The WizardMath model developed by the Microsoft all-Chinese team is a 70B open-source large model, which uses AI-generated instructions to fine-tune the large model, surpassing other large models such as ChatGPT, Claude Instant 1, and PaLM 2-540B in mathematical abilities. The open-source code and paper for the WizardMath model have been published on GitHub, garnering widespread attention from the academic community and developers. The model employs the Evil Instruction technique, which automatically generates open-domain instructions covering various difficulty levels to enhance the processing capabilities of the large model. The authors have also developed models for other fields based on this method, such as WizardCoder focusing on coding abilities. The research team consists of several Chinese scientists, including researchers from the Microsoft Asia Internet Engineering院 and Microsoft Research.