MathMixup: Revolutionizing LLM Mathematical Reasoning with Smart Data and Learning

MathMixup introduces a groundbreaking data synthesis approach to fuel the mathematical reasoning capabilities of Large Language Models (LLMs). This innovative method generates high-quality, difficulty-controllable problems, which are crucial for effective curriculum learning and improving LLM performance. The resulting MathMixupQA dataset, alongside its curriculum learning strategy, promises to significantly boost LLMs' mathematical prowess.

• •MathMixup employs a novel data synthesis paradigm for generating mathematical reasoning problems.
• •The approach incorporates automated self-checking and manual screening to ensure data quality.
• •Fine-tuned Qwen2.5-7B achieves state-of-the-art results on mathematical benchmarks.