Training & Fine-Tuningaka preference optimization

Preference Learning

Training methods that use human or model comparisons between outputs rather than absolute scores.

Preference learning is any training method that uses human or model comparisons between outputs (A is better than B) rather than absolute scores, including RLHF, DPO, and KTO. This approach captures nuanced quality judgments that are easier for humans to express as comparisons than as numerical ratings.

Where Preference Learning comes up

Training & Fine-Tuningaka preference optimization

Preference Learning

Training methods that use human or model comparisons between outputs rather than absolute scores.

Ask Melo about this ← all terms

Preference Learning

Related terms

Where Preference Learning comes up

Preference Learning

Related terms

Where Preference Learning comes up