语言大模型 / LLM - arXivDaily 专题

2312.12411 2023-12-20 cs.LG 88%

Future-proofing geotechnics workflows: accelerating problem-solving with large language models

Stephen Wu, Yu Otake, Daijiro Mizutani, Chang Liu, Kotaro Asano, Nana Sato, Hidetoshi Baba, Yusuke Fukunaga, Yosuke Higo, Akiyoshi Kamura, Shinnosuke Kodama, Masataka Metoki, Tomoka Nakamura, Yuto Nakazato, Taiga Saito, Akihiro Shioi, Masahiro Takenobu, Keigo Tsukioka, Ryo Yoshikawa

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

Comments Supplementary information will be available upon request

2312.11701 2023-12-20 eess.SY cs.CL cs.SY 88%

Opportunities and Challenges of Applying Large Language Models in Building Energy Efficiency and Decarbonization Studies: An Exploratory Overview

Liang Zhang, Zhelun Chen

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

2306.14048 2023-12-20 cs.LG 88%

H$_2$O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models

Zhenyu Zhang, Ying Sheng, Tianyi Zhou, Tianlong Chen, Lianmin Zheng, Ruisi Cai, Zhao Song, Yuandong Tian, Christopher Ré, Clark Barrett, Zhangyang Wang, Beidi Chen

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

2312.00407 2023-12-04 cs.CL 88%

CoLLiE: Collaborative Training of Large Language Models in an Efficient Way

Kai Lv, Shuo Zhang, Tianle Gu, Shuhao Xing, Jiawei Hong, Keyu Chen, Xiaoran Liu, Yuqing Yang, Honglin Guo, Tengxiao Liu, Yu Sun, Qipeng Guo, Hang Yan, Xipeng Qiu

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

Comments To appear at EMNLP 2023 Demo; Code is available at https://github.com/OpenLMLab/collie

2311.07014 2023-11-14 cs.CL cs.SD eess.AS 88%

Teach me with a Whisper: Enhancing Large Language Models for Analyzing Spoken Transcripts using Speech Embeddings

Fatema Hasan, Yulong Li, James Foulds, Shimei Pan, Bishwaranjan Bhattacharjee

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

Comments 11 pages

2310.09259 2023-11-03 cs.LG 88%

QUIK: Towards End-to-End 4-Bit Inference on Generative Large Language Models

Saleh Ashkboos, Ilia Markov, Elias Frantar, Tingxuan Zhong, Xincheng Wang, Jie Ren, Torsten Hoefler, Dan Alistarh

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

Comments 16 pages

2310.11453 2023-10-18 cs.CL 88%

BitNet: Scaling 1-bit Transformers for Large Language Models

Hongyu Wang, Shuming Ma, Li Dong, Shaohan Huang, Huaijie Wang, Lingxiao Ma, Fan Yang, Ruiping Wang, Yi Wu, Furu Wei

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

Comments Work in progress

2310.07343 2023-10-12 cs.CL 88%

How Do Large Language Models Capture the Ever-changing World Knowledge? A Review of Recent Advances

Zihan Zhang, Meng Fang, Ling Chen, Mohammad-Reza Namazi-Rad, Jun Wang

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

Comments EMNLP 2023 main conference, paper link at https://github.com/hyintell/awesome-refreshing-llms

2308.09376 2023-08-21 cs.CL cs.HC 88%

Leveraging Large Language Models for DRL-Based Anti-Jamming Strategies in Zero Touch Networks

Abubakar S. Ali, Dimitrios Michael Manias, Abdallah Shami, Sami Muhaidat

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

2308.04241 2023-08-14 cs.AI cs.CY 88%

AutoPCF: Efficient Product Carbon Footprint Accounting with Large Language Models

Zhu Deng, Jinjie Liu, Biao Luo, Can Yuan, Qingrun Yang, Lei Xiao, Wenwen Zhou, Zhu Liu

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

2307.13221 2023-07-26 cs.CV cs.AI cs.CE cs.DC econ.GN q-fin.EC 88%

Multilevel Large Language Models for Everyone

Yuanhao Gong

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

2305.02626 2023-05-05 cs.SE cs.AI cs.HC 88%

"Oops, Did I Just Say That?" Testing and Repairing Unethical Suggestions of Large Language Models with Suggest-Critique-Reflect Process

