arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2501.02795cs.CLcs.CV
收藏

InfiFusion:通过LLM融合增强跨模型推理的统一框架

InfiFusion: A Unified Framework for Enhanced Cross-Model Reasoning via LLM Fusion

Reallm Labs(睿尔姆实验室);The Hong Kong Polytechnic University(香港理工大学);Zhejiang University(浙江大学);Harbin Institute of Technology(哈尔滨工业大学)

Zhaoyi Yan, Yiming Zhang, Baoyi He, Yuhao Fu, Qi Zhou, Zhijie Sang, Chunlin Ji, Shengyu Zhang, Fei Wu, Hongxia Yang

AI总结:

针对传统模型融合方法灵活性和效率受限的问题,提出InfiFusion框架,通过增强ULD的Top-K选择和Logits标准化,采用成对/统一两种融合策略,在11个基准上优于主流模型且训练成本显著降低。

摘要:

我们提出InfiFusion,这是一种高效的训练流程,旨在将多个领域专用大型语言模型(LLMs)整合到单个枢纽模型中,有效利用每个源模型的优势。传统融合方法要么直接合并模型参数,要么依赖带有严格假设的知识蒸馏,限制了灵活性和效率。InfiFusion通过用Top-K选择和Logits标准化增强通用Logit蒸馏(ULD)来克服这些限制。我们提出两种融合策略:成对融合(InfiFusion$_p$),即每个源模型知识被单独蒸馏到枢纽模型后合并;统一融合(InfiFusion$_u$),即所有源模型知识同时被蒸馏到枢纽模型。InfiFusion在涵盖推理、编码、数学和指令遵循任务的11个广泛应用基准上优于Qwen-2.5-14B-Instruct和Phi-4等最先进模型。值得注意的是,InfiFusion在实现卓越性能的同时显著降低了计算成本,仅用160个H800 GPU小时完成完整训练,而传统LLM训练通常需要数百万小时。

英文摘要:

We introduce InfiFusion, an efficient training pipeline designed to integrate multiple domain-specialized Large Language Models (LLMs) into a single pivot model, effectively harnessing the strengths of each source model. Traditional fusion methods either merge model parameters directly or rely on knowledge distillation with rigid assumptions, limiting their flexibility and efficiency. InfiFusion overcomes these limitations by enhancing Universal Logit Distillation (ULD) with Top-K selection and Logits Standardization. We propose two fusion strategies: Pairwise Fusion (InfiFusion$_p$), where each source model knowledge is distilled individually into the pivot model followed by merging and Unified Fusion (InfiFusion$_u$), where knowledge from all source models is distilled simultaneously into the pivot model. InfiFusion outperforms the state-of-the-art models, such as Qwen-2.5-14B-Instruct and Phi-4, across 11 widely applied benchmarks covering reasoning, coding, mathematics, and instruction-following tasks. Notably, InfiFusion achieves this superior performance while significantly reduces computational costs, completing full training with only 160 H800 GPU hours compared to the millions typically required for traditional LLM training.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