prorl-reasoning-expansion

Train LLMs to discover novel reasoning strategies beyond base model capabilities using prolonged RL with KL control, reference policy resets, and diverse task suites.

Independent GitHub snapshot · Automated classification · Original documentation governs