Velox: Spatial Memory Makes World Models Efficient
Velox speeds long video world-model rollouts 2.43x at chunk 48 while staying near dense quality.
Velox is a training-free attention router that keeps video world-model rollouts responsive by selecting historical context with camera frustum slabs derived from streaming geometry. It matches FlashAttention-style token-block tiling and avoids the growing KV-cache cost of chunked video diffusion, while uncovered regions keep the model-derived route. On an H200 at chunk 48, attention is 4.36 times faster and denoising 2.43 times faster. Across 158 WBench navigation cases, quality stays within 0.15 API-free average points of dense LingBot-Fast-Cam.
- Velox is a training-free attention router using frustum-slab geometry.
- It selects region-level context so KV caches do not grow with rollout length.
- At chunk 48, attention is 4.36x faster and denoising 2.43x faster on an H200.
- Quality stays within 0.15 points of dense LingBot-Fast-Cam on 158 WBench cases.
- Vendors
- NVIDIA
- Products
- H200
- AI models
- VeloxLingBot-Fast-Cam
From the original article
1 min read at huggingface.co
Video world models turn actions and camera motion into long visual rollouts, making video a natural interface for interactive simulation. Such interaction requires responsive generation, yet coherent rollouts must retain visual memory of what has already been generated. Chunked video diffusion models usually keep this context in growing KV caches, so later chunks become progressively slower.
By Qiuhong Shen, Shizun Wang, Xinchao Wang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.