标题： 面向局部性的并行解码用于高效的自回归图像生成 显示英文标题

标题： Locality-aware Parallel Decoding for Efficient Autoregressive Image Generation

摘要： 我们提出了一种局部感知并行解码（LPD）方法，以加速自回归图像生成。 传统的自回归图像生成依赖于下一个块的预测，这是一个内存密集型过程，导致高延迟。 现有工作尝试通过转向多块预测来并行化下一个块的预测以加速该过程，但仅实现了有限的并行化。 为了在保持生成质量的同时实现高并行化，我们引入了两项关键技术：（1）灵活并行化的自回归建模，这是一种新颖的架构，能够实现任意生成顺序和并行度。 它使用可学习的位置查询标记来指导目标位置的生成，同时确保同时生成的标记之间的相互可见性，以保证并行解码的一致性。（2）局部感知生成顺序，这是一种新的调度方式，通过形成组来最小化组内依赖性和最大化上下文支持，从而提高生成质量。 通过这些设计，我们在ImageNet类别条件生成上将生成步骤从256减少到20（256$\times$256 res.）和从1024减少到48（512$\times$512 res.），而不会降低质量，并且比之前的并行化自回归模型至少降低了3.4$\times$的延迟。 显示英文摘要

摘要： We present Locality-aware Parallel Decoding (LPD) to accelerate autoregressive image generation. Traditional autoregressive image generation relies on next-patch prediction, a memory-bound process that leads to high latency. Existing works have tried to parallelize next-patch prediction by shifting to multi-patch prediction to accelerate the process, but only achieved limited parallelization. To achieve high parallelization while maintaining generation quality, we introduce two key techniques: (1) Flexible Parallelized Autoregressive Modeling, a novel architecture that enables arbitrary generation ordering and degrees of parallelization. It uses learnable position query tokens to guide generation at target positions while ensuring mutual visibility among concurrently generated tokens for consistent parallel decoding. (2) Locality-aware Generation Ordering, a novel schedule that forms groups to minimize intra-group dependencies and maximize contextual support, enhancing generation quality. With these designs, we reduce the generation steps from 256 to 20 (256$\times$256 res.) and 1024 to 48 (512$\times$512 res.) without compromising quality on the ImageNet class-conditional generation, and achieving at least 3.4$\times$ lower latency than previous parallelized autoregressive models.