Token Dynamics: Towards Efficient and Dynamic Video Token Representation for Video Large Language Models