SKILLER enables reusable skill extraction for small language models

SKILLER uses a natural-language-driven reinforcement learning framework to extract procedural skills for small language models. The goal is to enable compact models to execute complex tasks with the reliability of larger, more expensive closed-source models.