GitHub

SeRL: Self-Play Reinforcement Learning for Large Language Models with Limited Data

The former module generates additional instructions based on the available data at each training step, employing comprehensive online filtering strategies to ensure instruction quality, diversity, and ...