularly useful for enterprise applications. ## Performance Evaluation of ModelScopeGPT The performance of ModelScopeGPT is evaluated using MSAgent-Eval, which employs automatic metrics such as Action EM, ROUGE-L, and Argument F1. Human-AI interaction evaluation is conducted through Agent Arena, accessible at [Agent Arena](https://modelscope.cn/studios/LLMZOO/Chinese-Arena/summary). These evaluations ensure the reliability and accuracy of ModelScopeGPT in practical applications. ## Training Datasets for ModelScopeGPT ModelScopeGPT is trained on the MSAgent-Bench dataset, which contains 598,000 dialogues in both English and Chinese. This dataset covers various API categories, multi-turn conversations, API-oriented Q&A, and API-independent instructions, providing a robust foundation for ModelScopeGPT's functionalities. ### Citation sources: - [ModelScopeGPT](https://www.modelscope.cn/studios/damo/ModelScopeGPT) - Official URL Updated: 2025-03-26