arXiv 2512.218572025-12-26cs.CV

视觉自回归模型快速推断的邻接自适应动态草案树

Fast Inference of Visual Autoregressive Model with Adjacency-Adaptive Dynamical Draft Trees

School of Computer Science and Engineering, Southeast University（东南大学计算机科学与工程学院）；Key Laboratory of New Generation Artificial Intelligence Technology and Its Interdisciplinary Applications (Southeast University), Ministry of Education, China（东南大学新一代人工智能技术及交叉应用重点实验室（教育部））；New Laboratory of Pattern Recognition (NLPR), State Key Laboratory of Multimodal Artificial Intelligence Systems (MAIS), Institute of Automation, Chinese Academy of Sciences (CASIA)（模式识别新实验室（NLPR），多模态人工智能系统国家重点实验室（MAIS），中国科学院自动化研究所（CASIA））；School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）；School of Computing and Information Systems, Singapore Management University（新加坡国立大学计算机与信息学院）

Haodong Lei, Hongsong Wang, Xin Geng, Liang Wang, Pan Zhou

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

ADT-Tree通过邻接自适应动态草案树提升视觉自回归模型的推断速度，实现3.13倍和3.05倍的加速。

摘要：

自回归（AR）图像模型能够实现扩散级别的质量，但存在顺序推断的问题，需要大约2000步才能生成一个576x576的图像。投机解码与草案树加速了大语言模型，但在视觉AR模型上表现不佳，因为不同图像区域的token预测难度存在空间差异。我们发现将投机解码应用于视觉AR模型的关键障碍在于：由于不同图像区域预测难度不同，导致草案树之间接受率不一致。我们提出了一种邻接自适应动态草案树（ADT-Tree），这是一种邻接自适应的动态草案树，通过利用相邻token状态和先验接受率动态调整草案树的深度和宽度。ADT-Tree通过水平邻接初始化，然后通过二分法适应来优化深度和宽度，从而在简单区域生成更深的树，在复杂区域生成更宽的树。在MS-COCO 2017和PartiPrompts上的实证评估表明，ADT-Tree分别实现了3.13倍和3.05倍的加速。此外，它能够无缝集成到放松采样方法如LANTERN中，从而进一步加速。代码可在https://github.com/Haodong-Lei-Ray/ADT-Tree上获得。

英文摘要：

Autoregressive (AR) image models achieve diffusion-level quality but suffer from sequential inference, requiring approximately 2,000 steps for a 576x576 image. Speculative decoding with draft trees accelerates LLMs yet underperforms on visual AR models due to spatially varying token prediction difficulty. We identify a key obstacle in applying speculative decoding to visual AR models: inconsistent acceptance rates across draft trees due to varying prediction difficulties in different image regions. We propose Adjacency-Adaptive Dynamical Draft Trees (ADT-Tree), an adjacency-adaptive dynamic draft tree that dynamically adjusts draft tree depth and width by leveraging adjacent token states and prior acceptance rates. ADT-Tree initializes via horizontal adjacency, then refines depth/width via bisectional adaptation, yielding deeper trees in simple regions and wider trees in complex ones. The empirical evaluations on MS-COCO 2017 and PartiPrompts demonstrate that ADT-Tree achieves speedups of 3.13xand 3.05x, respectively. Moreover, it integrates seamlessly with relaxed sampling methods such as LANTERN, enabling further acceleration. Code is available at https://github.com/Haodong-Lei-Ray/ADT-Tree.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。