pass-at-k-training

Use Pass@k as the reward metric in RL training to balance exploration and exploitation, enabling models to learn diverse sampling strategies while maintaining correctness.

Independent GitHub snapshot · Automated classification · Original documentation governs