EP4375950B1 - Pixel-Based Machine-Learned Models for Multimodal Vision-Language Tasks