More Data or Better Data? A Critical Analysis of Data Selection and Synthesis for Mathematical Reasoning
This work addresses the problem of optimizing training data for LLMs in mathematical reasoning, offering practical insights for industrial applications, though it is incremental as it builds on existing data construction methods.
The study analyzed data selection and synthesis techniques for mathematical reasoning in LLMs, finding that structuring data in interpretable formats or distilling from stronger models often outperforms simply increasing data volume, providing actionable guidance for cost-effective data curation.
The reasoning capabilities of Large Language Models (LLMs) play a critical role in many downstream tasks, yet depend strongly on the quality of training data. Despite various proposed data construction methods, their practical utility in real-world pipelines remains underexplored. In this work, we conduct a comprehensive analysis of open-source datasets and data synthesis techniques for mathematical reasoning, evaluating them under a unified pipeline designed to mirror training and deployment scenarios. We further distill effective data selection strategies and identify practical methods suitable for industrial applications. Our findings highlight that structuring data in more interpretable formats, or distilling from stronger models often outweighs simply scaling up data volume. This study provides actionable guidance for integrating training data to enhance LLM capabilities, supporting both cost-effective data curation and scalable model enhancement. We hope this work will inspire further research on how to balance "more data" versus "better data" for real-world reasoning tasks.