reinforcement-distillation-negative-signals

Extract maximum value from limited reasoning traces by leveraging both successful and failed examples through REINFORCE-style distillation.

Independent GitHub snapshot · Automated classification · Original documentation governs