通过重参数化让扩散模型并行推理，大幅提速无需额外模型。
Diffusion Models are Secretly Exchangeable: Parallelizing DDPMs via Autospeculation
- 利用扩散模型与随机定位的关联，发现其增量可交换
- 理论证明可实现近似 $\tilde{O}(K^{1/3})$ 的并行加速
- 无需辅助模型即可实现高效并行采样，适合快速生成场景
去噪扩散概率模型（DDPMs）已成为强大的生成建模工具，但其串行计算特性导致显著的推理延迟。本文利用DDPMs与随机定位之间的联系，证明在适当的重参数化下，DDPM的增量满足交换性。这一通用洞察使得多种自回归模型的性能优化技术可直接迁移至扩散模型。为此，我们提出自动推测解码（Autospeculative Decoding, ASD），将广泛使用的推测解码算法扩展至DDPM，且无需任何辅助草稿模型。理论分析表明，ASD相较于传统的K步串行DDPM可实现$\tilde{O}(K^{1/3})$的并行运行时加速。实际实现也证明，该方法在多个领域显著加速了DDPM的推理过程。
原文摘要 · Abstract (English)
Denoising Diffusion Probabilistic Models (DDPMs) have emerged as powerful tools for generative modeling. However, their sequential computation requirements lead to significant inference-time bottlenecks. In this work, we utilize the connection between DDPMs and Stochastic Localization to prove that, under an appropriate reparametrization, the increments of DDPM satisfy an exchangeability property. This general insight enables near-black-box adaptation of various performance optimization techniques from autoregressive models to the diffusion setting. To demonstrate this, we introduce \emph{Autospeculative Decoding} (ASD), an extension of the widely used speculative decoding algorithm to DDPMs that does not require any auxiliary draft models. Our theoretical analysis shows that ASD achieves a $\tilde{O} (K^{\frac{1}{3}})$ parallel runtime speedup over the $K$ step sequential DDPM. We also demonstrate that a practical implementation of autospeculative decoding accelerates DDPM inference significantly in various domains.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。