V2PE: Improving Multimodal Long-Context Capability of Vision-Language Models with Variable Visual Position Encoding Paper • 2412.09616 • Published 12 days ago • 1