arXivDaily arXiv每日学术速递 周一至周五更新

AI 大模型

语言大模型 / LLM

大语言模型、预训练、指令微调、后训练和语言模型应用。

共收录 11667 信号源:cs.CL, cs.AI, cs.LG

1. 指令微调 11667 篇

2512.23049 2025-12-30 cs.CL 85%

Accelerating Language Model Workflows with Prompt Choreography

通过提示编排加速语言模型工作流

TJ Bai, Jason Eisner

机构 * Johns Hopkins University(约翰霍普金斯大学)

专题命中 指令微调 :language model(title,abstract);LLM(abstract);large language model(abstract);分类 cs.CL

AI总结 提示编排通过动态缓存和并行处理,显著提升多智能体工作流中语言模型的效率与速度

Comments to appear in TACL (final preprint of 2025-10-12); 10 pages + appendices

详情

展开后加载摘要…

URL PDF HTML 收藏
2507.02976 2025-12-30 cs.CR cs.LG cs.SE 85%

How Safe Are AI-Generated Patches? A Large-scale Study on Security Risks in LLM and Agentic Automated Program Repair on SWE-bench

AI生成的补丁有多安全?一项针对LLM和代理自动程序修复在SWE-bench上的大规模安全风险研究

Amirali Sajadi, Kostadin Damevski, Preetha Chatterjee

机构 * Drexel University(德雷塞尔大学) Virginia Commonwealth University(弗吉尼亚共同wealth大学)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.LG

AI总结 本研究评估了LLM和代理框架生成补丁的安全性,发现LLM引入新漏洞,代理工作流在自主权高时也产生漏洞,需考虑上下文因素进行风险评估。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.22742 2025-12-30 cs.DB cs.AI 85%

Robust LLM-based Column Type Annotation via Prompt Augmentation with LoRA Tuning

基于提示增强与LoRA微调的鲁棒列类型标注

Hanze Meng, Jianhao Cao, Rachel Pottinger

机构 * University of British Columbia(不列颠哥伦比亚大学)

专题命中 指令微调 :LLM(title);large language model(abstract);language model(abstract);prompting(abstract)

AI总结 本文提出基于提示增强与LoRA微调的鲁棒列类型标注方法,通过减少可训练参数提升模型稳定性与性能。

Comments 13 pages, 8 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2508.13246 2025-12-30 cs.CR cs.AI 85%

Involuntary Jailbreak: On Self-Prompting Attacks

强制性越狱:关于自我提示攻击

Yangyang Guo, Yangyan Li, Mohan Kankanhalli

机构 * National University of Singapore(新加坡国立大学) Alibaba Group(阿里巴巴集团)

专题命中 指令微调 :prompting(title);LLM(abstract);large language model(abstract);language model(abstract)

AI总结 研究揭示了大型语言模型中一种新的漏洞,通过简单提示策略可强制越狱多数主流模型,促使重新评估防护机制的鲁棒性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2503.17809 2025-12-30 stat.ML cs.LG math.ST stat.TH 85%

Poisson-Process Topic Model for Integrating Knowledge from Pre-trained Language Models

泊松过程主题模型:整合预训练语言模型的知识

Morgane Austern, Yuanchuan Guo, Zheng Tracy Ke, Tianle Liu

机构 * Harvard University(哈佛大学)

专题命中 指令微调 :language model(title,abstract);LLM(abstract);large language model(abstract);分类 cs.LG

AI总结 本文提出基于泊松过程的主题模型,利用预训练语言模型的嵌入信息,改进传统主题建模方法,通过净圆整和核平滑增强,实现更高效的主题估计和收敛性分析。

Comments 96 pages

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.08139 2025-12-23 cs.IT cs.LG math.IT 85%

SCA-LLM: Spectral-Attentive LLM-Based Wireless World Modeling for Agentic Communications

SCA-LLM:基于频谱-注意力的LLM无线世界建模用于智能通信

Ke He, Le He, Lisheng Fan, Xianfu Lei, Thang X. Vu, George K. Karagiannidis, Symeon Chatzinotas

