Sparse rollout training stabilizes reasoning-heavy language model learning

Original: Sparrow: Sparse Rollout for Stable and Efficient Long-context RL of Large Language Models

Source: arxiv.org ↗