Preference Optimization by Estimating the Ratio of the Data Distribution
This work addresses the challenge of preference optimization for large language models, offering a method that improves both win rate and entropy compared to existing approaches, though it is incremental as it builds on DPO.
The paper tackles the problem of aligning large language models with human preferences by proposing Bregman preference optimization (BPO), a generalized framework for ratio matching that subsumes direct preference optimization (DPO) and offers tractable objective functions. The result is improved performance, with BPO achieving a 55.9% length-controlled win rate on AlpacaEval2 using Llama-3-8B-Instruct, outperforming other DPO variants.
Direct preference optimization (DPO) is widely used as a simple and stable method for aligning large language models (LLMs) with human preferences. This paper investigates a generalized DPO loss that enables a policy model to match the target policy from a likelihood ratio estimation perspective. The ratio of the target policy provides a unique identification of the policy distribution without relying on reward models or partition functions. This allows the generalized loss to retain both simplicity and theoretical guarantees, which prior work such as $f$-PO fails to achieve simultaneously. We propose Bregman preference optimization (BPO), a generalized framework for ratio matching that provides a family of objective functions achieving target policy optimality. BPO subsumes DPO as a special case and offers tractable forms for all instances, allowing implementation with a few lines of code. We further develop scaled Basu's power divergence (SBA), a gradient scaling method that can be used for BPO instances. The BPO framework complements other DPO variants and is applicable to target policies defined by these variants. In experiments, unlike other probabilistic loss extensions such as $f$-DPO or $f$-PO, which exhibit a trade-off between generation fidelity and diversity, instances of BPO improve both win rate and entropy compared with DPO. When applied to Llama-3-8B-Instruct, BPO achieves state-of-the-art performance among Llama-3-8B backbones, with a 55.9\% length-controlled win rate on AlpacaEval2. Project page: https://github.com/aailab-kaist/BPO.