arXiv cs.CL· Yanggan Gu, Yuanyi Wang, Zhaoyi Yan, Yiming Zhang, Qi Zhou, Fei Wu, Hongxia Yang·· 9 天前AI 评分24
InfiFPO：通过偏好优化实现大语言模型隐式模型融合
InfiFPO: Implicit Model Fusion via Preference Optimization in Large Language Models
AI 导读
InfiFPO 是一种面向隐式模型融合的偏好优化方法，用序列级合成多源概率的融合源模型替换 DPO 中的参考模型，并引入概率裁剪与最大间隔融合策略。在 11 个基准上，以 Phi-4 为枢轴模型时平均性能从 79.95 提升至 83.33，数学、代码与推理任务提升明显。
正文 · 原文
Abstract:Model fusion combines multiple Large Language Models (LLMs) with different strengths into a more powerful, integrated model through lightweight training methods. Existing works on model fusion focus primarily on supervised fine-tuning (SFT), leaving preference alignment (PA) --a critical phase for enhancing LLM performance--largely unexplored. The current few fusion methods on PA phase, like WRPO, simplify the process by utilizing only response outputs from source models while discarding their probability information. To address this limitation, we propose InfiFPO, a preference optimization method for implicit model fusion. InfiFPO replaces the reference model in Direct Preference Optimization (DPO) with a fused source model that synthesizes multi-source probabilities at the sequence level, circumventing complex vocabulary alignment challenges in previous works and meanwhile maintaining the probability information. By introducing probability clipping and max-margin fusion strategies, InfiFPO enables the pivot model to align with human preferences while effectively distilling knowledge from source models. Comprehensive experiments on 11 widely-used benchmarks demonstrate that InfiFPO consistently outperforms existing model fusion and preference optimization methods. When using Phi-4 as the pivot model, InfiFPO improve its average performance from 79.95 to 83.33 on 11 benchmarks, significantly improving its capabilities in mathematics, coding, and reasoning tasks.
| Subjects: | Machine Learning (cs.LG); Computation and Language (cs.CL) |
| Cite as: | arXiv:2505.13878 [cs.LG] |
| (or arXiv:2505.13878v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2505.13878 arXiv-issued DOI via DataCite |
|
| Journal reference: | NeurIPS 2025 |
Submission history
From: Yanggan Gu [view email]
[v1]
Tue, 20 May 2025 03:32:37 UTC (263 KB)
[v2]
Wed, 22 Oct 2025 13:55:29 UTC (268 KB)
[v3]
Sat, 23 May 2026 08:09:30 UTC (263 KB)
[v4]
Thu, 8 Oct 2026 09:26:06 UTC (268 KB)
来源：arXiv cs.CL · arxiv.org