语言大模型 / LLM - arXivDaily 专题

2510.03274 2025-10-07 cs.LG cs.AI 92%

Quant-dLLM: Post-Training Extreme Low-Bit Quantization for Diffusion Large Language Models

Tianao Zhang, Zhiteng Li, Xianglong Yan, Haotong Qin, Yong Guo, Yulun Zhang

机构 * Shanghai Jiao Tong University（上海交通大学）； ETH Zürich（苏黎世联邦理工学院）； South China University of Technology（华南理工大学）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；post-training(title,abstract)；分类 cs.AI、cs.LG

2510.03847 2025-10-07 cs.AI cs.LG 91%

Small Language Models for Agentic Systems: A Survey of Architectures, Capabilities, and Deployment Trade offs

Raghav Sharma, Manan Mehta

专题命中效率与部署：language model(title,abstract)；small language model(title,abstract)；LLM(abstract)；SLM(abstract)

Comments 9 Pages

2509.08825 2025-10-07 cs.CL cs.AI cs.LG 91%

Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation

Joachim Baumann, Paul Röttger, Aleksandra Urman, Albert Wendsjö, Flor Miriam Plaza-del-Arco, Johannes B. Gruber, Dirk Hovy

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；prompting(abstract)

2509.25622 2025-10-07 cs.LG 90%

Layer-wise dynamic rank for compressing large language models

Zhendong Mi, Bian Sun, Grace Li Zhang, Shaoyi Huang

机构 * Stevens Institute of Technology（史蒂文斯理工学院）； Carnegie Mellon University（卡内基梅隆大学）； Technical University of Darmstadt（达姆施塔特技术大学）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；post-training(abstract)

Comments 10 pages, 5 figures

2510.03760 2025-10-07 cs.LG cs.AI 90%

EvoEngineer: Mastering Automated CUDA Kernel Code Evolution with Large Language Models

Ping Guo, Chenyu Zhu, Siyuan Chen, Fei Liu, Xi Lin, Zhichao Lu, Qingfu Zhang

机构 * Department of Computer Science（计算机科学系）； City University of Hong Kong（香港城市大学）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；LLM(abstract)；分类 cs.AI、cs.LG

Comments Under Review of ICLR 2026

2510.03662 2025-10-07 cs.LG cs.AI 88%

Operationalizing Data Minimization for Privacy-Preserving LLM Prompting

Jijie Zhou, Niloofar Mireshghallah, Tianshi Li

机构 * Northeastern University（东北大学）； Carnegie Mellon University（卡内基梅隆大学）

专题命中效率与部署：LLM(title)；prompting(title)；large language model(abstract)；language model(abstract)

2510.04147 2025-10-07 cs.CL 88%

Self Speculative Decoding for Diffusion Large Language Models

Yifeng Gao, Ziang Ji, Yuxuan Wang, Biqing Qi, Hanlin Xu, Linfeng Zhang

机构 * School of AI, Shanghai Jiao Tong University（上海交通大学人工智能学院）； University of Science and Technology of China（中国科学技术大学）； Xidian University（西安电子科技大学）； Shanghai AI Laboratory（上海人工智能实验室）； Huawei Technologies Ltd.（华为技术有限公司）

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)；分类 cs.CL

2510.03515 2025-10-07 cs.LG 88%

RAPID: An Efficient Reinforcement Learning Algorithm for Small Language Models

Lianghuan Huang, Sagnik Anupam, Insup Lee, Shuo Li, Osbert Bastani

机构 * University of Pennsylvania（宾夕法尼亚大学）

专题命中效率与部署：language model(title,abstract)；small language model(title,abstract)；分类 cs.LG

2510.04605 2025-10-07 cs.SE 88%

Exploring the Power of Diffusion Large Language Models for Software Engineering: An Empirical Investigation

Jingyao Zhang, Tianlin Li, Xiaoyu Zhang, Qiang Hu, Bin Shi

专题命中效率与部署：large language model(title,abstract)；language model(title,abstract)

2510.03541 2025-10-07 cs.CL 87%

What is a protest anyway? Codebook conceptualization is still a first-order concern in LLM-era classification

