facebookresearch/rlcd

Features

• Model Alignment and Feedback - Aligns models using reinforcement learning from contrast distillation.
• Self-Improvement Methods - Aligns models via reinforcement learning from contrastive distillation.