Yang et al. (2025b) identified a gradient anomaly in reinforcement learning training where low-probability tokens generate disproportionately large gradient magnitudes, which suppresses the learning of high-probability tokens.

Authors

• Person: Not available Organization: arXiv
  A Survey on the Theory and Mechanism of Large Language Models

Sources

• A Survey on the Theory and Mechanism of Large Language Models arxiv.org arXiv via serper

Referenced by nodes (1)

• reinforcement learning concept