AlphaDPO: Adaptive Reward Margin for Direct Preference Optimization | Read Paper on Bytez