SimPO: Simple Preference Optimization with a Reference-Free Reward - View it on GitHub