Patch-as-Decodable-Token: Towards Unified Multi-Modal Vision Tasks in MLLMs

Published in International Conference on Learning Representations (ICLR), 2026