Hugging Face Daily Papers·· 2026-10-15AI score23
Velox: Making world models more efficient with spatial memory
Velox: Spatial Memory Makes World Models Efficient
AI briefing
Velox is a training-free attention router that uses frustum tile correspondences in streaming geometry for region-level context selection, solving the problem that KV caches in chunked video diffusion models slow down as rollouts grow. At the 48th chunk, Velox accelerates attention by 4.36× and denoising by 2.43×, and on the WBench navigation test set across all 158 cases, its average score difference from dense LingBot-Fast-Cam remains within 0.15.
Open original arxiv.org