GCoT-Decoding: Unlocking Deep Reasoning Paths for Universal Question Answering

TL;DR

GCoT-decoding is a novel universal decoding strategy that improves reasoning path generation for large language models across various question-answering tasks, including both fixed and free answer formats.

Contribution

It introduces a two-stage branching method with Fibonacci sampling and heuristic error backtracking, enabling broader applicability and improved accuracy in reasoning paths.

Findings

Maintains strong performance on fixed QA tasks.

Significantly improves results on free QA tasks.

Demonstrates generality across diverse datasets.

Abstract

Chain-of-Thought reasoning can enhance large language models, but it requires manually designed prompts to guide the model. Recently proposed CoT-decoding enables the model to generate CoT-style reasoning paths without prompts, but it is only applicable to problems with fixed answer sets. To address this limitation, we propose a general decoding strategy GCoT-decoding that extends applicability to a broader range of question-answering tasks. GCoT-decoding employs a two-stage branching method combining Fibonacci sampling and heuristic error backtracking to generate candidate decoding paths. It then splits each path into a reasoning span and an answer span to accurately compute path confidence, and finally aggregates semantically similar paths to identify a consensus answer, replacing traditional majority voting. We conduct extensive experiments on six datasets covering both fixed and free QA tasks. Our method not only maintains strong performance on fixed QA but also achieves significant improvements on free QA, demonstrating its generality.