SPO: Multi-Dimensional Preference Sequential Alignment With Implicit Reward Modeling
This addresses the challenge of managing multiple reward models for human preference alignment in LLMs, offering a novel approach that could enhance model reliability, though it appears incremental in the context of existing alignment methods.
The paper tackles the problem of aligning large language models with multi-dimensional human preferences, such as helpfulness and harmlessness, by proposing Sequential Preference Optimization (SPO), which sequentially fine-tunes models without explicit reward modeling, resulting in significant performance improvements over baselines on various evaluation datasets.
Human preference alignment is critical in building powerful and reliable large language models (LLMs). However, current methods either ignore the multi-dimensionality of human preferences (e.g. helpfulness and harmlessness) or struggle with the complexity of managing multiple reward models. To address these issues, we propose Sequential Preference Optimization (SPO), a method that sequentially fine-tunes LLMs to align with multiple dimensions of human preferences. SPO avoids explicit reward modeling, directly optimizing the models to align with nuanced human preferences. We theoretically derive closed-form optimal SPO policy and loss function. Gradient analysis is conducted to show how SPO manages to fine-tune the LLMs while maintaining alignment on previously optimized dimensions. Empirical results on LLMs of different size and multiple evaluation datasets demonstrate that SPO successfully aligns LLMs across multiple dimensions of human preferences and significantly outperforms the baselines.