paper-with-me

홈 › Papers

A Principled Loss Function for Direct Language Model Alignment

2025-08-10 · Yuandong Tan arxiv

The alignment of large language models (LLMs) with human preferences is commonly achieved through Reinforcement Learning from Human Feedback (RLHF). Direct Preference Optimization (DPO) simplified this paradigm by establishing a direct mapping between the optimal policy and a reward function, eliminating the need for an explicit reward model. However, we argue that the DPO loss function is theoretically misaligned with its own derivation, as it promotes the indefinite maximization of a logits difference, which can lead to training instability and reward hacking. In this paper, we propose a novel loss function derived directly from the RLHF optimality condition. Our proposed loss targets a specific, finite value for the logits difference, which is dictated by the underlying reward, rather than its maximization. We provide a theoretical analysis, including a gradient-based comparison, to demonstrate that our method avoids the large gradients that plague DPO when the probability of dispreferred responses approaches zero. This inherent stability prevents reward hacking and leads to more effective alignment. We validate our approach by fine-tuning a Qwen2.5-7B model, showing significant win-rate improvements over a standard DPO baseline and achieving competitive performance against larger models like Llama-3.1-8B.

📄 PDF Abstract BibTeX arXiv:2508.07137

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

Averaging log-likelihoods in direct alignment

2024-06-27 · Nathan Grinsztajn, Yannis Flet-Berliac, Mohammad Gheshlaghi Azar, Florian Strub 외

To better align Large Language Models (LLMs) with human judgment, Reinforcement Learning from Human Feedback (RLHF) learns a reward model and then optimizes it using regularized RL. Recently, direct alignment methods wer…

Distributionally Robust Direct Preference Optimization

2025-02-04 · Zaiyan Xu, Sushil Vemuri, Kishan Panaganti, Dileep Kalathil 외

A major challenge in aligning large language models (LLMs) with human preferences is the issue of distribution shift. LLM alignment algorithms rely on static preference datasets, assuming that they accurately represent r…

BSO: Safety Alignment Is Density Ratio Matching

2026-05-12 · Tien-Phat Nguyen, Truong Nguyen, Thin Nguyen, Duy Minh Ho Nguyen 외 arxiv

Aligning language models for both helpfulness and safety typically requires complex pipelines-separate reward and cost models, online reinforcement learning, and primal-dual updates. Recent direct preference optimization…

Reinforcement Learning

Why DPO is a Misspecified Estimator and How to Fix It

2025-10-23 · Aditya Gopalan, Sayak Ray Chowdhury, Debangshu Banerjee arxiv

Direct alignment algorithms such as Direct Preference Optimization (DPO) fine-tune models based on preference data, using only supervised learning instead of two-stage reinforcement learning with human feedback (RLHF). W…

Reinforcement Learning

Principled Multimodal Representation Learning

2025-07-23 · Xiaohao Liu, Xiaobo Xia, See-Kiong Ng, Tat-Seng Chua arxiv

Multimodal representation learning seeks to create a unified representation space by integrating diverse data modalities to improve multimodal understanding. Traditional methods often depend on pairwise contrastive learn…

Representation LearningContrastive Learning