VL-JEPA: Joint Embedding Predictive Architecture for Vision-language Paper โข 2512.10942 โข Published 18 days ago โข 10