语言大模型 / LLM - arXivDaily 专题

2404.01430 2024-04-03 cs.CL cs.AI cs.LG 80%

Position-Aware Parameter Efficient Fine-Tuning Approach for Reducing Positional Bias in LLMs

Zheng Zhang, Fan Yang, Ziyan Jiang, Zheng Chen, Zhengyang Zhao, Chengyuan Ma, Liang Zhao, Yang Liu

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2403.12017 2024-03-19 cs.LG cs.AI cs.CL 80%

Supervised Fine-Tuning as Inverse Reinforcement Learning

Hao Sun

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2401.14151 2024-03-12 cs.LG cs.AI cs.CL 80%

True Knowledge Comes from Practice: Aligning LLMs with Embodied Environments via Reinforcement Learning

Weihao Tan, Wentao Zhang, Shanqi Liu, Longtao Zheng, Xinrun Wang, Bo An

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted by ICLR2024

2403.04746 2024-03-08 cs.CL cs.AI cs.LG 80%

LLMs in the Imaginarium: Tool Learning through Simulated Trial and Error

Boshi Wang, Hao Fang, Jason Eisner, Benjamin Van Durme, Yu Su

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Code and data available at https://github.com/microsoft/simulated-trial-and-error

2402.17930 2024-02-29 cs.AI cs.CL cs.LG 80%

Pragmatic Instruction Following and Goal Assistance via Cooperative Language-Guided Inverse Planning

Tan Zhi-Xuan, Lance Ying, Vikash Mansinghka, Joshua B. Tenenbaum

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Accepted to AAMAS 2024. 8 pages (excl. references), 5 figures/tables. (Appendix: 8 pages, 8 figures/tables). Code available at: https://github.com/probcomp/CLIPS.jl

2402.09997 2024-02-20 cs.AI cs.CL cs.LG 80%

LoraRetriever: Input-Aware LoRA Retrieval and Composition for Mixed Tasks in the Wild

Ziyu Zhao, Leilei Gan, Guoyin Wang, Wangchunshu Zhou, Hongxia Yang, Kun Kuang, Fei Wu

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2402.10210 2024-02-16 cs.LG cs.AI cs.CL cs.CV stat.ML 80%

Self-Play Fine-Tuning of Diffusion Models for Text-to-Image Generation

Huizhuo Yuan, Zixiang Chen, Kaixuan Ji, Quanquan Gu

专题命中指令微调：large language model(abstract)；language model(abstract)；RLHF(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments 28 pages, 8 figures, 10 tables

2311.06668 2024-02-15 cs.LG cs.AI cs.CL 80%

In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering

Sheng Liu, Haotian Ye, Lei Xing, James Zou

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2402.01684 2024-02-06 cs.CL cs.AI cs.LG 80%

A Framework to Implement 1+N Multi-task Fine-tuning Pattern in LLMs Using the CGC-LORA Algorithm

Chao Song, Zhihao Ye, Qiqiang Lin, Qiuying Peng, Jun Wang

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2310.00647 2024-01-23 cs.CV cs.MM 80%

Beyond Task Performance: Evaluating and Reducing the Flaws of Large Multimodal Models with In-Context Learning

Mustafa Shukor, Alexandre Rame, Corentin Dancette, Matthieu Cord

专题命中指令微调：large language model(abstract)；language model(abstract)；instruction tuning(abstract)；RLHF(abstract)

Comments ICLR 2024. Project Page: https://evalign-icl.github.io/

2311.06791 2023-12-07 cs.CV 80%

InfMLLM: A Unified Framework for Visual-Language Tasks

Qiang Zhou, Zhibin Wang, Wei Chu, Yinghui Xu, Hao Li, Yuan Qi

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；pretraining(abstract)

Comments 8

2310.01320 2023-10-25 cs.AI cs.CL cs.CY cs.LG cs.MA 80%

Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation

Shenzhi Wang, Chang Liu, Zilong Zheng, Siyuan Qi, Shuo Chen, Qisen Yang, Andrew Zhao, Chaofei Wang, Shiji Song, Gao Huang

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments 40 pages

2310.12100 2023-10-19 cs.CL cs.AI cs.CV cs.LG cs.MM 80%

Non-Intrusive Adaptation: Input-Centric Parameter-efficient Fine-Tuning for Versatile Multimodal Modeling

Yaqing Wang, Jialin Wu, Tanmaya Dabral, Jiageng Zhang, Geoff Brown, Chun-Ta Lu, Frederick Liu, Yi Liang, Bo Pang, Michael Bendersky, Radu Soricut

