AgenticMath: Enhancing LLM Reasoning via Agentic-based Math Data Generation

2025-10-22

AI Summary

Existing mathematical reasoning datasets suffer from high answer error rates and low information density, hindering LLM performance. To address this, we propose the first agent-driven, four-stage mathematical data generation framework: (1) seed-based filtering for high-information-density problem selection, (2) multi-agent rephrasing to enhance diversity, (3) chain-of-thought (CoT)-guided high-quality answer generation, and (4) automated evaluation-driven sample optimization. This end-to-end pipeline integrates multi-agent collaboration, CoT reasoning, and dynamic evaluation, enabling effective supervised fine-tuning of 3B–8B models using only 30–60K samples. Experiments demonstrate consistent superiority over strong baselines trained on 400K–2.3M samples across both in-domain (e.g., MATH, AMC) and cross-domain (e.g., GSM8K, SVAMP) mathematical reasoning benchmarks, significantly improving data quality and generalization capability.

Technology Category

• Natural Language Processing: Generation
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Multiagent Systems: Mechanism Design

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

The creation of high-quality datasets to improve Large Language Model (LLM) reasoning remains a significant challenge, as current methods often suffer from generating low-quality/incorrect answers and limited information richness from available data sources. To address this, we propose AgenticMath, a novel agentic pipeline for generating high-quality mathematical question-answer pairs to enhance the supervised fine-tuning of LLMs. Our method operates through four stages: (1) Seed Question Filter that selects questions with high information richness, complexity, and clarity; (2) an Agentic Question Rephrase step that employs a multi-agent system to generate diverse, logically consistent paraphrases; (3) an Answer Augment step where rewrite answers using chain-of-thought reasoning to enhance numerical and logical correctness, without reliance on human-provided labels; and (4) a final Question and Answer Evaluation that retains only the most superior pairs. Extensive experiments demonstrate that, fine-tuning 3B-8B parameter LLMs on AgenticMath generated datasets (comprising only 30-60K math samples) achieves competitive or superior performance on diverse in domain and out-of-domain mathematical reasoning benchmarks compared to baselines trained on much more data (e.g., 400K or 2.3M samples). Our work demonstrates that targeted, high-quality data generation is a more efficient path to improving mathematical reasoning in LLMs than large-scale, low-quality alternatives.

Problem

• Generating high-quality math datasets for LLM reasoning enhancement
• Addressing low-quality answers and limited data richness in training
• Improving mathematical reasoning efficiency through targeted data generation

Innovation

• Agentic pipeline generates math question-answer pairs
• Multi-agent system rephrases questions for diversity
• Chain-of-thought reasoning enhances answer correctness