Pingchuan Ma, Zongjie Li, Ao Sun, Shuai Wang

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

2304.12512 2023-04-26 cs.AI 88%

Semantic Compression With Large Language Models

Henry Gilbert, Michael Sandborn, Douglas C. Schmidt, Jesse Spencer-Smith, Jules White

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

2210.15718 2022-10-31 cs.CL cs.IR 88%

QUILL: Query Intent with Large Language Models using Retrieval Augmentation and Multi-stage Distillation

Krishna Srinivasan, Karthik Raman, Anupam Samanta, Lingrui Liao, Luca Bertelli, Mike Bendersky

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

Comments EMNLP 2022 Industry Track

2206.07682 2022-10-27 cs.CL 88%

Emergent Abilities of Large Language Models

Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, William Fedus

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

Comments Transactions on Machine Learning Research (TMLR), 2022

2109.15082 2021-10-01 cs.CL 88%

Towards Efficient Post-training Quantization of Pre-trained Language Models

Haoli Bai, Lu Hou, Lifeng Shang, Xin Jiang, Irwin King, Michael R. Lyu

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

2511.17178 2026-02-19 cs.RO 88%

Efficient Robot Design with Multi-Objective Black-Box Optimization and Large Language Models

基于多目标黑盒优化与大语言模型的高效机器人设计

Kento Kawaharazuka, Yoshiki Obinata, Naoaki Kanazawa, Haoyu Jia, Kei Okada

机构 * The University of Tokyo（东京大学）

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

AI总结本文提出利用大语言模型提升黑盒优化在机器人本体设计中的效率，通过结合LLMs与传统黑盒优化方法，实现更高效的解决方案探索。

Comments Accepted to IEEE Access, website: https://haraduka.github.io/urdf-llm-opt/ , video: https://www.youtube.com/watch?v=N9iMjx7of1w

2608.18733 2026-08-20 cs.SE cs.AI cs.LG 新提交 88%

Flama: a Python framework for development and deployment of production-ready APIs, machine learning, and LLM services

Flama：用于开发和部署生产级API、机器学习及大语言模型服务的Python框架

José A. Perdiguero López, Miguel A. Durán-Olivencia

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

AI总结 Flama是一款开源Python框架，基于ASGI构建，统一REST API开发、ML服务与LLM推理，含七大子系统及多项内置功能，可支持多后端LLM部署等场景。

Comments 83 pages, 6 figures, 1 table. Software available at https://github.com/vortico/flama, up-to-date documentation at https://flama.dev

2608.18149 2026-08-20 cs.AR cs.AI cs.DC cs.LG cs.PF 新提交 88%

TokenPowerSandbox: Evidence-Gated CPU-First Screening for Energy-Aware LLM Serving

TokenPowerSandbox：面向能源感知大语言模型服务的证据门控CPU优先筛选

Chenxu Niu

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

AI总结 TokenPowerSandbox是面向能源感知LLM服务的证据门控CPU优先筛选工作流，结合CPU投影器、GPU探测等技术，在H100上评估Qwen2.5-7B-Instruct，验证了能源与延迟预测的关联及弃权门控的必要性。

2608.14684 2026-08-18 cs.LG cs.AI 新提交 88%

Mitigating Rubric Interference in LLM Judges via On-Policy Self-Distillation

通过策略内自蒸馏缓解大语言模型评审员中的评分标准干扰

Dingyao Yu, Tong Zhang, Yutao Mou, Yunxiao Zhang, Wei Ye, Shikun Zhang

机构 * Peking University（北京大学）； Weixin Al, Tencent Inc（腾讯公司微信智能）

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

AI总结本研究针对LLM评审员多评分标准评估时的干扰问题，提出SARA方法，通过策略内自蒸馏提升评估一致性，且该一致性可跨数据集迁移。

2608.10459 2026-08-12 cs.CL cs.AI 新提交 88%

MD-ProTector: Positioning Multiple Data-Driven Prototypes for LLM-Generated Text Detection

