Discriminative Policy Optimization for Token-Level Reward Models | Read Paper on Bytez