机构 * Interdisciplinary Centre for Security, Reliability and Trust (SnT), University of Luxembourg(安全、可靠性与信任跨学科研究中心(SnT),卢森堡大学) School of Computer Science of Guangzhou University(广州大学计算机科学学院) School of Information Science and Technology, Institute of Mobile Communications, Southwest Jiaotong University(信息科学与技术学院,移动通信研究所,西南交通大学) Department of Electrical and Computer Engineering, Aristotle University of Thessaloniki(电气与计算机工程系,塞萨洛尼基阿瑞斯托大学) Cyber Security Systems and Applied AI Research Center, Lebanese American University (LAU)(网络安全与应用人工智能研究中心,黎巴嫩美国大学)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.LG

AI总结 SCA-LLM通过频谱-注意力适配器将信道状态信息与LLM结合,实现无线世界建模,提升预测性能和零样本泛化能力。

详情

展开后加载摘要…

URL PDF HTML 收藏
2506.09707 2025-12-22 eess.AS cs.CL cs.HC 85%

Fine-Tuning Large Audio-Language Models with LoRA for Precise Temporal Localization of Prolonged Exposure Therapy Elements

通过LoRA微调大音频-语言模型实现精确的延长暴露疗法元素时间定位

Suhas BN, Andrew M. Sherrill, Jyoti Alaparthi, Dominik Mattioli, Rosa I. Arriaga, Chris W. Wiese, Saeed Abdullah

机构 * 1College of Information Sciences \& Technology, The Pennsylvania State University, USA 2Department of Psychiatry \& Behavioral Sciences, Emory University, USA 3School of Interactive Computing, Georgia Institute of Technology, USA 4School of Psychology, Georgia Institute of Technology, USA

专题命中 指令微调 :language model(title,abstract);LLM(abstract);prompting(abstract);分类 cs.CL

AI总结 通过LoRA微调大音频-语言模型,实现对延长暴露疗法关键元素的精确时间定位,提升治疗师忠实性评估的效率和准确性。

Comments 5 pages, 2 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.15957 2025-12-19 cs.CV cs.AI 85%

Seeing is Believing (and Predicting): Context-Aware Multi-Human Behavior Prediction with Vision Language Models

看见即信仰(并预测):基于视觉语言模型的上下文感知多人类行为预测

Utsav Panchal, Yuchen Liu, Luigi Palmieri, Ilche Georgievski, Marco Aiello

机构 * Institute of Architecture of Application Systems, University of Stuttgart, Germany(应用系统建筑研究所,斯图加特大学,德国) Bosch Research, Germany(博世研究,德国)

专题命中 指令微调 :language model(title,abstract);SFT(abstract);preference optimization(abstract);分类 cs.AI

AI总结 CAMP-VLM通过结合视觉语言模型与上下文特征,提升了多人类行为预测的准确性,其在预测精度上比基线模型高66.9%。

Comments Accepted at IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2026

详情

展开后加载摘要…

URL PDF HTML 收藏
2504.12913 2025-12-19 cs.CL 85%

MAIN: Mutual Alignment Is Necessary for instruction tuning

MAIN:互斥对齐是指令微调的必要条件

Fanyi Yang, Jianfeng Liu, Xin Zhang, Haoyu Liu, Xixin Cao, Yuefeng Zhan, Hao Sun, Weiwei Deng, Feng Sun, Qi Zhang

机构 * Peking University(北京大学) Microsoft Corporation(微软公司)

专题命中 指令微调 :instruction tuning(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

AI总结 本文提出MAIN框架,通过互斥约束增强指令与响应的一致性,提升LLM在多种基准上的性能。

Comments Accepted by EMNLP 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.23188 2025-12-16 cs.CL 85%

Diagnose, Localize, Align: A Full-Stack Framework for Reliable LLM Multi-Agent Systems under Instruction Conflicts

诊断、定位、对齐:一种用于在指令冲突下可靠LLM多智能体系统的全栈框架

Guancheng Wan, Leixin Sun, Longxu Dou, Zitong Shi, Fang Wu, Eric Hanchen Jiang, Wenke Huang, Guibin Zhang, Hejia Geng, Xiangru Tang, Zhenfei Yin, Yizhou Sun, Wei Wang

机构 * University of California, Los Angeles(加州大学洛杉矶分校) Sea AI Lab(Sea AI 实验室) Stanford University(斯坦福大学) University of Oxford(牛津大学) Yale University(耶鲁大学) NTU(南洋理工大学) NUS(新加坡国立大学) Boston University(波士顿大学)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

