SFT or RL? An Early Investigation into Training R1-Like Reasoning Large Vision-Language Models | Read Paper on Bytez