Format-Adapter: Improving Reasoning Capability of LLMs by Adapting Suitable Format

2025-06-29

AI Summary

Large language models (LLMs) exhibit inconsistency in multi-answer generation due to mismatches between their internal reasoning processes and fixed, manually designed reasoning formats. Method: This paper proposes a fully automated, annotation-free adaptive method: (1) LLMs first generate diverse reasoning formats autonomously; (2) an error-driven automatic evaluation mechanism quantitatively assesses, filters, and refines these formats based on reasoning error measurements. Contribution/Results: To our knowledge, this is the first approach enabling task-driven, end-to-end automation of reasoning format construction and selection—eliminating reliance on human-crafted templates. Evaluated on mathematical and commonsense reasoning benchmarks, the method achieves an average accuracy gain of 4.3% over strong baselines, while significantly improving consistency and robustness in multi-answer generation.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Generating and voting multiple answers is an effective method to mitigate reasoning inconsistencies of large language models (LLMs). Prior works have shown that multiple reasoning formats outperform a single format when generating multiple answers. However, previous works using multiple formats rely on formats labeled by humans, which could be unsuitable for all tasks and have high labeling costs. To address this issue, we adapt suitable formats to the given tasks by generating and selecting formats. We first propose how to measure the reasoning error when generating multiple answers. Then, we introduce Format-Adapter, which utilizes LLMs to generate and select suitable reasoning formats by minimizing the error measurement we present. We conduct experiments on math and commonsense reasoning tasks, where Format-Adapter achieves a 4.3% performance improvement on average over previous works, demonstrating the effectiveness.

Problem

• Mitigating reasoning inconsistencies in LLMs
• Reducing human-labeled format dependency
• Improving reasoning accuracy via format adaptation

Innovation

• Generates and votes multiple answers
• Adapts formats by generating and selecting
• Measures reasoning error for format selection