Q-Mask: Query-driven Causal Masks for Text Anchoring in OCR-Oriented Vision-Language Models Paper • 2604.00161 • Published Mar 31 • 2