Published on:

5 October 2023

Primary Category:

Computation and Language

Paper Authors:

Ke Wang,

Houxing Ren,

Aojun Zhou,

Zimu Lu,

Sichun Luo,

Weikang Shi,

Renrui Zhang,

Linqi Song,

Mingjie Zhan,

Hongsheng Li

•

Presents MathCoder method to enhance math reasoning of LLMs using code

•

Creates MathCodeInstruct dataset with math problems and code solutions

•

Proposes training approach to teach models to generate solutions with language, code, execution

•

MathCoder models achieve SOTA on math datasets among open-source LLMs

Enabling LLMs to solve math problems with code

This paper presents a method to improve the mathematical reasoning abilities of large language models by integrating code generation and execution. The key ideas are creating a dataset of math problems paired with solutions that interleave natural language, code, and execution results, and a training approach that teaches models to produce such solutions.

Enhancing math reasoning in language models with code execution

Teaching Smaller Language Models To Reason Over Unseen Questions

Improving math reasoning via multi-view fine-tuning

Multi-modal math reasoning evaluation

Process-free mathematical reasoning via Monte Carlo Tree Search

Competition problems evaluate reasoning in language models

No comments yet, be the first to start the conversation...

Sign up to comment on this paper