Uni-DPO: A Unified Paradigm for Dynamic Preference Optimization of LLMs

TL;DR

Uni-DPO introduces a dynamic preference optimization framework that adaptively reweights preference pairs based on data quality and model performance, leading to improved efficiency and superior results in training large language models.

Contribution

It proposes a novel unified framework for preference optimization that considers data quality and model performance, enhancing data utilization and model performance.

Findings

Outperforms baseline methods across multiple benchmarks.

Achieves 6.7 points higher on Arena-Hard with Gemma-2-9B-IT.

Demonstrates robustness and generalization across tasks.

Abstract

Direct Preference Optimization (DPO) has emerged as a cornerstone of reinforcement learning from human feedback (RLHF) due to its simplicity and efficiency. However, existing DPO-based methods typically treat all preference pairs equally, overlooking substantial variations in data quality and learning difficulty, which leads to inefficient data utilization and suboptimal performance. To address this limitation, we propose Uni-DPO, a unified dynamic preference optimization framework that jointly considers (a) the inherent quality of preference pairs and (b) the model's evolving performance during training. By adaptively reweighting samples based on both factors, Uni-DPO enables more effective use of preference data and achieves superior performance. Extensive experiments across models and benchmarks demonstrate the effectiveness and generalization of Uni-DPO. On textual tasks, Gemma-2-9B-IT fine-tuned with Uni-DPO surpasses the leading LLM, Claude 3 Opus, by 6.7 points on Arena-Hard. On mathematical and multimodal tasks, Uni-DPO consistently outperforms baseline methods across all benchmarks, providing strong empirical evidence of its effectiveness and robustness.