Online Merging Optimizers for Boosting Rewards and Mitigating Tax in Alignment

Documenti analoghi

• Large Language Models are Superpositions of All Characters: Attaining Arbitrary Role-play via Self-Alignment
  di: Lu, Keming, et al.
  Pubblicazione: (2024)
• ProcessBench: Identifying Process Errors in Mathematical Reasoning
  di: Zheng, Chujie, et al.
  Pubblicazione: (2024)
• The Lessons of Developing Process Reward Models in Mathematical Reasoning
  di: Zhang, Zhenru, et al.
  Pubblicazione: (2025)
• Safety Alignment as Continual Learning: Mitigating the Alignment Tax via Orthogonal Gradient Projection
  di: Sun, Guanglong, et al.
  Pubblicazione: (2026)
• Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax
  di: Su, Zeli, et al.
  Pubblicazione: (2026)