Andrew Halterman, Katherine A. Keith

机构 * Michigan State University（密歇根州立大学）； Williams College（威廉姆斯学院）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；prompting(abstract)

2505.18350 2025-10-07 cs.LG cs.AI cs.CL 87%

How Many Parameters Does Your Task Really Need? Task Specific Pruning with LLM-Sieve

Waleed Reda, Abhinav Jangda, Krishna Chintalapudi

机构 * Microsoft（微软）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2510.03283 2025-10-07 cs.LG cs.AI cs.CL cs.DC 87%

MACE: A Hybrid LLM Serving System with Colocated SLO-aware Continuous Retraining Alignment

Yufei Li, Yu Fu, Yue Dong, Cong Liu

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments 14 pages, 15 figures

2510.04980 2025-10-07 cs.AI cs.CL 86%

LLM-Hanabi: Evaluating Multi-Agent Gameplays with Theory-of-Mind and Rationale Inference in Imperfect Information Collaboration Game

Fangzhou Liang, Tianshi Zheng, Chunkit Chan, Yauwai Yim, Yangqiu Song

机构 * Department of Computer Science and Engineering, HKUST, Hong Kong SAR, China（计算机科学与工程系，香港科技大学，香港特别行政区，中国）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI

Comments EMNLP 2025 Wordplay

2510.03275 2025-10-07 cs.LG cs.AI cs.CV 86%

SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size

Junhao Xia, Ming Zhao, Limin Xiao, Xiujun Zhang

机构 * Dept of Electronic Engineering, Tsinghua University, Beijing, China（电子工程系，清华大学，北京，中国）； Beijing National Research Center for Information Science and Technology, Beijing, China（北京信息科学与技术国家研究中心，北京，中国）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.AI、cs.LG

2507.04724 2025-10-07 cs.MA cs.AI 85%

Who's the Mole? Modeling and Detecting Intention-Hiding Malicious Agents in LLM-Based Multi-Agent Systems

Yizhe Xie, Congcong Zhu, Xinyue Zhang, Tianqing Zhu, Dayong Ye, Minghao Wang, Chi Liu

机构 * Faculty of Data Science, City University of Macau（数据科学学院，澳门城市大学）； Minzu University of China（民族大学）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.AI

2409.09071 2025-10-07 cs.DC cs.AI 85%

Elastic On-Device LLM Service

Wangsong Yin, Rongjie Yi, Daliang Xu, Gang Huang, Mengwei Xu, Xuanzhe Liu

机构 * Peking University, Beijing, China（北京大学）； Beijing University of Posts（北京邮电大学）； Beiyou Shenzhen Institute, Shenzhen, China（贝友深圳研究所）

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)；分类 cs.AI

Comments MobiCom'25

2510.04076 2025-10-07 cs.RO cs.SY eess.SY 85%

From Shadow to Light: Toward Safe and Efficient Policy Learning Across MPC, DeePC, RL, and LLM Agents

Amin Vahidi-Moghaddam, Sayed Pedram Haeri Boroujeni, Iman Jebellat, Ehsan Jebellat, Niloufar Mehrabi, Zhaojian Li

专题命中效率与部署：LLM(title,abstract)；large language model(abstract)；language model(abstract)

2510.03561 2025-10-07 cs.CL cs.AI cs.LG 85%

Reactive Transformer (RxT) -- Stateful Real-Time Processing for Event-Driven Reactive Language Models

Adam Filipek

专题命中效率与部署：language model(title,abstract)；large language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments 25 pages, 13 figures

2502.05370 2025-10-07 cs.LG cs.AI cs.DC 84%

Taming Latency-Memory Trade-Off in MoE-Based LLM Serving via Fine-Grained Expert Offloading

Hanfei Yu, Xingqi Cui, Hong Zhang, Hao Wang, Hao Wang

机构 * Stevens Institute of Technology（史蒂文斯理工学院）； Rice University（Rice大学）； University of Waterloo（多伦多大学）； Rutgers University（罗格斯大学）

专题命中效率与部署：LLM(title)；large language model(abstract)；language model(abstract)；分类 cs.AI、cs.LG

2510.03727 2025-10-07 cs.AI cs.CL cs.CV cs.LG 82%

Bridging the Gap Between Multimodal Foundation Models and World Models

Xuehai He

机构 * Computer Science and Engineering University of California, Santa Cruz（计算机科学与工程大学加州大学圣克ruz分校）