MD-ProTector：为大语言模型生成文本检测定位多个数据驱动原型

Jinmo Han, Jimin Hong, Chanyeong Moon, Ju Yeon Kang, Seonuk Kim, Nam Soo Kim

机构 * Seoul National University（首尔大学）

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

AI总结该研究针对LLM生成文本检测的类别内部差异问题，提出MD-ProTector模型，通过原型定位损失优化，在多基准测试中取得优于同类编码器方法的检测性能。

2608.08878 2026-08-11 cs.LG cs.AI cs.PF 新提交 88%

DistillCache: KL-Guided Adaptive KV-Cache Eviction for Memory-Efficient LLM Inference

DistillCache：基于KL引导的自适应KV缓存驱逐的内存高效大语言模型推理

Asaad Althoubi

机构 * Oklahoma State University（俄克拉荷马州立大学）

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

AI总结 DistillCache是一种强化学习框架，将KV缓存驱逐建模为序列决策问题，在25%缓存预算下，在LongBench上保留全缓存94.2%的准确率，优于多种基线方法，可提升推理吞吐量。

Comments 20 pages, 5 figures

2608.02515 2026-08-10 cs.CL cs.LG 版本更新 88%

LiveMem: Maintaining Memory State Continuity in Long-Running LLM Inference

LiveMem：在长期运行的大语言模型推理中维持记忆状态连续性

Zhichen Liu, Ruihan Sun, Hengjie Yang, Zipeng Wu, Zhaohan Chen, Xiaofan Zhang, Yang Xu

机构 * Southern University of Science and Technology（南方科技大学）； Xidian University（西安电子科技大学）

专题命中效率与部署：LLM(title,summary_cn)；post-training(abstract)；分类 cs.CL、cs.LG

AI总结该研究针对长期运行LLM推理中上下文切换导致状态不连续的问题，提出LiveMem方法，通过引入独立于活跃上下文的持久记忆状态，实现状态连续性，在LongMemEval等测试中表现领先。

2608.04714 2026-08-06 cs.SE cs.AI cs.LG 新提交 88%

What We Observe as LLM Behavior Can Be a Side-effect of Inference Backend

我们所观察到的大语言模型（LLM）行为可能是推理后端的副作用

Shahed Masoudian, Passant Shafaei, Monorama Swain, Markus Schedl

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

AI总结该研究发现推理后端是影响LLM基准测试分数的不可忽视因素，约39%的分数变异性源于后端，建议披露后端信息及完整生成配置以提升结果可靠性。

2608.03494 2026-08-05 cs.CL cs.LG 新提交 88%

Beyond Initialization Loss: A Systematic Study of Token Embedding Initialization Strategies for LLM Vocabulary Extension

超越初始化损失：大语言模型词汇扩展的词元嵌入初始化策略的系统研究

Raviraj Joshi, Utkarsh Vaidya, Sanjay Singh Chauhan, Niranjan Wartikar

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

AI总结该研究针对LLM词汇扩展，对比20余种初始化策略，发现子词组合方法更优，最优配置使CPT步骤减超6倍，轻量CPT可可靠选择最优初始化策略。

2608.02947 2026-08-05 cs.LG cs.CL 新提交 88%

ATFlash: Per-RoPE-Wavelength Attention Windows for Compute/Memory-Efficient LLM Inference

ATFlash：用于计算/内存高效LLM推理的逐RoPE波长注意力窗口

Shun-ichiro Hayashi, Daichi Mukunoki, Tetsuya Hoshino, Takahiro Katagiri

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

AI总结 ATFlash提出逐RoPE波长注意力窗口，修剪查询-键内积项，在保持长上下文任务性能的同时，可移植至FlashAttention-4等框架，在多款模型上实现LLM推理的计算与内存效率提升。

2607.28959 2026-08-03 cs.LG cs.AI 新提交 88%

Efficient LLM Adversarial Training via Low-Rank Defense and Circuit-Guided Surrogates

基于低秩防御与电路引导代理的高效大语言模型对抗训练

Weiyi He, Yuping Lin, Jiliang Tang, Yue Xing

机构 * Michigan State University（密歇根州立大学）

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

AI总结本研究针对大语言模型对抗训练计算成本高的问题，从防御侧优化表示微调、攻击侧构建轻量代理模型两方面提出策略，使每步对抗训练FLOPs降48.1%，仅需0.0118%可训练参数。

