Provably Mitigating Corruption, Overoptimization, and Verbosity Simultaneously in Offline and Online RLHF/DPO Alignment
TL;DR: We proposed the first RLHF and DPO algorithms that can simultaneously mitigate corruption, overoptimization and verbosity in large language model (LLM) alignment, with theoretical generalization error rates.
Abstract: Reinforcement learning from human feedback (RLHF) and direct preference optimization (DPO) are emerging and important techniques to align large language models (LLM) with human preference. However, the quality of RLHF and DPO training is seriously compromised by Corrupted preference, reward Overoptimization, and bias towards Verbosity. 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-COV and DPO-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.
Primary Area: Deep Learning->Large Language Models
Keywords: RLHF, DPO, large language model, alignment
Submission Number: 11148
Loading