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。

今日/当前日期收录 13 信号源:cs.CL, cs.AI, cs.LG
2606.19348 2026-06-19 cs.CL cs.AI 新提交 专题 95

DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence

DeepSeek-V4: 迈向高效百万令牌上下文智能

DeepSeek-AI, Anyi Xu, Bangcai Lin, Bing Xue, Bingxuan Wang, Bingzheng Xu, Bochao Wu, Bowei Zhang, Chaofan Lin, Chen Dong, Chenchen Ling, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chengyu Hou, Chenhao Xu, Chenze Shao, Chong Ruan, Conner Sun, Damai Dai, Daya Guo, Dejian Yang, Deli Chen, Donghao Li, Dongjie Ji, Erhang Li, Fang Wei, Fangyun Lin, Fangzhou Yuan, Feiyu Xia, Fucong Dai, Guangbo Hao, Guanting Chen, Guoai Cao, Guolai Meng, Guowei Li, Han Yu, Han Zhang, Hanwei Xu, Hao Li, Haofen Liang, Haoling Zhang, Haoming Luo, Haoran Wei, Haotian Yuan, Haowei Zhang, Haowen Luo, Haoyu Chen, Haozhe Ji, Hengqing Zhang, Honghui Ding, Hongxuan Tang, Huanqi Cao, Huazuo Gao, Hui Qu, Hui Zeng, J Yang, JQ Zhu, Jia Luo, Jia Song, Jia Yu, Jialiang Huang, Jialu Cai, Jian Liang, Jiangting Zhou, Jiasheng Ye, Jiashi Li, Jiaxin Xu, Jiewen Hu, Jieyu Yang, Jin Chen, Jin Yan, Jingchang Chen, Jingli Zhou, Jingting Xiang, Jingyang Yuan, Jingyuan Cheng, Jingzi Zhou, Jinhua Zhu, Jiping Yu, Joseph Sun, Jun Ran, Junguang Jiang, Junjie Qiu, Junlong Li, Junmin Zheng, Junxiao Song, Kai Dong, Kaige Gao, Kang Guan, Kexing Zhou, Kezhao Huang, Kuai Yu, Lean Wang, Lecong Zhang, Lei Wang, Leyi Xia, Li Zhang, Liang Zhao, Lihua Guo, Lingxiao Luo, Linwang Ma, Linyan Zhu, Litong Wang, Liyu Cai, Liyue Zhang, Longhao Chen, MS Di, MY Xu, Max Mei, Miaojun Wang, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Mingming Li, Mingxu Zhou, Minmin Han, Ning Wang, Panpan Huang, Panpan Wang, Peixin Cong, Peiyi Wang, Peng Zhang, Qiancheng Wang, Qihao Zhu, Qingyang Li, Qinyu Chen, Qiushi Du, Qiwei Jiang, Rui Tian, Ruifan Xu, Ruijie Lu, Ruiling Xu, Ruiqi Ge, Ruisong Zhang, Ruizhe Pan, Runji Wang, Runqian Chen, Runqiu Yin, Runxin Xu, Ruomeng Shen, Ruoyu Zhang, Ruyi Chen, SH Liu, Shanghao Lu, Shangmian Sun, Shangyan Zhou, Shanhuang Chen, Shaofei