Accelerating Diffusion Language Models Inference via Local Determinism Propagation

ICLR 2026 Conference Submission19207 Authors

Keywords: Diffusion Language Model, Discrete Diffusion Models, Inference Acceleration

TL;DR: We introduce LocalLeap, a training-free method that accelerates diffusion language models through anchor-guided parallel decoding, using high-confidence tokens to safely decode their local neighborhoods while maintaining output quality.

Abstract: Diffusion large language models (dLLMs) represent a significant advancement in text generation, offering parallel token decoding capabilities. However, existing open-source implementations suffer from quality-speed trade-offs that impede their practical deployment. Conservative sampling strategies typically decode only the most confident token per step to ensure quality (i.e., greedy decoding), at the cost of inference efficiency due to repeated redundant refinement iterations--a phenomenon we term delayed decoding. Through systematic analysis of dLLM decoding dynamics, we characterize this delayed decoding behavior and propose a training-free adaptive parallel decoding strategy, named LocalLeap, to address these inefficiencies. LocalLeap is built on two fundamental empirical principles: local determinism propagation centered on high-confidence anchors and progressive spatial consistency decay. By applying these principles, LocalLeap identifies anchors and performs localized relaxed parallel decoding within bounded neighborhoods, achieving substantial inference step reduction through early commitment of already-determined tokens without compromising output quality. Comprehensive evaluation on various benchmarks demonstrates that LocalLeap achieves 6.94$\times$ throughput improvements and reduces decoding steps to just 14.2\% of the original requirement, achieving these gains with negligible performance impact.

Primary Area: generative models

Submission Number: 19207