Aligning Compound AI Systems via System-level DPO
This work addresses the problem of ensuring effective deployment of multi-component AI systems for real-world applications, representing an incremental advancement in alignment methods.
The paper tackles the challenge of aligning compound AI systems with human preferences by introducing SysDPO, a framework that extends Direct Preference Optimization to enable joint system-level alignment, achieving improved performance in applications like aligning language and diffusion models and LLM collaboration systems.
Compound AI systems, comprising multiple interacting components such as LLMs, foundation models, and external tools, have demonstrated remarkable improvements compared to single models in various tasks. To ensure their effective deployment in real-world applications, aligning these systems with human preferences is crucial. However, aligning the compound system via policy optimization, unlike the alignment of a single model, is challenging for two main reasons: (i) non-differentiable interactions between components make end-to-end gradient-based optimization method inapplicable, and (ii) system-level preferences cannot be directly transformed into component-level preferences. To address these challenges, we first formulate compound AI systems as Directed Acyclic Graphs (DAGs), explicitly modeling both component interactions and the associated data flows. Building on this formulation, we introduce $\textbf{SysDPO}$, a framework that extends Direct Preference Optimization (DPO) to enable joint system-level alignment. We propose two variants, SysDPO-Direct and SysDPO-Sampling, tailored for scenarios depending on whether we construct a system-specific preference dataset. We empirically demonstrate the effectiveness of our approach across two applications: the joint alignment of a language model and a diffusion model, and the joint alignment of an LLM collaboration system.