Cai, Shaoheng Nie, Shaoqing Wu, Shaoyuan Chen, Shengding Hu, Shengyu Liu, Shiqiang Hu, Shirong Ma, Shiyu Wang, Shuiping Yu, Shunfeng Zhou, Shuting Pan, Shuying Yu, Songyang Zhou, Tao Ni, Tao Yun, Tian Jin, Tian Pei, Tian Ye, Tianle Lin, Tianran Ji, Tianyi Cui, Tianyuan Yue, Tingting Yu, Tun Wang, W Zhang, WL Xiao, Wangding Zeng, Wei An, Weilin Zhao, Wen Liu, Wenfeng Liang, Wenjie Pang, Wenjing Luo, Wenjing Yao, Wenjun Gao, Wenkai Yang, Wenlve Huang, Wenqing Hou, Wentao Zhang, Wenting Ma, Xi Gao, Xiang He, Xiangwen Wang, Xianzu Wang, Xiao Bi, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaokang Zhang, Xiaotao Nie, Xiaowen Sun, Xiaoxiang Wang, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xingchen Liu, Xingkai Yu, Xingyou Li, Xinyu Yang, Xinyu Zhang, Xu Chen, Xuanyu Wang, Xuecheng Su, Xueyin Chen, Xuheng Lin, Xuwei Fu, YC Yan, YQ Wang, YW Ma, Yanfeng Luo, Yang Zhang, Yanhong Xu, Yanru Ma, Yanwen Huang, Yao Li, Yao Li, Yao Xu, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yi Qian, Yi Shao, Yi Yu, Yichao Zhang, Yifan Ding, Yifan Shi, Yijia Wu, Yiliang Xiong, Yiling Ma, Ying He, Ying Tang, Ying Zhou, Yingjia Luo, Yinmin Zhong, Yishi Piao, Yisong Wang, Yixiang Zhang, Yixiao Chen, Yixuan Tan, Yixuan Wei, Yiyang Ma, Yiyuan Liu, Yonglun Yang, Yongqiang Guo, Yongtong Wu, Yu Wu, YuKun Li, Yuan Cheng, Yuan Ou, Yuanfan Xu, Yuanhao Li, Yuduan Wang, Yuehan Yang, Yuer Xu, Yuhan Wu, Yuhao Meng, Yuheng Zou, Yukun Zha, Yunfan Xiong, Yupeng Chen, Yuping Lin, Yuqian Cao, Yuqian Wang, Yushun Zhang, Yuting Yan, Yutong Lin, Yuxian Gu, Yuxiang Luo, Yuxiang You, Yuxuan Liu, Yuxuan Zhou, Yuyang Zhou, Yuzhen Huang, ZF Wu, Zehao Wang, Zehua Zhao, Zehui Ren, Zekai Zhang, Zhangli Sha, Zhe Fu, Zhe Ju, Zhean Xu, Zhenda Xie, Zhengyan Zhang, Zheren Gao, Zhewen Hao, Zhibin Gou, Zhicheng Ma, Zhigang Yan, Zhihong Shao, Zhixian Huang, Zhixuan Chen, Zhiyu Wu, Zhizhou Ren, Zhongyu Wu, Zhuoshu Li, Zhuping Zhang, Zian Xu, Zihao Wang, Zihua Qu, Zihui Gu, Zijia Zhu, Zilin Li, Zipeng Zhang, Ziwei Xie, Ziyi Gao, Ziyi Wan, Zizheng Pan, Zongqing Yao

