Earlier Tokens Contribute More: Learning Direct Preference Optimization From Temporal Decay Perspective

Documenti analoghi

• Length Desensitization in Direct Preference Optimization
  di: Liu, Wei, et al.
  Pubblicazione: (2024)
• Libra: Assessing and Improving Reward Model by Learning to Think
  di: Zhou, Meng, et al.
  Pubblicazione: (2025)
• Preference Curriculum: LLMs Should Always Be Pretrained on Their Preferred Data
  di: Zhang, Xuemiao, et al.
  Pubblicazione: (2025)
• Token-weighted Direct Preference Optimization with Attention
  di: Huang, Chengyu, et al.
  Pubblicazione: (2026)
• ReMamba: Equip Mamba with Effective Long-Sequence Modeling
  di: Yuan, Danlong, et al.
  Pubblicazione: (2024)