Search papers, labs, and topics across Lattice.
The paper introduces RLHF-COV and DPO-COV algorithms designed to simultaneously address corrupted preference data, reward overoptimization, and verbosity biases in aligning LLMs with human preferences. The algorithms achieve this by incorporating length regularization and leveraging theoretical guarantees on generalization error rates, even with corrupted data. The authors prove the equivalence of RLHF-COV and DPO-COV, mirroring the known equivalence of vanilla RLHF and DPO, and demonstrate the effectiveness of DPO-COV in offline and online settings.
Finally, a single algorithm, DPO-COV, tackles the trifecta of corrupted preferences, reward overoptimization, and verbosity that plague RLHF and DPO, and it even comes with theoretical guarantees.
Reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) are important techniques to align large language models (LLM) with human preference. However, the quality of RLHF and DPO training is seriously compromised by \textit{\textbf{C}orrupted} preference, reward \textit{\textbf{O}veroptimization}, and bias towards \textit{\textbf{V}erbosity}. To our knowledge, most existing works tackle only one of these important issues, and the few other works require much computation to estimate multiple reward models and lack theoretical guarantee of generalization ability. In this work, we propose RLHF-\textbf{COV} and DPO-\textbf{COV} algorithms that can simultaneously mitigate these three issues, in both offline and online settings. This ability is theoretically demonstrated by obtaining length-regularized generalization error rates for our DPO-COV algorithms trained on corrupted data, which match the best-known rates for simpler cases with clean data and without length regularization. Moreover, our DPO-COV algorithm is simple to implement without reward estimation, and is proved to be equivalent to our RLHF-COV algorithm, which directly implies the equivalence between the vanilla RLHF and DPO algorithms. Experiments demonstrate the effectiveness of our DPO-COV algorithms under both offline and online settings.
