Logic-RL
Other AI toolsLogic-RL is an open-source project that enhances the reasoning capabilities of large language models through rule-based reinforcement learning, aiming to reproduce the effects of R1 Zero on logic puzzles such as the Knights and Knaves puzzles. The project provides instructions for data preparation and training execution, and includes a demonstration of self-generated data. The repository also mentions Short-RL for stable length control and the release information for ADORA, a scalable paradigm for guided learning trajectories. It is suitable for reference use by those researching rule-based reinforcement learning and logical reasoning training.