DiffCoT: Diffusion-styled Chain-of-Thought Reasoning in LLMs

TL;DR

DiffCoT introduces a diffusion-inspired iterative framework for chain-of-thought reasoning in large language models, enhancing robustness and error correction in multi-step problem solving.

Contribution

It reformulates CoT reasoning as a denoising process with a causal diffusion schedule, enabling correction of intermediate errors and improved reasoning performance.

Findings

DiffCoT outperforms existing CoT methods on multiple benchmarks.

It improves robustness and error correction in multi-step reasoning.

Experiments show consistent gains across diverse models.

Abstract

Chain-of-Thought (CoT) reasoning improves multi-step mathematical problem solving in large language models but remains vulnerable to exposure bias and error accumulation, as early mistakes propagate irreversibly through autoregressive decoding. In this work, we propose DiffCoT, a diffusion-styled CoT framework that reformulates CoT reasoning as an iterative denoising process. DiffCoT integrates diffusion principles at the reasoning-step level via a sliding-window mechanism, enabling unified generation and retrospective correction of intermediate steps while preserving token-level autoregression. To maintain causal consistency, we further introduce a causal diffusion noise schedule that respects the temporal structure of reasoning chains. Extensive experiments on three multi-step CoT reasoning benchmarks across diverse model backbones demonstrate that DiffCoT consistently outperforms existing CoT preference optimization methods, yielding improved robustness and error-correction capability in CoT reasoning.