AI总结 本文提出了一种全栈框架,通过诊断、定位和对齐三个阶段提升LLM多智能体系统在指令冲突下的可靠性。

Comments Upon further review, we realized that the version submitted to arXiv was not the final draft and omits crucial results and discussion. To avoid confusion and ensure the integrity of the record, we request withdrawal and will resubmit once the complete work is ready

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.10501 2025-12-15 cs.AI 85%

Zero-shot 3D Map Generation with LLM Agents: A Dual-Agent Architecture for Procedural Content Generation

无监督3D地图生成与LLM代理:一种双代理架构用于程序化内容生成

Lim Chien Her, Ming Yan, Yunshu Bai, Ruihao Li, Hao Zhang

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.AI

AI总结 本文提出一种双代理架构,利用LLM代理实现无监督3D地图生成,通过迭代推理优化参数配置,提升PCG指令遵循能力。

Comments 12 pages, 6 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.10043 2025-12-12 cs.LG 85%

Local LLM Ensembles for Zero-shot Portuguese Named Entity Recognition

本地大语言模型集成用于零样本葡萄牙命名实体识别

João Lucas Luz Lima Sarcinelli, Diego Furtado Silva

机构 * Instituto de Ciências Matemáticas e Computação, Universidade de São Paulo(数学与计算科学学院,圣保罗大学)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.LG

AI总结 本文提出了一种本地大语言模型集成方法,用于零样本葡萄牙命名实体识别,通过选择最优模型组合提升性能,无需标注数据。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.09329 2025-12-11 cs.LG cs.CE 85%

Self Distillation Fine-Tuning of Protein Language Models Improves Versatility in Protein Design

蛋白质语言模型的自我蒸馏微调提升了蛋白质设计的通用性

Amin Tavakoli, Raswanth Murugan, Ozan Gokdemir, Arvind Ramanathan, Frances Arnold, Anima Anandkumar

专题命中 指令微调 :language model(title,abstract);large language model(abstract);SFT(abstract);分类 cs.LG

AI总结 通过自我蒸馏微调提升蛋白质语言模型的通用性,生成更稳定和功能性的蛋白质序列。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.04350 2025-12-05 cs.CL 85%

ClusterFusion: Hybrid Clustering with Embedding Guidance and LLM Adaptation

ClusterFusion: 嵌入引导的混合聚类与LLM适应

Yiming Xu, Yuan Yuan, Vijay Viswanathan, Graham Neubig

机构 * Adobe(Adobe公司) Carnegie Mellon University(卡内基梅隆大学)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

AI总结 ClusterFusion通过将LLM作为聚类核心,结合嵌入引导方法,实现领域知识与用户偏好的整合,提升文本聚类在标准任务和特定领域的性能。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.02892 2025-12-03 cs.CL 85%

Fast-Decoding Diffusion Language Models via Progress-Aware Confidence Schedules

通过进度感知置信度调度实现扩散语言模型的快速解码

Amr Mohamed, Yang Zhang, Michalis Vazirgiannis, Guokan Shang

机构 * MBZUAI Ecole Polytechnique(巴黎高等理工学院)

专题命中 指令微调 :language model(title,abstract);large language model(abstract);instruction tuning(abstract);分类 cs.CL

AI总结 SchED通过进度感知置信度调度实现扩散语言模型的快速解码,显著提升解码效率并保持高准确性。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.02682 2025-12-03 cs.MA cs.AI 85%

Beyond Single-Agent Safety: A Taxonomy of Risks in LLM-to-LLM Interactions

超越单体安全:LLM到LLM交互中的风险分类

Piercosma Bisconti, Marcello Galisai, Federico Pierucci, Marcantonio Bracale, Matteo Prandi

机构 * icaro-lab(ICARO实验室)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.AI

AI总结 本文提出从模型级安全向系统级安全的转变,引入ESRH框架,阐述LLM交互中的集体风险并提出InstitutionalAI架构。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.00231 2025-12-02 cs.SE cs.AI 85%

CodeFlowLM: Incremental Just-In-Time Defect Prediction with Pretrained Language Models and Exploratory Insights into Defect Localization

CodeFlowLM:基于预训练语言模型的增量即需缺陷预测与缺陷定位的探索性洞察

Monique Louise Monteiro, George G. Cabral, Adriano L. I. OLiveira

机构 * cin.ufpe.br(佛罗里达大学佩德罗斯分校计算机学院) Recife-PE – Brazil(巴西佩德罗斯)