专题命中效率与部署：foundation model(title,abstract)；分类 cs.CL、cs.AI、cs.LG

Comments PhD thesis

2505.16724 2025-10-07 cs.LG cs.AI cs.HC 81%

Advancing Brainwave Modeling with a Codebook-Based Foundation Model

Konstantinos Barmpas, Na Lee, Yannis Panagakis, Dimitrios A. Adamos, Nikolaos Laskaris, Stefanos Zafeiriou

机构 * Imperial College London（帝国理工学院伦敦分校）； Cogitat ； National and Kapodistrian University of Athens（国家与资本主义大学雅典分校）； Archimedes Research Unit（阿奇米德研究单位）； Aristotle University of Thessaloniki（亚里士多德大学塞萨洛尼基分校）

专题命中效率与部署：foundation model(title,abstract)；分类 cs.AI、cs.LG

2501.14794 2025-10-07 cs.DC cs.AI cs.LG 81%

Characterizing Mobile SoC for Accelerating Heterogeneous LLM Inference

Le Chen, Dahu Feng, Erhu Feng, Yingrui Wang, Rong Zhao, Yubin Xia, Pinjie Xu, Haibo Chen

机构 * Institute of Parallel and Distributed Systems, Shanghai Jiao Tong University（并行与分布式系统研究所，上海交通大学）； Tsinghua University（清华大学）； SenseTime Research（商汤科技研究院）

专题命中效率与部署：LLM(title,abstract)；分类 cs.AI、cs.LG

2505.17322 2025-10-07 cs.CL cs.AI cs.LG 80%

From Compression to Expression: A Layerwise Analysis of In-Context Learning

Jiachen Jiang, Yuxin Dong, Jinxin Zhou, Zhihui Zhu

机构 * The corresponding author.（通讯作者）

专题命中效率与部署：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2510.04577 2025-10-07 cs.SD cs.LG cs.MM eess.AS 79%

Language Model Based Text-to-Audio Generation: Anti-Causally Aligned Collaborative Residual Transformers

Juncheng Wang, Chao Xu, Cheng Yu, Zhe Hu, Haoyu Xie, Guoqi Yu, Lei Shang, Shujun Wang

机构 * The Hong Kong Polytechnic University（香港理工大学）； Alibaba Group（阿里巴巴集团）

专题命中效率与部署：language model(title,abstract)；分类 cs.LG

Comments Accepted to EMNLP 2025

2510.04071 2025-10-07 cs.CL 79%

What Makes Diffusion Language Models Super Data Learners?

Zitian Gao, Haoming Luo, Lynx Chen, Jason Klein Liu, Ran Tao, Joey Zhou, Bryan Dai

机构 * Ubiquant

专题命中效率与部署：language model(title,abstract)；分类 cs.CL

Comments Technical report, work in progress

2510.03555 2025-10-07 cs.CV cs.AI 79%

GAS-MIL: Group-Aggregative Selection Multi-Instance Learning for Ensemble of Foundation Models in Digital Pathology Image Analysis

Peiran Quan, Zifan Gu, Zhuo Zhao, Qin Zhou, Donghan M. Yang, Ruichen Rong, Yang Xie, Guanghua Xiao

专题命中效率与部署：foundation model(title,abstract)；分类 cs.AI

2510.03962 2025-10-07 cs.LG cs.AI 79%

SPEAR: Soft Prompt Enhanced Anomaly Recognition for Time Series Data

Hanzhe Wei, Jiajun Wu, Jialin Yang, Henry Leung, Steve Drew

机构 * Department of Electrical and Software Engineering（电气与软件工程系）

专题命中效率与部署：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.AI、cs.LG

Comments Accepted to 2025 IEEE International Conference on Autonomous and Trusted Computing (ATC 2025)

2503.10714 2025-10-07 cs.CL cs.AI 79%

ZSMerge: Zero-Shot KV Cache Compression for Memory-Efficient Long-Context LLMs

Xin Liu, Xudong Wang, Pei Liu, Guoming Tang

机构 * Information Hub, The Hong Kong University of Science and Technology (Guangzhou)（香港科学与技术大学信息中心）； School of Data Science, The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳）数据科学学院）

