LMFusion: Adapting Pretrained Language Models for Multimodal Generation | Read Paper on Bytez