Lumen: Unleashing Versatile Vision-Centric Capabilities of Large Multimodal Models Y ang Jiao