PreferenceLossType
Enums · AiDotNet.Enums
Types of loss functions for preference optimization methods.
For Beginners
This controls how the model learns from preference data. Start with Sigmoid (standard DPO) and only switch if you have specific needs.
How It Works
Different preference optimization methods use different loss formulations. Each has tradeoffs in terms of stability, sample efficiency, and robustness.
Fields
| Field | Summary |
|---|---|
PreferenceLossType Conservative |
Conservative DPO loss (cDPO). |
PreferenceLossType Hinge |
Hinge loss for preference optimization. |
PreferenceLossType IPO |
Identity Preference Optimization (IPO) loss. |
PreferenceLossType KTO |
Kahneman-Tversky Optimization loss. |
PreferenceLossType OddsRatio |
Odds ratio preference loss (used in ORPO). |
PreferenceLossType Robust |
Robust DPO loss with outlier handling. |
PreferenceLossType Sigmoid |
Standard sigmoid loss used in DPO. |
PreferenceLossType Simple |
Simple preference optimization loss (used in SimPO). |
int value__ |