TRL supports the Direct Preference Optimization (DPO) Trainer for training language models, as described in the paper Direct Preference Optimization: Your Language Model is Secretly a Reward Model by ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results