专题命中指令微调：large language model(abstract)；language model(abstract)；instruction tuning(abstract)；分类 cs.CL、cs.AI、cs.LG

2308.13032 2023-09-12 cs.CL cs.AI cs.CE cs.IR cs.LG 80%

Financial News Analytics Using Fine-Tuned Llama 2 GPT Model

Bohdan M. Pavlyshenko

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

2304.15010 2023-05-01 cs.CV cs.AI cs.CL cs.LG cs.MM 80%

LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model

Peng Gao, Jiaming Han, Renrui Zhang, Ziyi Lin, Shijie Geng, Aojun Zhou, Wei Zhang, Pan Lu, Conghui He, Xiangyu Yue, Hongsheng Li, Yu Qiao

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments Code and models are available at https://github.com/ZrrSkywalker/LLaMA-Adapter

2304.08968 2023-04-19 cs.CL cs.AI cs.CR cs.LG 80%

Stochastic Parrots Looking for Stochastic Parrots: LLMs are Easy to Fine-Tune and Hard to Detect with other LLMs

Da Silva Gameiro Henrique, Andrei Kucharavy, Rachid Guerraoui

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI、cs.LG

Comments 15 pages, 6 figures; 10 pages, 7 figures Supplementary Materials; under review at ECML 2023

2212.10929 2022-12-22 cs.CL cs.AI cs.LG 80%

SPT: Semi-Parametric Prompt Tuning for Multitask Prompted Learning

M Saiful Bari, Aston Zhang, Shuai Zheng, Xingjian Shi, Yi Zhu, Shafiq Joty, Mu Li

专题命中指令微调：large language model(abstract)；language model(abstract)；pretraining(abstract)；分类 cs.CL、cs.AI、cs.LG

2411.00056 2024-11-04 cs.CL cs.AI 80%

Generating Diverse Negations from Affirmative Sentences

Darian Rodriguez Vasquez, Afroditi Papadaki

专题命中指令微调：LLM(abstract)；large language model(abstract)；language model(abstract)；分类 cs.CL、cs.AI

Comments Accepted at "Adaptive Foundation Models: Evolving AI for Personalized and Efficient Learning" workshop at NeurIPS 2024

2608.21762 2026-08-25 cs.CV cs.CL 新提交 79%

Learning to Look Again: Loss-Gap Supervision for Free-form Crop Routing in Vision-Language Models

重新学习观察：面向视觉语言模型的自由格式裁剪重路由的损失间隙监督

Jinchang Zhu, Rong Fu, Yi Ding, Chenghao Wu, Ying Liu, Menglin Yang

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

AI总结提出GapSight框架，利用损失间隙监督让VLMs选择性重读图像局部区域，在多个基准测试中显著提升了LLaVA、InternVL2.5等VLMs的细节类任务性能。

2606.01734 2026-08-25 cs.CV cs.LG cs.RO 版本更新 79%

FlatVPR: Plug-and-play Geo-linear Residual Adapter for Geometric Rectification of Foundation Model Feature Manifolds

FlatVPR: 用于基础模型特征流形几何校正的即插即用地线性残差适配器

Rai Hisada, Kanji Tanaka

机构 * Fundamental Engineering for Knowledge-Based Society, Graduate School of Engineering, University of Fukui（知识社会基础工程，工程研究生院，福井大学）

专题命中指令微调：foundation model(title,abstract)；分类 cs.LG

AI总结提出FlatVPR范式，通过可学习残差适配器和Pullback Flatness Loss抑制特征流形曲率，实现稀疏锚点下的线性插值重建，在NCLT数据集上显著提升视觉位置识别精度。

Comments 16 pages, 4 figures, technical report; v2: Added supplementary materials for downstream tasks and included a supplementary roadmap in Section I

2601.23182 2026-08-25 cs.CL 版本更新 79%

FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation

FourierSampler: 通过频率引导生成解锁扩散语言模型的非自回归潜力

Siyang He, Qiqi Wang, Xiaoran Liu, Hongnan Ma, Yiwei Shi, Yuerong Song, Ying Zhu, Tianyi Liang, Zengfeng Huang, Ziwei He, Xipeng Qiu

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

AI总结 FourierSampler通过频域滑动窗口机制，在扩散语言模型中实现非自回归生成，提升生成效果并超越自回归模型

Comments 15 pages, 6 figures, under review

2608.19778 2026-08-21 cs.MA cs.AI 新提交 79%

Distilling Aggregate Mobility Statistics into a Language Model Policy for Post-Event Crowd Simulation

