Small Language Models Need Strong Verifiers to Self-Correct Reasoning

TL;DR

This paper investigates how small language models can improve their reasoning accuracy through self-correction, using a novel training pipeline with critiques and a strong verifier, leading to notable performance gains.

Contribution

It introduces a new method for training small LMs to self-correct reasoning errors using self-generated critiques and a strong verifier, enhancing their reasoning capabilities.

Findings

Improved self-correction abilities on multiple reasoning datasets.

Significant gains when paired with a strong GPT-4 verifier.

Limitations observed with weak self-verifiers.

Abstract

Self-correction has emerged as a promising solution to boost the reasoning performance of large language models (LLMs), where LLMs refine their solutions using self-generated critiques that pinpoint the errors. This work explores whether small (<= 13B) language models (LMs) have the ability of self-correction on reasoning tasks with minimal inputs from stronger LMs. We propose a novel pipeline that prompts smaller LMs to collect self-correction data that supports the training of self-refinement abilities. First, we leverage correct solutions to guide the model in critiquing their incorrect responses. Second, the generated critiques, after filtering, are used for supervised fine-tuning of the self-correcting reasoner through solution refinement. Our experimental results show improved self-correction abilities of two models on five datasets spanning math and commonsense reasoning, with notable performance gains when paired with a strong GPT-4-based verifier, though limitations are identified when using a weak self-verifier for determining when to correct.