Mitigating Hallucinations in Large Vision-Language Models with Instruction Contrastive Decoding F i n d i n g s - A C L