WorldCache: Accelerating World Models for Free via Heterogeneous Token Caching
Researchers have developed WorldCache, a new caching framework designed to accelerate diffusion-based world models. This system addresses challenges like token heterogeneity and non-uniform temporal dynamics inherent in these models. By employing curvature-guided heterogeneous token prediction and chaotic-prioritized adaptive skipping, WorldCache achieves up to 3.7x speedups while preserving 98% of rollout quality. AI
IMPACT Accelerates diffusion world models, enabling more efficient simulation and long-horizon predictions.