将聚合出行统计数据提炼为语言模型策略以用于事件后人群模拟

Tatsuya Amano, Hirozumi Yamaguchi

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

AI总结该研究针对仅能获取聚合出行统计数据的场景，通过微调语言模型人群智能体并结合迭代比例拟合与低秩适配器，降低了事件后人群模拟的目的地份额误差。

Comments 4 pages, 3 figures. Accepted as a short paper at ACM SIGSPATIAL 2026

2608.18597 2026-08-20 cs.LG 新提交 79%

Off-Manifold Collapse in Guided Protein Language Models

引导式蛋白质语言模型中的流形外坍塌

Shuibai Zhang, Xinchi Liu, Fred Zhangzhi Peng, Zhihan Yang, Shutong Wu, Yingzi Ma, Jiawei Zhang

机构 * Duke University（杜克大学）； Cornell University（康奈尔大学）； University of Wisconsin--Madison（威斯康星大学麦迪逊分校）

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

AI总结该研究发现引导式蛋白质语言模型存在流形外坍塌问题，提出无需训练的马氏过滤方法，可低成本提升生成序列的属性得分与结构合理性，且可跨引导方法迁移。

Comments 12 pages

2608.17306 2026-08-19 cs.CV cs.AI 新提交 79%

Learning What Not to Learn: Adversarial Disentangled Prompt Tuning for Robust Vision-Language Models

学习无需学习的内容：用于鲁棒视觉-语言模型的对抗性解耦提示调优

Yang Chen, Zhan Zhuang, Yanbin Wei, Zebin Chen, Hua Liu, Yu Zhang

机构 * Southern University of Science and Technology（南方科技大学）； City University of Hong Kong（香港城市大学）； Hong Kong University of Science and Technology（香港科技大学）

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

AI总结针对现有对抗性提示调优存在的鲁棒泛化过拟合问题，提出ADAPT框架，通过双提示机制解耦鲁棒与伪鲁棒特征，提升模型对未见类别对抗样本的鲁棒性。

2605.13163 2026-08-18 cs.CR cs.CV cs.LG 79%

LoREnc: Low-Rank Encryption for Securing Foundation Models and LoRA Adapters

LoREnc：低秩加密用于保护基础模型和LoRA适配器

Beomjin Ahn, Jungmin Kwon, Chanyong Jung, Jaewook Chung

机构 * Samsung Research（三星研究院）； Samsung Electronics（三星电子）； Amazon Web Services（亚马逊网络服务）； University of Michigan（密歇根大学）

专题命中指令微调：foundation model(title,abstract)；分类 cs.LG

AI总结 LoREnc通过谱截断和补偿技术，在不重新训练的情况下保护基础模型和LoRA适配器，防止模型恢复攻击和知识产权泄露，实验表明其在1%计算开销下有效。

Comments Accepted to ICIP 2026

Journal ref 2026 IEEE International Conference on Image Processing (ICIP)

2603.12478 2026-08-18 cs.CV cs.LG 版本更新 79%

Less Data, Faster Convergence: Goal-Driven Data Optimization for Multimodal Instruction Tuning

数据更少，收敛更快：面向多模态指令微调的目标驱动数据优化

Rujie Wu, Haozhe Zhao, Hai Ci, Yizhou Wang

机构 * Peking University（北京大学）； University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）； National University of Singapore（新加坡国立大学）

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

AI总结本文提出目标驱动数据优化框架GDO，通过优化训练样本实现更快收敛和更高精度，适用于多模态指令微调任务。

Comments Accepted to ECCV 2026

2608.11359 2026-08-13 cs.LG 新提交 79%

Market-Information-Aware Gated-LoRA of Foundation Models for Transferable Day-Ahead Electricity Price Forecasting

面向可迁移日前电价预测的、融合市场信息的基础门控低秩适配（Gated-LoRA）模型

Hang Fan, Wei Wei, Shengwei Mei

专题命中指令微调：foundation model(title,abstract)；分类 cs.LG

AI总结本文提出融合市场信息的Gated-LoRA框架，将Chronos-2模型迁移至日前电价预测，经中国四省现货市场验证，可显著降低预测误差，为数据稀缺电力市场提供迁移方案。

2608.10525 2026-08-12 cs.CV cs.AI 新提交 79%

Dynamic Context Adapters: Efficiently Infusing History into Vision-and-Language Models

