VL-JEPA: Joint Embedding Predictive Architecture for Vision-Language