专题命中效率与部署：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI

2408.05606 2025-10-07 cs.IR cs.AI cs.LG math.OC 79%

Exploring Applications of State Space Models and Advanced Training Techniques in Sequential Recommendations: A Comparative Study on Efficiency and Performance

Mark Obozov, Makar Baderko, Stepan Kulibaba, Nikolay Kutuzov, Alexander Gasnikov

机构 * Phystech（菲斯特技术）

专题命中效率与部署：large language model(abstract)；language model(abstract)；preference optimization(abstract)；分类 cs.AI、cs.LG

Comments arXiv admin note: text overlap with arXiv:2403.07691 by other authors

2502.16706 2025-10-07 cs.LG cs.AI cs.CL cs.SE 78%

DISC: Dynamic Decomposition Improves LLM Inference Scaling

Jonathan Light, Wei Cheng, Benjamin Riviere, Wu Yue, Masafumi Oyamada, Mengdi Wang, Yisong Yue, Santiago Paternain, Haifeng Chen

机构 * Rensselaer Polytechnic Institute（拉特克利夫理工学院）； NEC Laboratories America（NEC美洲实验室）； Princeton University（普林斯顿大学）； California Institute of Technology（加州理工学院）； NEC Corporation（NEC公司）

专题命中效率与部署：LLM(title)；分类 cs.CL、cs.AI、cs.LG

Comments 10 pages, Accepted to NeurIPS 2025 (Conference on Neural Information Processing Systems)

1. 效率与部署 63 篇

Quant-dLLM: Post-Training Extreme Low-Bit Quantization for Diffusion Large Language Models

Small Language Models for Agentic Systems: A Survey of Architectures, Capabilities, and Deployment Trade offs

Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation

Layer-wise dynamic rank for compressing large language models

EvoEngineer: Mastering Automated CUDA Kernel Code Evolution with Large Language Models

Operationalizing Data Minimization for Privacy-Preserving LLM Prompting

Self Speculative Decoding for Diffusion Large Language Models

RAPID: An Efficient Reinforcement Learning Algorithm for Small Language Models

Exploring the Power of Diffusion Large Language Models for Software Engineering: An Empirical Investigation

What is a protest anyway? Codebook conceptualization is still a first-order concern in LLM-era classification

How Many Parameters Does Your Task Really Need? Task Specific Pruning with LLM-Sieve

MACE: A Hybrid LLM Serving System with Colocated SLO-aware Continuous Retraining Alignment

LLM-Hanabi: Evaluating Multi-Agent Gameplays with Theory-of-Mind and Rationale Inference in Imperfect Information Collaboration Game

SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size

Who's the Mole? Modeling and Detecting Intention-Hiding Malicious Agents in LLM-Based Multi-Agent Systems

Elastic On-Device LLM Service

From Shadow to Light: Toward Safe and Efficient Policy Learning Across MPC, DeePC, RL, and LLM Agents

Reactive Transformer (RxT) -- Stateful Real-Time Processing for Event-Driven Reactive Language Models

Taming Latency-Memory Trade-Off in MoE-Based LLM Serving via Fine-Grained Expert Offloading

Bridging the Gap Between Multimodal Foundation Models and World Models

Advancing Brainwave Modeling with a Codebook-Based Foundation Model

Characterizing Mobile SoC for Accelerating Heterogeneous LLM Inference

From Compression to Expression: A Layerwise Analysis of In-Context Learning

Language Model Based Text-to-Audio Generation: Anti-Causally Aligned Collaborative Residual Transformers

What Makes Diffusion Language Models Super Data Learners?

GAS-MIL: Group-Aggregative Selection Multi-Instance Learning for Ensemble of Foundation Models in Digital Pathology Image Analysis

SPEAR: Soft Prompt Enhanced Anomaly Recognition for Time Series Data

ZSMerge: Zero-Shot KV Cache Compression for Memory-Efficient Long-Context LLMs

Exploring Applications of State Space Models and Advanced Training Techniques in Sequential Recommendations: A Comparative Study on Efficiency and Performance

DISC: Dynamic Decomposition Improves LLM Inference Scaling