Accelerating Pre-training of Multimodal LLMs via Chain-of-Sight | Read Paper on Bytez