SmolVLM: Redefining small and efficient multimodal models | Read Paper on Bytez