专题命中 指令微调 :language model(title,abstract);LLM(abstract);large language model(abstract);分类 cs.AI

AI总结 CodeFlowLM通过增量学习提升即时软件缺陷预测性能,同时探索LLMs在缺陷定位中的潜力与局限。

详情

展开后加载摘要…

URL PDF HTML 收藏
2512.00219 2025-12-02 cs.CL 85%

Minimal-Edit Instruction Tuning for Low-Resource Indic GEC

低资源印地语语法错误纠正的最小编辑指令微调

Akhil Rajeev P

机构 * Indian Heritage Language Computing Team Special and Strategic Projects (SSP) Group(印度遗产语言计算团队特殊与战略项目组) Centre for Development of Advanced Computing (C DAC)(高级计算发展中心)

专题命中 指令微调 :instruction tuning(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

AI总结 本文提出一种无需数据增强的低资源印地语语法错误纠正方法,通过指令微调和保守解码实现高效纠错。

Comments Submitted to AACL-IJCNLP Bhasha Workshop Shared Task1 :GEC

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.10507 2025-11-27 cs.CL 85%

AdvancedIF: Rubric-Based Benchmarking and Reinforcement Learning for Advancing LLM Instruction Following

AdvancedIF:基于规则的基准测试与强化学习以推进大语言模型指令跟随

Yun He, Wenzhe Li, Hejia Zhang, Songlin Li, Karishma Mandyam, Sopan Khosla, Yuanhao Xiong, Nanshu Wang, Xiaoliang Peng, Beibin Li, Shengjie Bi, Shishir G. Patil, Qi Qi, Shengyu Feng, Julian Katz-Samuels, Richard Yuanzhe Pang, Sujan Gonugondla, Hunter Lang, Yue Yu, Yundi Qian, Maryam Fazel-Zarandi, Licheng Yu, Amine Benhalloum, Hany Awadalla, Manaal Faruqui

机构 * Meta Superintelligence Labs(Meta超智能实验室) Princeton University(普林斯顿大学)

专题命中 指令微调 :LLM(title);large language model(abstract);language model(abstract);post-training(abstract)

AI总结 本研究提出AdvancedIF基准和RIFL方法,通过规则生成和奖励塑造提升大语言模型的指令跟随能力,在AdvancedIF上实现6.7%的提升。

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.20689 2025-11-27 q-bio.NC cs.AI 85%

Morality in AI. A plea to embed morality in LLM architectures and frameworks

人工智能中的道德。呼吁将道德嵌入大语言模型架构和框架中

Gunter Bombaerts, Bram Delisse, Uzay Kaymak

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.AI

AI总结 本文呼吁将道德嵌入大语言模型架构和框架中,通过自上而下设计原则,提出技术路径以实现道德处理。

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.19489 2025-11-26 cs.SE cs.AI 85%

Evolution without an Oracle: Driving Effective Evolution with LLM Judges

无Oracle的进化:通过LLM裁判驱动有效进化

Zhe Zhao, Yuheng Yang, Haibin Wen, Xiaojie Qiu, Zaixi Zhang, Qingfu Zhang

机构 * Stanford University(斯坦福大学) City University of Hong Kong(香港城市大学) Princeton University(普林斯顿大学)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.AI

AI总结 本文提出MADE框架,通过问题规范减少LLM反馈噪声,实现无Oracle的进化优化,在软件需求满足和复杂指令跟随任务中取得显著成效。

Comments 14 pages, 5 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.10229 2025-11-14 cs.CL 85%

LangGPS: Language Separability Guided Data Pre-Selection for Joint Multilingual Instruction Tuning

Yangfan Ye, Xiaocheng Feng, Xiachong Feng, Lei Huang, Weitao Ma, Qichen Hong, Yunfei Lu, Duyu Tang, Dandan Tu, Bing Qin

专题命中 指令微调 :instruction tuning(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

Comments AAAI2026 Main Track Accepted

详情

展开后加载摘要…

URL PDF HTML 收藏
2509.02615 2025-11-13 astro-ph.IM cs.AI 85%

Radio Astronomy in the Era of Vision-Language Models: Prompt Sensitivity and Adaptation