专题命中 预训练 :百万token上下文MoE模型,架构优化

AI总结 提出DeepSeek-V4系列MoE模型,通过混合注意力架构、流形约束超连接和Muon优化器,实现百万令牌上下文的高效推理,在核心任务上超越前代。

2606.20381 2026-06-19 cs.AI 新提交 专题 90

Rethinking Shrinkage Bias in LLM FP4 Pretraining: Geometric Origin, Systemic Impact, and UFP4 Recipe

重新思考LLM FP4预训练中的收缩偏差:几何起源、系统影响与UFP4方案

Qian Zhao, Kunlong Chen, Changxin Tian, Zhonghui Jiang, Haitao Zhang, Chaofan Yu, Peijie Jiang, Mingliang Gong, Jia Liu, Ziqi Liu, Zhiqiang Zhang, Jun Zhou

专题命中 预训练 :研究LLM FP4预训练中的收缩偏差与优化方案。

AI总结 本文发现E2M1格式因几何不对称导致收缩偏差,该偏差经随机哈达玛变换放大,造成训练不稳定;提出均匀网格E1M2/INT4及UFP4训练方案,在多种模型上实现更低损失。

Comments 18 pages, 12 figures

2606.20089 2026-06-19 cs.CL cs.AI 新提交 专题 90

IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources

IHUBERT: 面向波斯语资源的基于向量的语义去重与领域平衡预训练

Arash Ghafouri, Mahdi Firouzmandi, Hossein Saberi, Mohammad Reza Hasani Ahangar

专题命中 预训练 :波斯语预训练语言模型

AI总结 提出IHUBERT,一个基于RoBERTa-base的波斯语预训练模型,通过多阶段预处理(包括基于向量数据库的语义去重和领域平衡)在45GB语料上训练,在多项NLU任务上取得领先结果,尤其抽取式问答表现突出。

2606.19993 2026-06-19 cs.LG 新提交 专题 85

Activation- and Influence-Aware Ranks (AIR): Function-Preserving SVD Compression for LLMs

激活与影响感知秩 (AIR):保持功能的SVD压缩用于大语言模型

Nico Harder, Daniel Becking, Karsten Mueller, Wojciech Samek

专题命中 预训练 :提出LLM压缩框架,提升模型效率

AI总结 提出AIR框架,基于SVD和反向信号影响度量,通过单次交替最小二乘扫描实现权重矩阵的低秩近似,在参数保留≤60%时困惑度比SVD-LLM(W)改善>18%,并减少90%校准数据。

Comments Accepted at the ICML 2026 Workshop on Resource-Adaptive Foundation Model Inference (AdaptFM), Seoul, South Korea (non-archival)

2606.19491 2026-06-19 cs.LG stat.ML 新提交 专题 85

Algebraic Dead Directions in LayerNorm Transformers: A Forward-Pass-Only Diagnostic at LLM Scale

LayerNorm Transformer 中的代数死方向:一种仅需前向传播的大语言模型规模诊断方法

Tejas Pradeep Shirodkar, P. J. Narayanan

专题命中 预训练 :研究LayerNorm变换器的死方向,涉及预训练模型诊断。

AI总结 本文发现 LayerNorm 的逆尺度方向是后最终归一化中心激活协方差矩阵的精确代数核,可仅从参数中读取死方向,无需前向或后向传播,并在 14 个预训练模型上验证了其有效性。

Comments 34 pages, 7 figures, 6 tables. Empirical companion to arXiv:2606.05957

2606.19468 2026-06-19 cs.CL 新提交 专题 85

Characterizing Narrative Content in Web-scale LLM Pretraining Data

网络规模LLM预训练数据中的叙事内容特征化

Teagan Johnson, Elliott Ash, Andrew Piper, Maria Antoniak

专题命中 预训练 :细粒度研究LLM预训练语料库的叙事特征。

AI总结 首次细粒度研究LLM预训练语料库Dolma的叙事特征,提出涵盖三个核心叙事元素(能动性、场景、事件)的框架,构建NarraBERT模型并发布NarraDolma数据集,揭示叙事结构在异构数据中可测量且分布不均。

Comments 8 pages of main content, 28 total pages. 30 figures

2510.06048 2026-06-19 cs.LG 版本更新 专题 85

BLISS: A Lightweight Bilevel Influence Scoring Method for Data Selection in Language Model Pretraining

BLISS: 一种用于语言模型预训练数据选择的轻量级双层影响评分方法

Jie Hao, Rui Yu, Wei Zhang, Huixia Wang, Jie Xu, Mingrui Liu

专题命中 预训练 :提出数据选择方法用于语言模型预训练

AI总结 提出一种无需外部预训练模型的轻量级数据选择方法BLISS,通过双层优化和代理模型估计训练样本的长期影响,实现高效数据筛选,在C4数据集上预训练多种规模模型,显著加速收敛并提升下游任务性能。

2606.19989 2026-06-19 cs.DC cs.LG 新提交 专题 80

Online Dynamic Batching with Formal Guarantees for LLM Training

面向LLM训练的具有形式保证的在线动态批处理

Dian Li, Zekun Wang, Yaoru Wang, Jiahong Yan

专题命中 预训练 :提出在线动态批处理系统加速LLM训练

AI总结 提出在线动态批处理(ODB)系统,在数据加载器侧将批构建延迟到样本真实成本可观测时,解决离线批采样中预处理成本不可见问题,实现1.58-4.43x吞吐量提升,并提供无死锁有界终止的形式化保证。

