reinforcement-distillation-negative-signals

【效率与协作】Extract maximum value from limited reasoning traces by leveraging both successful and failed examples through REINFORCE-style distillation.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准