Glossary · Model Training

Reward Model

A separate model trained to score how well an output matches human preferences, used to guide reinforcement learning during RLHF-style alignment training.