Breaking Block Boundaries: Anchor-based History-stable Decoding for Diffusion Large Language Models
This addresses a specific bottleneck in decoding for diffusion large language models, offering a training-free solution that enhances inference efficiency and performance across multiple domains.
The paper tackles the problem of block constraints in semi-autoregressive decoding for diffusion large language models, which delay cross-block stable tokens, and proposes Anchor-based History-stable Decoding (AHD) to improve efficiency and performance, achieving an 80% reduction in decoding steps and a 3.67% performance gain on the BBH benchmark.
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%.