ToRA

ToRA, developed by researchers from Tsinghua University and Microsoft, is a series of Tool-integrated Reasoning Agents aimed at addressing challenging mathematical problems. By seamlessly combining natural language reasoning with external computational tools such as computation libraries and symbolic solvers, ToRA enhances the analytical capabilities of AI systems. This integration allows ToRA to effectively tackle complex mathematical reasoning tasks, outperforming existing open-source models on multiple datasets. Notably, ToRA-Code-34B is the first open-source model to achieve over 50% accuracy on the MATH dataset, surpassing GPT-4's Chain-of-Thought results.