# IPO

A preference-optimization objective that regularizes more strictly than DPO and avoids overfitting to preference labels.

## What it is used for

Preference tuning when DPO overfits.