VoCo-LLaMA: Towards Vision Compression with Large Language Models | Read Paper on Bytez