DPO
Direct preference optimization
Adapting a model using examples of preferred and less-preferred responses. It learns from those comparisons without a separate reinforcement-learning loop of the usual RLHF kind.
By Mega AI Hub Editorial · Training & data