2406.05881 2026-07-24 cs.LG cs.CL cs.RO 88%

Training Fast Robot Policies with Slow Foundation Models

用慢速基础模型训练快速机器人策略

Utsav Singh, Pramit Bhattacharyya, Vinay P. Namboodiri, Amrit Singh Bedi

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

AI总结本文提出VGRS方法，通过利用慢速基础模型训练快速机器人策略，结合LLM生成奖励与视觉分析诊断，提升机器人在复杂任务中的表现和部署效率。

2512.08967 2026-07-17 cs.LG cs.AI 版本更新 88%

CluCERT: Certifying LLM Robustness via Clustering-Guided Denoising Smoothing

CluCERT：通过聚类引导去噪平滑认证大语言模型的鲁棒性

Zixia Wang, Gaojie Jin, Jia Hu, Ronghui Mu

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

AI总结针对大语言模型易受对抗攻击问题，提出CluCERT框架，通过聚类引导去噪平滑认证其鲁棒性。引入语义聚类过滤器，结合细化模块和快速同义词替换策略，提升效率，实验证明该方法在鲁棒性边界和计算效率上优于现有方法。

2606.29581 2026-07-16 cs.LG cs.AI 版本更新 88%

The Joint Effect of Quantization and Sampling Temperature on LLM Safety Alignment: A Factorial Analysis

量化和采样温度对LLM安全对齐的联合效应：一项因子分析

Hari Prasad, Ritam Pal

机构 * Conscious Engines（意识引擎）

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

AI总结通过因子分析评估9个模型在3种精度和6种温度下的安全对齐，发现量化通常安全中性，而高温增加决策不稳定性，且两者无系统性叠加效应。

Comments 11 pages, 5 Figures

1. 效率与部署 22280 篇

Future-proofing geotechnics workflows: accelerating problem-solving with large language models

Opportunities and Challenges of Applying Large Language Models in Building Energy Efficiency and Decarbonization Studies: An Exploratory Overview

H$_2$O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models

CoLLiE: Collaborative Training of Large Language Models in an Efficient Way

Teach me with a Whisper: Enhancing Large Language Models for Analyzing Spoken Transcripts using Speech Embeddings

QUIK: Towards End-to-End 4-Bit Inference on Generative Large Language Models

BitNet: Scaling 1-bit Transformers for Large Language Models

How Do Large Language Models Capture the Ever-changing World Knowledge? A Review of Recent Advances

Leveraging Large Language Models for DRL-Based Anti-Jamming Strategies in Zero Touch Networks

AutoPCF: Efficient Product Carbon Footprint Accounting with Large Language Models

Multilevel Large Language Models for Everyone

"Oops, Did I Just Say That?" Testing and Repairing Unethical Suggestions of Large Language Models with Suggest-Critique-Reflect Process

Semantic Compression With Large Language Models

QUILL: Query Intent with Large Language Models using Retrieval Augmentation and Multi-stage Distillation

Emergent Abilities of Large Language Models

Towards Efficient Post-training Quantization of Pre-trained Language Models

Efficient Robot Design with Multi-Objective Black-Box Optimization and Large Language Models

Flama: a Python framework for development and deployment of production-ready APIs, machine learning, and LLM services

TokenPowerSandbox: Evidence-Gated CPU-First Screening for Energy-Aware LLM Serving

Mitigating Rubric Interference in LLM Judges via On-Policy Self-Distillation

MD-ProTector: Positioning Multiple Data-Driven Prototypes for LLM-Generated Text Detection

DistillCache: KL-Guided Adaptive KV-Cache Eviction for Memory-Efficient LLM Inference

LiveMem: Maintaining Memory State Continuity in Long-Running LLM Inference

What We Observe as LLM Behavior Can Be a Side-effect of Inference Backend

Beyond Initialization Loss: A Systematic Study of Token Embedding Initialization Strategies for LLM Vocabulary Extension

ATFlash: Per-RoPE-Wavelength Attention Windows for Compute/Memory-Efficient LLM Inference

Efficient LLM Adversarial Training via Low-Rank Defense and Circuit-Guided Surrogates

Training Fast Robot Policies with Slow Foundation Models

CluCERT: Certifying LLM Robustness via Clustering-Guided Denoising Smoothing

The Joint Effect of Quantization and Sampling Temperature on LLM Safety Alignment: A Factorial Analysis