通过风险控制提升扩散世界模型推理速度，兼顾精度与效率。
WorldDynCache: Risk-Controlled Latent Dynamics Approximation for Diffusion World Model
- 引入潜空间风险估计算法，动态评估跳步带来的误差累积。
- 在不增加推理开销前提下，实现4.92倍（HunyuanVoyager）和2.15倍（Aether）加速。
- 适合对生成质量敏感且需高效推理的多模态未来预测场景。
扩散世界模型能生成高质量未来画面，但重复的Transformer计算使推理速度极慢。现有缓存方法依赖中间特征复用、局部漂移或短历史预测更新，易忽略跨跳步积累的潜空间演化缺陷，以及条件/阶段依赖的演化方向变化。本文提出WorldDynCache，包含两个核心组件：轻量级潜空间转移风险估计算法，通过对比真实锚点处的反事实误差校准预测；以及条件与阶段感知的升维潜空间代理，无需额外Transformer计算即可逼近潜变量演化。在HunyuanVoyager-13B与Aether-5B上，分别实现4.92倍与2.15倍加速，且在WorldScore、PSNR、SSIM、LPIPS指标上均优于对比缓存方法。
原文摘要 · Abstract (English)
Diffusion world models generate high-quality futures, but re- peated transformer evaluations make inference prohibitively slow. Existing caches reuse intermediate features, selectively update tokens, or reuse and extrapolate denoising outputs ac- cording to local drift or short native-space histories. These criteria can miss both approximation-induced latent transition defects that accumulate across skipped steps and phase- or condition-dependent changes in the direction of latent evo- lution. We propose WorldDynCache, a risk-controlled latent dynamics approximation framework with two core compo- nents. First, a lightweight latent-transition risk estimator tracks the accumulated future impact of approximation defects and calibrates its predictions against counterfactual defects ob- served at exact anchors. Second, a condition- and phase- aware lifted latent surrogate approximates latent evolution without extra transformer evaluations. On HunyuanVoyager- 13B and Aether-5B, WorldDynCache achieves 4.92 times and 2.15 times speedups, respectively, while attaining the best gen- eration quality among the compared caching methods across WorldScore, PSNR, SSIM, and LPIPS.
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。