Breaking Block Boundaries: Anchor-based History-stable Decoding for Diffusion Large Language Models

TL;DR

This paper introduces Anchor-based History-stable Decoding (AHD), a dynamic, training-free decoding strategy for diffusion large language models that enhances efficiency and performance by monitoring token stability in real time.

Contribution

The paper presents AHD, a novel decoding method that leverages real-time stability monitoring to improve decoding speed and accuracy without additional training.

Findings

AHD reduces decoding steps by 80% on the BBH benchmark.

AHD improves performance by 3.67% on the BBH benchmark.

AHD reverses performance degradation in existing acceleration strategies.

Abstract

Diffusion Large Language Models (dLLMs) have recently become a promising alternative to autoregressive large language models (ARMs). Semi-autoregressive (Semi-AR) decoding is widely employed in base dLLMs and advanced decoding strategies due to its superior performance. However, our observations reveal that Semi-AR decoding suffers from inherent block constraints, which cause the decoding of many cross-block stable tokens to be unnecessarily delayed. To address this challenge, we systematically investigate the identification of stable tokens and present three key findings: (1) naive lookahead decoding is unreliable, (2) token stability closely correlates with convergence trend, and (3) historical information is isolated. Building on these insights, we propose Anchor-based History-stable Decoding (AHD), a training-free, plug-and-play dynamic decoding strategy. Specifically, AHD monitors the stability trend of tokens in real time through dynamic anchors. Once a token reaches stability, it initiates early cross-block decoding to enhance efficiency and performance. Extensive experiments across language, vision-language, and audio-language domains demonstrate that AHD simultaneously improves both performance and inference efficiency. Notably, AHD effectively reverses the performance degradation typically observed in existing advanced decoding acceleration strategies. For instance, on the BBH benchmark, our approach reduces decoding steps by 80% while improving performance by 3.67%.