Direct Multi-Turn Preference Optimization for Language Agents

TL;DR

This paper introduces DMPO, a novel loss function for multi-turn language agents that improves preference optimization by addressing partition function challenges and length disparities, validated through extensive experiments.

Contribution

It proposes DMPO, a new loss function that enhances multi-turn preference optimization by modifying the RL objective and incorporating length normalization.

Findings

DMPO outperforms existing methods on three multi-turn agent datasets.

Theoretical analysis supports the effectiveness of the proposed loss.

Extensive experiments demonstrate DMPO's superiority in preference optimization.

Abstract

Adapting Large Language Models (LLMs) for agent tasks is critical in developing language agents. Direct Preference Optimization (DPO) is a promising technique for this adaptation with the alleviation of compounding errors, offering a means to directly optimize Reinforcement Learning (RL) objectives. However, applying DPO to multi-turn tasks presents challenges due to the inability to cancel the partition function. Overcoming this obstacle involves making the partition function independent of the current state and addressing length disparities between preferred and dis-preferred trajectories. In this light, we replace the policy constraint with the state-action occupancy measure constraint in the RL objective and add length normalization to the Bradley-Terry model, yielding a novel loss function named DMPO for multi-turn agent tasks with theoretical explanations. Extensive experiments on three multi-turn agent task datasets confirm the effectiveness and superiority of the DMPO loss. The code is available at https://github.com/swt-user/DMPO.