Identifying and Mitigating Position Bias of Multi-image Vision-Language Models | Read Paper on Bytez