[CVPR 2026] Official implementation of "On Token's Dilemma: Dynamic MoE with Drift-Aware Token Assignment for Continual Learning of Large Vision Language Models" - View it on GitHub