The paper introduces Comparison-based Preference Optimization (ComPO), a zeroth-order method that aligns large language models with human preferences using comparison oracles instead of direct differentiable loss optimization. It provides theoretical convergence guarantees for both offline and online variants under smoothness, gradient sparsity, and oracle compatibility assumptions, and establishes performance bounds under local coverage and in-distribution reward accuracy. Experiments on several LLMs (Mistral, Llama, Gemma-2, Qwen3, Gemma-3) show that ComPO outperforms existing direct alignment methods, achieving higher length-controlled win rates and diagnostics that suggest mitigation of likelihood displacement.
By Peter Chen, Xi Chen, Wotao Yin, Tianyi Lin
arXiv:2504. 06659v2 Announce Type: replace-cross Abstract: Despite advances in Preference Alignment (PA) for Large Language Models (LLMs), mainstream methods like reinforcement learning with human feedback face notable challenges.
By Xiaohua Feng, Yuyuan Li, Huwei Ji, Jiaming Zhang, Li Zhang, Tianyu Du, Chaochao Chen
arXiv:2509. 23982v2 Announce Type: replace-cross Abstract: Preference alignment is a critical step in making Large Language Models (LLMs) useful and aligned with (human) preferences.
By Lucio La Cava, Andrea Tagarelli
arXiv:2410. 15595v4 Announce Type: replace Abstract: With the rapid advancement of large language models (LLMs), aligning policy models with human preferences has become increasingly critical.
By Wenyi Xiao, Zechuan Wang, Leilei Gan, Shuai Zhao, Zongrui Li, Ruirui Lei, Wanggui He, Luu Anh Tuan, Long Chen, Hao Jiang, Zhou Zhao, Fei Wu
The paper introduces BALIGN, a balanced data selection strategy designed to reduce catastrophic forgetting—referred to as the alignment tax—in large language models during preference-based alignment. By analyzing preference optimization gradients, the authors identify three data-centric features that influence parameter drift: the reference model's log-probability margin, token length differences between chosen and rejected responses, and TF‑IDF similarity to general capability corpora. BALIGN aggregates these features into a composite risk score to filter out high-risk preference samples, thereby preserving foundational capabilities while maintaining alignment gains with minimal computational overhead.
By Minsu Kim, Jianxun Lian, Xing Xie, Steven Euijong Whang
Aligning large language models to human preferences is crucial for real-world deployment but frequently incurs an alignment tax, leading to the catastrophic forgetting of pre-trained general capabilit...
arXiv:2510. 05342v2 Announce Type: replace-cross Abstract: Direct Preference Optimization (DPO) has emerged as a simple and effective method for aligning large language models.
By Hyung Gyu Rho
arXiv:2607. 03248v1 Announce Type: cross Abstract: The alignment of large language models with human preferences is commonly achieved through Reinforcement Learning from Human Feedback or Direct Preference Optimization.
By Jialiang Wang, Xianming Liu, Xiong Zhou, Hui Liu, Haoliang Li
arXiv:2604. 27733v2 Announce Type: replace Abstract: Aligning Large Language Models (LLMs) with human intent, whether through explicit reward modeling or direct methods such as DPO, fundamentally relies on minimizing a surrogate loss as a proxy for the true pairwise ranking objective.
By Mehryar Mohri, Yutao Zhong
arXiv:2609.05899v1 Announce Type: cross
Abstract: Aligning large language models with human preferences remains a challenge, primarily due to the critical role of preference data quality in effective...
By Peng Lai, He Zhu, Zhiwen Ruan, Dongdong Zhang, Yun Chen, Peng Li, Furu Wei, Yang Liu, Guanhua Chen
GroupDPO introduces a memory‑efficient approach to group‑wise direct preference optimization for aligning large language models. By using first‑order linearization with per‑response coefficients, the method decouples samples during backpropagation, dramatically reducing peak memory usage and enabling scalable training with larger groups. Experiments in both offline and online settings show that leveraging multiple responses consistently outperforms single‑pair training, and adding a negative log‑likelihood term on positive responses is essential for performance gains and training stability.
By Jixuan Leng, Si Si, Hsiang-Fu Yu, Vinod Raman, Inderjit S. Dhillon
arXiv:2608.23149v1 Announce Type: cross
Abstract: The alignment of Large Language Models heavily relies on English-centric high-quality preference data, which often leads to suboptimal performance in...
By Seungyoon Lee, Minhyuk Kim, Jungseob Lee, Heuiseok Lim