Preference Optimization for Vision Language Models