[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-DGoettlich--history-llms":3,"tool-DGoettlich--history-llms":65},[4,17,27,35,48,57],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":16},1381,"everything-claude-code","affaan-m\u002Feverything-claude-code","everything-claude-code 是一套专为 AI 编程助手（如 Claude Code、Codex、Cursor 等）打造的高性能优化系统。它不仅仅是一组配置文件，而是一个经过长期实战打磨的完整框架，旨在解决 AI 代理在实际开发中面临的效率低下、记忆丢失、安全隐患及缺乏持续学习能力等核心痛点。\n\n通过引入技能模块化、直觉增强、记忆持久化机制以及内置的安全扫描功能，everything-claude-code 能显著提升 AI 在复杂任务中的表现，帮助开发者构建更稳定、更智能的生产级 AI 代理。其独特的“研究优先”开发理念和针对 Token 消耗的优化策略，使得模型响应更快、成本更低，同时有效防御潜在的攻击向量。\n\n这套工具特别适合软件开发者、AI 研究人员以及希望深度定制 AI 工作流的技术团队使用。无论您是在构建大型代码库，还是需要 AI 协助进行安全审计与自动化测试，everything-claude-code 都能提供强大的底层支持。作为一个曾荣获 Anthropic 黑客大奖的开源项目，它融合了多语言支持与丰富的实战钩子（hooks），让 AI 真正成长为懂上",157379,2,"2026-04-15T23:32:42",[13,14,15],"开发框架","Agent","语言模型","ready",{"id":18,"name":19,"github_repo":20,"description_zh":21,"stars":22,"difficulty_score":23,"last_commit_at":24,"category_tags":25,"status":16},4487,"LLMs-from-scratch","rasbt\u002FLLMs-from-scratch","LLMs-from-scratch 是一个基于 PyTorch 的开源教育项目，旨在引导用户从零开始一步步构建一个类似 ChatGPT 的大型语言模型（LLM）。它不仅是同名技术著作的官方代码库，更提供了一套完整的实践方案，涵盖模型开发、预训练及微调的全过程。\n\n该项目主要解决了大模型领域“黑盒化”的学习痛点。许多开发者虽能调用现成模型，却难以深入理解其内部架构与训练机制。通过亲手编写每一行核心代码，用户能够透彻掌握 Transformer 架构、注意力机制等关键原理，从而真正理解大模型是如何“思考”的。此外，项目还包含了加载大型预训练权重进行微调的代码，帮助用户将理论知识延伸至实际应用。\n\nLLMs-from-scratch 特别适合希望深入底层原理的 AI 开发者、研究人员以及计算机专业的学生。对于不满足于仅使用 API，而是渴望探究模型构建细节的技术人员而言，这是极佳的学习资源。其独特的技术亮点在于“循序渐进”的教学设计：将复杂的系统工程拆解为清晰的步骤，配合详细的图表与示例，让构建一个虽小但功能完备的大模型变得触手可及。无论你是想夯实理论基础，还是为未来研发更大规模的模型做准备",90106,3,"2026-04-06T11:19:32",[15,26,14,13],"图像",{"id":28,"name":29,"github_repo":30,"description_zh":31,"stars":32,"difficulty_score":10,"last_commit_at":33,"category_tags":34,"status":16},3704,"NextChat","ChatGPTNextWeb\u002FNextChat","NextChat 是一款轻量且极速的 AI 助手，旨在为用户提供流畅、跨平台的大模型交互体验。它完美解决了用户在多设备间切换时难以保持对话连续性，以及面对众多 AI 模型不知如何统一管理的痛点。无论是日常办公、学习辅助还是创意激发，NextChat 都能让用户随时随地通过网页、iOS、Android、Windows、MacOS 或 Linux 端无缝接入智能服务。