Fine-grained control over large language models (LLMs) remains a significant challenge, hindering their adaptability to diverse user needs. While Reinforcement Learning from Human Feedback (RLHF) shows promise in aligning LLMs, its reliance on scalar rewards often limits its ability to capture diverse user preferences in real-world applications. To address this limitation, we introduce the Directional Preference Alignment (DPA) framework. Unlike the scalar-reward RLHF, DPA incorporates multi-objective reward modeling to represent diverse preference profiles. Additionally, DPA models user preferences as directions (i.e., unit vectors) in the reward space to achieve user-dependent preference control. Our method involves training a multi-objective reward model and then fine-tuning the LLM with a preference-conditioned variant of Rejection Sampling Finetuning (RSF), an RLHF method adopted by Llama 2. This method enjoys a better performance trade-off across various reward objectives. In comparison with the scalar-reward RLHF, DPA offers users intuitive control over LLM generation: they can arithmetically specify their desired trade-offs (e.g., more helpfulness with less verbosity). We also validate the effectiveness of DPA with real-world alignment experiments on Mistral-7B. Our method provides straightforward arithmetic control over the trade-off between helpfulness and verbosity while maintaining competitive performance with strong baselines such as Direct Preference Optimization (DPO).

翻译：大语言模型的细粒度控制仍然是一个重大挑战，阻碍了其适应多样化用户需求的能力。虽然基于人类反馈的强化学习（RLHF）在模型对齐方面展现出潜力，但其依赖标量奖励的特性往往限制了在真实应用场景中捕捉多样化用户偏好的能力。为克服这一局限，我们提出了定向偏好对齐（DPA）框架。与标量奖励的RLHF不同，DPA采用多目标奖励建模来表征多样化的偏好特征。此外，DPA将用户偏好建模为奖励空间中的方向（即单位向量），从而实现用户相关的偏好控制。我们的方法包括训练多目标奖励模型，然后使用Llama 2采用的RLHF方法——拒绝采样微调（RSF）的偏好条件变体来微调大语言模型。该方法在多种奖励目标之间实现了更优的权衡性能。与标量奖励的RLHF相比，DPA为用户提供了对大模型生成的直观控制：用户可以通过算术方式指定期望的权衡关系（例如，更少冗长但更有帮助性）。我们还在Mistral-7B上通过真实场景的对齐实验验证了DPA的有效性。我们的方法在保持与直接偏好优化（DPO）等强基线方法竞争性性能的同时，实现了对帮助性与冗长度之间权衡的直观算术控制。