uclaml/SPIN

Features

• Reinforcement Learning - Self-play fine-tuning to enhance language model performance.
• Self-Improvement Methods - Uses self-play to refine language models without external human data.