EasyDeepLearn

ORPO — Odds Ratio Preference Optimization.

hard

Answer

  • Hong et al. 2024.
  • Combines SFT and preference optimization in one loss without a reference model.
  • Loss = SFT loss + λ · LORL_{\mathrm{OR}} where LORL_{\mathrm{OR}} is preference log odds ratio: -log σ(log(odds(yw)/odds(yl))\operatorname{log}(\mathrm{odds}(y_{w}) / \mathrm{odds}(y_{l}))).
  • No reference policy needed, no separate SFT stage.
  • Simpler pipeline than SFT + DPO.
  • Standard for open-source model alignment when reference model is unavailable / undesired.
Check yourself — multiple choice
  • Same as DPO
  • SFT loss + preference log-odds-ratio loss in one step; no reference model needed; simpler than SFT + DPO; open-source alignment when ref model unavailable
  • Random
  • Not real

ORPO: SFT + odds-ratio preference in one loss; no ref model.

#llm#alignment

Practise Reinforcement Learning

214 interview questions in this topic.

Related questions