Training Data & Post-TrainingHong et al.
ORPO
AIMade by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made
ORPO — folds a preference penalty into SFT; no reference model needed
Why it matters
A further simplification of the alignment pipeline
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2403.07691Added in Week 35, 2026 · See that week
On the map — ⑫ Training Data line
- Previous stopGRPO
- This recordORPO
- Next stopRunning out of data?