Mariia Drozdova, Erica Lastufka, Vitaliy Kinakh, Taras Holotyak, Daniel Schaerer, Slava Voloshynovskiy

机构 * University of Geneva(日内瓦大学)

专题命中 指令微调 :language model(title,abstract);pretraining(abstract);prompting(abstract);分类 cs.AI

Comments Machine Learning and the Physical Sciences Workshop, NeurIPS 2025

详情

展开后加载摘要…

URL PDF HTML 收藏
2511.06490 2025-11-11 cs.CV cs.AI 85%

Zooming into Comics: Region-Aware RL Improves Fine-Grained Comic Understanding in Vision-Language Models

Yule Chen, Yufan Ren, Sabine Süsstrunk

机构 * School of Computer and Communication Sciences, EPFL(瑞士联邦理工学院计算机与通信科学学院) Department of Computer Science and Engineering, Chalmers University of Technology(楚科奇技术大学计算机科学与工程系) IVRL, EPFL(EPFL智能机器人实验室)

专题命中 指令微调 :language model(title,abstract);post-training(abstract);SFT(abstract);分类 cs.AI

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.23595 2025-10-31 cs.AI 85%

Multi-Agent Evolve: LLM Self-Improve through Co-evolution

Yixing Chen, Yiding Wang, Siqi Zhu, Haofei Yu, Tao Feng, Muhan Zhang, Mostofa Patwary, Jiaxuan You

机构 * University of Illinois at Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校) Peking University(北京大学) NVIDIA(英伟达)

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.AI

Comments 29 pages, 4 figures

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.24331 2025-10-29 cs.LG cs.CV 85%

What do vision-language models see in the context? Investigating multimodal in-context learning

Gabriel O. dos Santos, Esther Colombini, Sandra Avila

机构 * Instituto de Computação, Universidade Estadual de Campinas (UNICAMP)(计算机学院,Campinas州立大学(UNICAMP))

专题命中 指令微调 :language model(title,abstract);large language model(abstract);instruction tuning(abstract);分类 cs.LG

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.05024 2025-10-29 cs.LG 85%

Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignment

Nevan Wichers, Aram Ebtekar, Ariana Azarbal, Victor Gillioz, Christine Ye, Emil Ryd, Neil Rathi, Henry Sleight, Alex Mallen, Fabien Roger, Samuel Marks

专题命中 指令微调 :prompting(title,abstract);large language model(abstract);language model(abstract);分类 cs.LG

Comments v2 Updates references. v3 Updates references; Adds IFEval results; Improves appendix readability; Adds author contributions

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.23160 2025-10-28 cs.CL 85%

ENTP: Enhancing Low-Quality SFT Data via Neural-Symbolic Text Purge-Mix

Zile Yang, Ling Li, Na Di, Jinlong Pang, Yao Zhou, Hao Cheng, Bo Han, Jiaheng Wei

机构 * The Hong Kong University of Science and Technology (Guangzhou)(香港科学与技术大学(广州)) University of California, Santa Cruz(加州大学圣克鲁兹分校) Hong Kong Baptist University(香港 Baptist 大学)

专题命中 指令微调 :SFT(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

详情

展开后加载摘要…

URL PDF HTML 收藏
2412.12710 2025-10-28 cs.CL 85%

Enhancing Naturalness in LLM-Generated Utterances through Disfluency Insertion

Syed Zohaib Hassan, Pierre Lison, Pål Halvorsen

专题命中 指令微调 :LLM(title,abstract);large language model(abstract);language model(abstract);分类 cs.CL

Comments 8 pages. Limitations, ethical considerations, and references are additional

详情

展开后加载摘要…

URL PDF HTML 收藏
2510.18250 2025-10-22 cs.AI 85%

ssToken: Self-modulated and Semantic-aware Token Selection for LLM Fine-tuning

Xiaohan Qin, Xiaoxing Wang, Ning Liao, Cancheng Zhang, Xiangdong Zhang, Mingquan Feng, Jingzhi Wang, Junchi Yan

机构 * Shanghai Jiao Tong University(上海交通大学) Shanghai Innovation Institute(上海创新研究院)

专题命中 指令微调 :LLM(title);large language model(abstract);language model(abstract);SFT(abstract)

详情

展开后加载摘要…

URL PDF HTML 收藏