DPO

Also known as: Direct Preference Optimization

Direct Preference Optimization (DPO) aligns a model to preference pairs (chosen vs rejected responses) without training a separate reward model or running PPO.

It reparameterizes the RLHF objective into a classification-style loss on the policy. DPO is widely used as a lighter alternative or complement to classic RLHF.

← Back to AI Glossary