上下文学习(In-context learning, ICL)已成为大语言模型(Large Language Models, LLMs)少样本学习的有效解决方案。然而,目前学界对大语言模型如何通过ICL利用示例明确任务并学习对应计算函数的机制仍探索不足。借鉴人类从示例中的内容-标签映射开展学习的方式,我们将ICL提示中的token划分为内容token、停用词token和模板token三类。我们的研究目标是识别出表征会直接影响大语言模型性能的token类型,我们将这一属性称为性能关键性。通过在测试样例的注意力机制中消融对应表征,我们发现,与人类关注信息性词汇的模式不同,信息性内容token的表征对性能的影响弱于模板token和停用词token。我们提供证据表明,性能关键token的表征会聚合来自内容token的信息。此外,我们通过实验证明,词汇语义、重复度和结构线索是这类token的核心区分特征。本研究揭示了大语言模型如何从示例中学习执行任务的机制,深化了我们对不同类型token在大语言模型中所发挥作用的认知。
英文摘要:
In-context learning (ICL) has emerged as an effective solution for few-shot learning with large language models (LLMs). However, how LLMs leverage demonstrations to specify a task and learn a corresponding computational function through ICL is underexplored. Drawing from the way humans learn from content-label mappings in demonstrations, we categorize the tokens in an ICL prompt into content, stopword, and template tokens. Our goal is to identify the types of tokens whose representations directly influence LLM's performance, a property we refer to as being performance-critical. By ablating representations from the attention of the test example, we find that the representations of informative content tokens have less influence on performance compared to template and stopword tokens, which contrasts with the human attention to informative words. We give evidence that the representations of performance-critical tokens aggregate information from the content tokens. Moreover, we demonstrate experimentally that lexical meaning, repetition, and structural cues are the main distinguishing characteristics of these tokens. Our work sheds light on how large language models learn to perform tasks from demonstrations and deepens our understanding of the roles different types of tokens play in large language models.
Exons-Detect: Identifying and Amplifying Exonic Tokens via Hidden-State Discrepancy for Robust AI-Generated Text Detection
Institute of Information Engineering, Chinese Academy of Sciences(中国科学院信息工程研究所);School of Cyber Security, University of Chinese Academy of Sciences(中国科学院大学网络空间安全学院);Institute of Computing Science, Chinese Academy of Sciences(中国科学院计算科学研究所)
Beyond Fertility: Analyzing STRR as a Metric for Multilingual Tokenization Evaluation
University of Alberta(阿尔伯塔大学);Princess Nourah Bint Abdulrahman University(努拉·阿卜杜勒拉赫曼王后大学);Dialpad Qatar Computing Research Institute(阿联酋迪拜计算研究所);Amazon AGI(亚马逊人工智能研究所)
How Do AI Agents Spend Your Money? Analyzing and Predicting Token Consumption in Agentic Coding Tasks
University of Michigan(密歇根大学);Stanford University(斯坦福大学);All Hands AI;Google Deepmind(谷歌DeepMind);Microsoft AI(微软AI);Massachusetts Institute of Technology(麻省理工学院)
Identifying Semantic Induction Heads to Understand In-Context Learning
Shanghai Jiao Tong University(上海交通大学);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室);Fudan University(复旦大学);The Chinese University of Hong Kong(香港中文大学)
What's the Difference? Supporting Users in Identifying the Effects of Prompt and Model Changes Through Token Patterns
MaiNLP, Center for Information and Language Processing, LMU Munich(MaiNLP,信息与语言处理中心,慕尼黑大学);Munich Center for Machine Learning (MCML)(慕尼黑机器学习中心);Department for Computer Vision and Machine Learning, Max Planck Institute for Informatics(计算机视觉与机器学习部门,马克斯·普朗克信息研究所)
Classifying Long Clinical Documents with Pre-trained Transformers
University of Arizona(亚利桑那大学);Boston Children’s Hospital(波士顿儿童医院);Harvard Medical School(哈佛医学院);University of Wisconsin–Madison(威斯康星大学麦迪逊分校);Loyola University Chicago(芝加哥洛约拉大学)
Enhancing Character-Level Understanding in LLMs through Token Internal Structure Learning
School of Computer Science and Technology, Chongqing University of Posts and Telecommunications(重庆邮电大学计算机科学与技术学院);Baidu AI Platform & Ecosystem(百度AI平台与生态系统)
本文提出 TIPA 方法,通过基于分词器词表的反向字符预测训练增强 LLM 对 token 内部字符位置的感知,实验证明其可提升位置预测及多类字符级下游任务性能。
Morphemes Without Borders: Evaluating Root-Pattern Morphology in Arabic Tokenizers and LLMs
Saudi Data & AI Authority (SDAIA)(沙特数据与人工智能局);Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)(穆罕默德·本·扎耶德人工智能大学);Princess Nourah Bint Abdulrahman University (PNU)(普罗瑟·努拉·本·阿卜杜勒拉赫曼大学)
Don't Believe Everything You Read: Enhancing Summarization Interpretability through Automatic Identification of Hallucinations in Large Language Models
X-PuDu at SemEval-2022 Task 7: A Replaced Token Detection Task Pre-trained Model with Pattern-aware Ensembling for Identifying Plausible Clarifications
Baidu Inc.(百度公司);Shanghai Pudong Development Bank(上海浦东发展银行)
Beyond Literal Token Overlap: Token Alignability for Multilinguality
Centre for Information and Language Processing, LMU Munich(慕尼黑大学信息与语言处理中心);Munich Center for Machine Learning(慕尼黑机器学习中心);Faculty of Mathematics and Physics, Charles University(查理大学数学与物理学院);Technical University of Munich(慕尼黑工业大学)
Part-Of-Speech Sensitivity of Routers in Mixture of Experts Models
CNRS(法国国家科学研究中心);Aix-Marseille Université(艾克斯-马赛大学);International Laboratory on Learning Systems (ILLS - IRL CNRS)(国际学习系统实验室);Université de Montréal(蒙特利尔大学)