PersonaMath: Boosting Mathematical Reasoning via Persona-Driven Data Augmentation

TL;DR

PersonaMath introduces a persona-driven data augmentation method and a new dataset to significantly improve open-source LLMs' mathematical reasoning, achieving state-of-the-art results with less data.

Contribution

The paper presents a novel persona-driven data augmentation technique and a new dataset, PersonaMathQA, to enhance mathematical reasoning in open-source LLMs.

Findings

PersonaMath-7B achieves 61.2% accuracy on MATH

Outperforms baselines with less data

High dataset quality and diversity

Abstract

While closed-source Large Language Models (LLMs) demonstrate strong mathematical problem-solving abilities, open-source models still face challenges with such tasks. To bridge this gap, we propose a data augmentation approach and introduce PersonaMathQA, a dataset derived from MATH and GSM8K, on which we train the PersonaMath models. Our approach consists of two stages: the first stage focuses on learning from Persona Diversification, and the second stage emphasizes learning from Reflection. In the first stage, we regenerate detailed chain-of-thought (CoT) solutions as instructions using a closed-source LLM and introduce a persona-driven data augmentation technique. This technique innovatively classifies personas based on occupations, significantly enhancing the dataset's diversity and quality. In the second stage, we incorporate reflection to fully leverage more challenging and valuable questions. Evaluation of our PersonaMath models on MATH and GSM8K reveals that the PersonaMath-7B model (based on Qwen2.5-7B) achieves an accuracy of 61.2% on MATH and 87.8% on GSM8K, surpassing all baseline methods and achieving state-of-the-art performance. Notably, our dataset contains only 128.9K data points-merely 32.6% of MetaMathQA and 49.5% of MathInstruct-yet our model outperforms these baselines, demonstrating the high quality and diversity of our dataset, which enables more efficient model training. We open-source the PersonaMathQA dataset, PersonaMath models, and our code for public usage.