MuMath-Code: Combining Tool-Use Large Language Models with Multi-perspective Data Augmentation for Mathematical Reasoning
Introduction
MuMath-Code is a series of tool-use math reasoning LLMs, which are trained using a two-stage strategy: firstly trained on MuMath-Data, then on the proposed multi-perspective and code integrated math reasoning dataset called MuMath-Code-Data. Through such a two-stage training strategy, our MuMath-Code models outperforms the state-of-the-art… See the full description on the dataset page: https://huggingface.co/datasets/weihao1/MuMath-Code-Data.
