MirrorAPI
收藏资源简介:
MirrorAPI数据集是由清华大学计算机科学与技术系、清华大学人工智能研究院等机构的研究人员创建的,包含7,000+个API的请求-响应对,涵盖了49个类别。该数据集用于训练专门的语言模型,以准确模拟真实API响应,作为工具环境的'镜子'。数据集通过从RapidAPI收集真实的请求-响应对,并使用监督微调和链式思维推理来提高模拟保真度。
The MirrorAPI dataset was developed by researchers from institutions including the Department of Computer Science and Technology at Tsinghua University, the Institute for Artificial Intelligence at Tsinghua University, and others. It contains over 7,000 API request-response pairs spanning 49 categories. This dataset is designed for training specialized language models to accurately simulate real API responses, acting as a "mirror" for tool environments. The dataset is constructed by collecting real request-response pairs from RapidAPI, and uses supervised fine-tuning and chain-of-thought reasoning to improve simulation fidelity.

- 1StableToolBench-MirrorAPI: Modeling Tool Environments as Mirrors of 7,000+ Real-World APIs清华大学计算机科学与技术系、清华大学人工智能研究院、RayNeo、Google、多伦多大学、中国人民大学高薪人工智能学院 · 2025年



