Reversible Diffusion Decoding for Diffusion Language Models

TL;DR

Reversible Diffusion Decoding (RDD) introduces reversibility into diffusion language models, enabling recovery from stagnation and errors during parallel token generation, thus improving robustness and quality with minimal extra computation.

Contribution

The paper presents RDD, a novel reversible decoding framework that enhances diffusion language models by allowing backtracking and selective reinitialization during generation.

Findings

RDD improves generation robustness and quality.

RDD achieves these benefits with minimal computational overhead.

Experiments validate RDD's effectiveness over baseline models.

Abstract

Diffusion language models enable parallel token generation through block-wise decoding, but their irreversible commitments can lead to stagnation, where the reverse diffusion process fails to make further progress under a suboptimal context.We propose Reversible Diffusion Decoding (RDD), a decoding framework that introduces reversibility into block-wise diffusion generation. RDD detects stagnation as a state-dependent failure of the reverse process and enables efficient backtracking to earlier blocks without recomputation via cached model states. To avoid repeated failure trajectories, RDD applies confidence-guided re-masking to selectively reinitialize uncertain tokens while preserving reliable context.This reversible formulation allows decoding to recover from early commitment errors while maintaining the parallel efficiency of diffusion-based generation. Experiments show that RDD improves generation robustness and quality over baselines with minimal computational overhead.