dppo-divergence-policy

【效率与协作】Replace PPO's heuristic ratio-based clipping with Divergence Proximal Policy Optimization (DPPO) that directly constrains policy divergence using either Total Variation or KL, enabling lightweight approximations (Binary, Top-K) for vocabulary-scale computations while improving stability and efficiency.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准