Beyond One-Preference-for-All: Multi-Objective Direct Preference Optimization
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: large language model, human feedback, multi objective
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: We present MODPO, a method that fine-tunes language models for diverse preferences without complex reinforcement learning, saving computational effort.
Abstract: Language models (LMs), despite aligning well with an average labeler through reinforcement learning from human feedback (RLHF), may not universally suit diverse human preferences. Recent approaches therefore opt for customization by collecting multi-dimensional feedback and creating distinct rewards for each dimension (e.g., helpfulness, harmlessness, honesty). LMs can then be tailored to different preferences using multi-objective RL (MORL) with different reward weightings. Yet, RL fine-tuning is unstable and resource-heavy, especially for MORLHF with diverse and usually conflicting objectives. In this paper, we present Multi-Objective Direct Preference Optimization (MODPO), an RL-free algorithm that extends Direct Preference Optimization (DPO) for multiple alignment objectives. Essentially, MODPO trains different LMs to represent different collective reward models that combine all objectives with specific weightings. With a simple cross-entropy loss, the LMs optimized against the MODPO objective are analytically the exact solutions of the original MORLHF objective. Empirical results in safety alignment and long-form question answering confirm that MODPO matches or outperforms existing methods, efficiently producing a Pareto-optimal set of LMs that cater to diverse preferences with 3 times less computational resources compared with MORLHF.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1127
Loading