Comments 29 pages, 3 figures, 21 tables

2606.19528 2026-06-19 cs.LG cs.AI 新提交 专题 80

Techniques for Peak Memory Reduction for LoRA Fine-tuning of LLMs on Edge Devices

边缘设备上LLM LoRA微调峰值内存降低技术

Hassan Dbouk, Matthias Reisser, Prathamesh Mandke, Likhita Arun Navali, Christos Louizos

专题命中 预训练 :降低LLM LoRA微调峰值内存的技术

AI总结 针对边缘设备上LLM LoRA微调的内存瓶颈,提出四种互补技术(量化、检查点、softmax近似、logits掩码),在Llama-3.2 3B和Qwen-2.5 3B上实现高达26倍和28倍的峰值内存降低。

Comments Hassan Dbouk and Matthias Reisser contributed equally to this work

2602.04396 2026-06-19 cs.LG cs.AI 版本更新 专题 80

LoRDO: Distributed Low-Rank Optimization with Infrequent Communication

LoRDO: 分布式低秩优化与低频通信

Andrej Jovanović, Alex Iacob, Mher Safaryan, Ionut-Vlad Modoranu, Lorenzo Sani, William F. Shen, Xinchi Qiu, Dan Alistarh, Nicholas D. Lane

专题命中 预训练 :LoRDO框架实现分布式低秩优化与低频通信

AI总结 提出LoRDO框架,统一低秩优化与低频同步,通过全秩准双曲更新恢复子空间探索,在125M-720M模型规模下实现与低秩DDP近似的性能,通信量减少约10倍。

Comments Accepted at ICML 2026

2606.19625 2026-06-19 cs.CL cs.LG 新提交 专题 75

Where Does Social Reasoning Come From? Capability Provenance in Language Models

社会推理从何而来?语言模型中的能力来源

Glenn Matlin, Chandreyi Chakraborty, Saehee Eom, Mika Okamoto, Rayan Castilla, Louis Jaburi, Alvin Deng, Taywon Min, Lucia Quirke, Stella Biderman, Mark Riedl

专题命中 预训练 :通过训练数据归因分析社会推理与STEM推理来源。

AI总结 通过训练数据归因方法,发现OLMo3-7B中社会推理和STEM推理依赖于不同的预训练语料区域,且推理层面的差异比知识层面更显著。

Comments Under review at COLM 2026 (Conference)

2606.19379 2026-06-19 cs.LG cs.AI cs.CL 新提交 专题 70

How Linear Is a Transformer Feed-Forward Block? Per-Block Linear Recoverability Is Learned, Not Architectural

Transformer 前馈块有多线性?逐块线性可恢复性是学习得到的,而非架构决定的

Stuart Whipp

专题命中 预训练 :分析Transformer前馈块的线性度,与模型架构相关。

AI总结 通过精确最小二乘线性近似,测量训练后 Transformer 各前馈块的线性可恢复性,发现其高度异质且非单调,是学习得到的属性而非架构决定,并可用于压缩和诊断。

Comments 14 pages, 5 figures

2606.19367 2026-06-19 cs.LG 新提交 专题 70

Weibull Weight-Scale Parameter Evolution under AdamW Training Dynamics

Weibull 权重尺度参数在 AdamW 训练动态下的演化

Tiexin Ding

专题命中 预训练 :研究AdamW训练动态,以Pythia模型为例。

AI总结 研究 AdamW 训练中 Weibull 权重尺度参数 λ 增长、过冲和松弛的原因,推导出三种力(对齐、注入、衰减)的分解,并在 Pythia-70M 模型上验证对齐力主导上升阶段,贡献 88-94%。

Comments 21 pages, 14 figures