[ICLR 2026] Official PyTorch Implementation of RLP: Reinforcement as a Pretraining Objective - View it on GitHub