动态上下文适配器：将历史信息高效注入视觉-语言模型

Yuhang Song, Bor-Jiun Lin, Jiaxu Liu, Te-Chuan Chiu, Anh Nguyen, Chun-Yi Lee

机构 * University of Liverpool（利物浦大学）； National Tsinghua University（国立清华大学）； Imperial College London（伦敦帝国学院）； National Taiwan University（国立台湾大学）

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

AI总结针对视觉-语言模型整合历史上下文时的计算与信息损失问题，提出动态上下文适配器（DCA），实现高效历史注入，降低注意力FLOPs超25%、内存13%，提升长时序任务性能。

2608.09521 2026-08-11 cs.AI 新提交 79%

One Adapter Pair per Model: A Universal Activation Interface for Language Models

每个模型一个适配器对：语言模型的通用激活接口

Su-Hyeon Kim, Jiwan Mun, Yo-Sub Han

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

AI总结本研究提出通用激活总线框架，为兼容语言模型提供通用激活接口，每个模型配备轻量级适配器对，可实现激活工具跨模型共享与复用，为可复用工具建立稳定的激活契约。

2608.09124 2026-08-11 cs.AI 新提交 79%

ChronoState: Hidden Elapsed-Time Conditioning for Temporal-State Action Selection in Frozen-Backbone Language Models

ChronoState：用于冻结骨干语言模型中时间-状态动作选择的隐式流逝时间条件

Sam Siavoshian, Omar Ramadan, Amir K. Saeed, Benjamin A. Johnson, Amin Mohamed El-Amin Diab, Benjamin M. Rodriguez

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

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

AI总结该研究提出 ChronoState 基准，证实冻结骨干语言模型可在直接监督下将隐式流逝时间与符号状态组合，但其泛化性有限，且性能不及提示注入时间戳基线。

Comments Submitted to SPIE Defense and Commercial Sensing 2027

1. 指令微调 11653 篇

Position-Aware Parameter Efficient Fine-Tuning Approach for Reducing Positional Bias in LLMs

Supervised Fine-Tuning as Inverse Reinforcement Learning

True Knowledge Comes from Practice: Aligning LLMs with Embodied Environments via Reinforcement Learning

LLMs in the Imaginarium: Tool Learning through Simulated Trial and Error

Pragmatic Instruction Following and Goal Assistance via Cooperative Language-Guided Inverse Planning

LoraRetriever: Input-Aware LoRA Retrieval and Composition for Mixed Tasks in the Wild

Self-Play Fine-Tuning of Diffusion Models for Text-to-Image Generation

In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering

A Framework to Implement 1+N Multi-task Fine-tuning Pattern in LLMs Using the CGC-LORA Algorithm

Beyond Task Performance: Evaluating and Reducing the Flaws of Large Multimodal Models with In-Context Learning

InfMLLM: A Unified Framework for Visual-Language Tasks

Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation

Non-Intrusive Adaptation: Input-Centric Parameter-efficient Fine-Tuning for Versatile Multimodal Modeling

Financial News Analytics Using Fine-Tuned Llama 2 GPT Model

LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model

Stochastic Parrots Looking for Stochastic Parrots: LLMs are Easy to Fine-Tune and Hard to Detect with other LLMs

SPT: Semi-Parametric Prompt Tuning for Multitask Prompted Learning

Generating Diverse Negations from Affirmative Sentences

Learning to Look Again: Loss-Gap Supervision for Free-form Crop Routing in Vision-Language Models

FlatVPR: Plug-and-play Geo-linear Residual Adapter for Geometric Rectification of Foundation Model Feature Manifolds

FourierSampler: Unlocking Non-Autoregressive Potential in Diffusion Language Models via Frequency-Guided Generation

Distilling Aggregate Mobility Statistics into a Language Model Policy for Post-Event Crowd Simulation

Off-Manifold Collapse in Guided Protein Language Models

Learning What Not to Learn: Adversarial Disentangled Prompt Tuning for Robust Vision-Language Models

LoREnc: Low-Rank Encryption for Securing Foundation Models and LoRA Adapters

Less Data, Faster Convergence: Goal-Driven Data Optimization for Multimodal Instruction Tuning

Market-Information-Aware Gated-LoRA of Foundation Models for Transferable Day-Ahead Electricity Price Forecasting

Dynamic Context Adapters: Efficiently Infusing History into Vision-and-Language Models

One Adapter Pair per Model: A Universal Activation Interface for Language Models

ChronoState: Hidden Elapsed-Time Conditioning for Temporal-State Action Selection in Frozen-Backbone Language Models