Step Guided Reasoning: Improving Mathematical Reasoning using Guidance Generation and Step Reasoning

TL;DR

This paper introduces Step Guided Reasoning, a training-free method that enhances mathematical reasoning in large language models by guiding their step-by-step reflection during inference, leading to significant performance improvements.

Contribution

It proposes a novel, training-free framework that improves mathematical reasoning in pre-trained LLMs by incorporating a reflective, step-by-step guidance process during inference.

Findings

Qwen2-72B-Instruct outperforms math-specific models on MMLU-STEM.

Increased average scores in math domain for Qwen2 models.

Significant performance gains without additional training data.

Abstract

Mathematical reasoning has been challenging for large language models (LLMs), and the introduction of step-by-step Chain-of-Thought (CoT) inference has significantly advanced the mathematical capabilities of LLMs. However, current approaches either necessitate extensive inference datasets for training or depend on few-shot methods that frequently compromise computational accuracy. To address these fundamental limitations, we propose Step Guided Reasoning, a novel training-free adaptation framework that efficiently equips general-purpose pre-trained language models with enhanced mathematical reasoning capabilities. In this approach, LLMs reflect on small reasoning steps, similar to how humans deliberate and focus attention on what to do next. By incorporating this reflective process into the inference stage, LLMs can effectively guide their reasoning from one step to the next. Through extensive experiments, we demonstrate the significant effect of Step Guided Reasoning in enhancing mathematical performance in state-of-the-art language models -- Qwen2-72B-Instruct outperforms its math-specific counterpart, Qwen2.5-72B-Math-Instruct, on MMLU-STEM with a score of 90.9%, compared to 87.3%. The average scores of Qwen2-7B-Instruct and Qwen2-72B-Instruct increase from 27.1% to 36. 3% and from 36. 5% to 47.4% in the math domain, respectively.