DSB: Dynamic Sliding Block Scheduling for Diffusion LLMs

2026-02-05

AI Summary

Existing diffusion-based large language models employ fixed chunk scheduling strategies that fail to adapt to the dynamic variations in textual semantic complexity, thereby limiting both generation quality and inference efficiency. This work proposes a Dynamic Sliding Block (DSB) scheduling method, which introduces, for the first time, a training-free dynamic block adjustment mechanism that adaptively aligns with semantic complexity. Furthermore, DSB integrates a novel cache optimization technique—DSB Cache—to enhance key-value (KV) cache management, significantly improving parallel decoding efficiency. Extensive experiments across multiple mainstream models and benchmarks demonstrate consistent gains in both generation quality and inference speed, confirming the method’s generality and effectiveness.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Diffusion large language models (dLLMs) have emerged as a promising alternative for text generation, distinguished by their native support for parallel decoding. In practice, block inference is crucial for avoiding order misalignment in global bidirectional decoding and improving output quality. However, the widely-used fixed, predefined block (naive) schedule is agnostic to semantic difficulty, making it a suboptimal strategy for both quality and efficiency: it can force premature commitments to uncertain positions while delaying easy positions near block boundaries. In this work, we analyze the limitations of naive block scheduling and disclose the importance of dynamically adapting the schedule to semantic difficulty for reliable and efficient inference. Motivated by this, we propose Dynamic Sliding Block (DSB), a training-free block scheduling method that uses a sliding block with a dynamic size to overcome the rigidity of the naive block. To further improve efficiency, we introduce DSB Cache, a training-free KV-cache mechanism tailored to DSB. Extensive experiments across multiple models and benchmarks demonstrate that DSB, together with DSB Cache, consistently improves both generation quality and inference efficiency for dLLMs. Code is released at https://github.com/lizhuo-luo/DSB.

Problem

• diffusion LLMs
• block scheduling
• semantic difficulty
• parallel decoding
• inference efficiency

Innovation

• Dynamic Sliding Block
• Diffusion LLMs
• Block Scheduling
• Training-free Inference
• KV-cache Optimization