// ⚖ Alignment & Preference Optimization
ORPO (Odds Ratio Preference Optimization)
Merges SFT and preference optimization in one step without a reference model.
ORPO 2024 Intermediate extended
IN PLAIN WORDS
Adds an odds-ratio penalty between chosen and rejected answers to the SFT loss — single-stage, reference-free.
Used in: Lightweight alignment pipelines
01Original paper
ORPO: Monolithic Preference Optimization without Reference Model
Jiwoo Hong, Noah Lee, James Thorne · 2024
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Learn first
ORPO (Odds Ratio Preference Optimization)
Then you can learn
- Nothing in the atlas builds on it yet
Related:SimPO (Simple Preference Optimization)
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
Official docs
04Where the course teaches it
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
orpo)