Resprompt: Residual Connection Prompting Advances Multi-Step Reasoning in Large Language Models

TL;DR

This paper introduces Residual Connection Prompting (RESPROMPT), a novel method that enhances multi-step reasoning in large language models by reconstructing reasoning graphs within prompts, significantly improving accuracy especially on complex problems.

Contribution

RESPROMPT is the first prompting strategy to incorporate residual connections that explicitly model reasoning graphs, advancing multi-step reasoning capabilities in LLMs.

Findings

RESPROMPT improves reasoning accuracy by 12.5% on LLaMA-65B and 6.8% on LLaMA2-70B.

It outperforms standard CoT prompting on complex questions with five or more reasoning steps.

Extensive ablation studies identify effective ways to build residual connections.

Abstract

Chain-of-thought (CoT) prompting, which offers step-by-step problem-solving rationales, has impressively unlocked the reasoning potential of large language models (LLMs). Yet, the standard CoT is less effective in problems demanding multiple reasoning steps. This limitation arises from the complex reasoning process in multi-step problems: later stages often depend on the results of several steps earlier, not just the results of the immediately preceding step. Such complexities suggest the reasoning process is naturally represented as a graph. The almost linear and straightforward structure of CoT prompting, however, struggles to capture this complex reasoning graph. To address this challenge, we propose Residual Connection Prompting (RESPROMPT), a new prompting strategy that advances multi-step reasoning in LLMs. Our key idea is to reconstruct the reasoning graph within prompts. We achieve this by integrating necessary connections-links present in the reasoning graph but missing in the linear CoT flow-into the prompts. Termed "residual connections", these links are pivotal in morphing the linear CoT structure into a graph representation, effectively capturing the complex reasoning graphs inherent in multi-step problems. We evaluate RESPROMPT on six benchmarks across three diverse domains: math, sequential, and commonsense reasoning. For the open-sourced LLaMA family of models, RESPROMPT yields a significant average reasoning accuracy improvement of 12.5% on LLaMA-65B and 6.8% on LLaMA2-70B. Breakdown analysis further highlights RESPROMPT particularly excels in complex multi-step reasoning: for questions demanding at least five reasoning steps, RESPROMPT outperforms the best CoT based benchmarks by a remarkable average improvement of 21.1% on LLaMA-65B and 14.3% on LLaMA2-70B. Through extensive ablation studies and analyses, we pinpoint how to most effectively build residual connections.