GeoRK2：面向扩散 Transformer 加速的几何引导 Runge-Kutta 积分
GeoRK2: Geometry-Guided Runge-Kutta Integration for Diffusion Transformer Acceleration
把二阶 Runge-Kutta 积分器与由噪声预测导出的曲率感知几何流结合，用激活协方差估计特征度量来约束大步长下的误差传播，可直接替换现有 DiT/视频扩散模型的采样器，实现免训练 4-5 倍加速。
原文依据 GeoRK2 couples a second-order Runge-Kutta (RK2) integrator with a curvature-aware geometric flow derived from the model's noise predictions
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论扩散 Transformer 模型能够达到最优的图像合成质量，但其迭代采样过程极其缓慢。尽管减少采样步数可以加速推理，激进的步数调度往往会扭曲中间特征并降低样本保真度。我们提出 GeoRK2，一个连接数值分析与信息几何的免训练框架。GeoRK2 将二阶 Runge-Kutta（RK2）积分器与由模型噪声预测导出的曲率感知几何流相耦合，在流形感知的积分下促进稳定的特征演化。通过利用由激活协方差估计得到的经验特征协方差诱导度量来刻画内在特征几何，GeoRK2 在大步长积分下约束误差传播，同时保持结构保真度。作为一种完全即插即用的方法，GeoRK2 无需重新训练，并与主流预训练扩散 Transformer 兼容。在代表性扩散骨干（如 DiT-XL、HunyuanVideo 和 FLUX.1-dev）上针对图像生成与文本到视频生成任务开展的全面实验表明，GeoRK2 相比基线框架实现 4-5 倍的推理加速，且仅有极小的感知差异（DFID 0.81），验证了其有效性与通用性。
英文原文摘要 1314 字符
Diffusion transformer models deliver state-of-the-art image synthesis quality but suffer from prohibitively slow iterative sampling. Although reducing sampling steps accelerates inference, aggressive step schedules often distort intermediate features and degrade sample fidelity. We present GeoRK2, a training-free framework that bridges numerical analysis and information geometry. GeoRK2 couples a second-order Runge-Kutta (RK2) integrator with a curvature-aware geometric flow derived from the model's noise predictions, promoting stable feature evolution under manifold-aware integration. By leveraging an empirical feature covariance-induced metric estimated from activation covariances to capture intrinsic feature geometry, GeoRK2 constrains error propagation under large-step integration while preserving structural fidelity. As a fully plug-and-play method, GeoRK2 requires no retraining and is compatible with mainstream pre-trained diffusion transformers. Comprehensive experiments on image generation and text-to-video generation tasks across representative diffusion backbones (e.g., DiT-XL, HunyuanVideo, and FLUX.1-dev) demonstrate that GeoRK2 achieves 4-5x faster inference than baseline frameworks with only marginal perceptual differences (DFID 0.81), confirming its effectiveness and generality.