concept
advancing artificial intelligence and ai applications is a direct consequence of how well models can reason. for specific tasks such as agent actions, base models cannot reason. so, in plain terms, RELIGN re-aligns a base model into a reasoning model.
one capable of thinking in steps, much like a human would about a complex problem. the result is a more powerful model that improves the performance of anything built on top of it.
relign is a fully open-sourced RL library tailored specifically for the research and development of reasoning engines. It supports state-of-the-art reinforcement learning algorithms like PPO (and soon GRPO) and useful abstractions for Chain of Thought (CoT) and MCTS inference strategies. All these can be evaluated on popular reasoning benchmarks.
Last updated