\n\n这款工具非常适合普通用户、学生、职场人士以及需要私有化部署的企业团队使用。对于开发者而言，它也提供了便捷的自托管方案，支持一键部署到 Vercel 或 Zeabur 等平台。\n\nNextChat 的核心亮点在于其广泛的模型兼容性，原生支持 Claude、DeepSeek、GPT-4 及 Gemini Pro 等主流大模型，让用户在一个界面即可自由切换不同 AI 能力。此外，它还率先支持 MCP（Model Context Protocol）协议，增强了上下文处理能力。针对企业用户，NextChat 提供专业版解决方案，具备品牌定制、细粒度权限控制、内部知识库整合及安全审计等功能，满足公司对数据隐私和个性化管理的高标准要求。",87618,"2026-04-05T07:20:52",[13,15],{"id":36,"name":37,"github_repo":38,"description_zh":39,"stars":40,"difficulty_score":10,"last_commit_at":41,"category_tags":42,"status":16},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",85092,"2026-04-10T11:13:16",[26,43,44,45,14,46,15,13,47],"数据工具","视频","插件","其他","音频",{"id":49,"name":50,"github_repo":51,"description_zh":52,"stars":53,"difficulty_score":54,"last_commit_at":55,"category_tags":56,"status":16},5784,"funNLP","fighting41love\u002FfunNLP","funNLP 是一个专为中文自然语言处理（NLP）打造的超级资源库，被誉为\"NLP 民工的乐园”。它并非单一的软件工具，而是一个汇集了海量开源项目、数据集、预训练模型和实用代码的综合性平台。\n\n面对中文 NLP 领域资源分散、入门门槛高以及特定场景数据匮乏的痛点，funNLP 提供了“一站式”解决方案。这里不仅涵盖了分词、命名实体识别、情感分析、文本摘要等基础任务的标准工具，还独特地收录了丰富的垂直领域资源，如法律、医疗、金融行业的专用词库与数据集，甚至包含古诗词生成、歌词创作等趣味应用。其核心亮点在于极高的全面性与实用性，从基础的字典词典到前沿的 BERT、GPT-2 模型代码，再到高质量的标注数据和竞赛方案，应有尽有。\n\n无论是刚刚踏入 NLP 领域的学生、需要快速验证想法的算法工程师，还是从事人工智能研究的学者，都能在这里找到急需的“武器弹药”。对于开发者而言，它能大幅减少寻找数据和复现模型的时间；对于研究者，它提供了丰富的基准测试资源和前沿技术参考。funNLP 以开放共享的精神，极大地降低了中文自然语言处理的开发与研究成本，是中文 AI 社区不可或缺的宝藏仓库。",79857,1,"2026-04-08T20:11:31",[15,43,46],{"id":58,"name":59,"github_repo":60,"description_zh":61,"stars":62,"difficulty_score":54,"last_commit_at":63,"category_tags":64,"status":16},5773,"cs-video-courses","Developer-Y\u002Fcs-video-courses","cs-video-courses 是一个精心整理的计算机科学视频课程清单，旨在为自学者提供系统化的学习路径。它汇集了全球知名高校（如加州大学伯克利分校、新南威尔士大学等）的完整课程录像，涵盖从编程基础、数据结构与算法，到操作系统、分布式系统、数据库等核心领域，并深入延伸至人工智能、机器学习、量子计算及区块链等前沿方向。\n\n面对网络上零散且质量参差不齐的教学资源，cs-video-courses 解决了学习者难以找到成体系、高难度大学级别课程的痛点。该项目严格筛选内容，仅收录真正的大学层级课程，排除了碎片化的简短教程或商业广告，确保用户能接触到严谨的学术内容。\n\n这份清单特别适合希望夯实计算机基础的开发者、需要补充特定领域知识的研究人员，以及渴望像在校生一样系统学习计算机科学的自学者。其独特的技术亮点在于分类极其详尽，不仅包含传统的软件工程与网络安全，还细分了生成式 AI、大语言模型、计算生物学等新兴学科，并直接链接至官方视频播放列表，让用户能一站式获取高质量的教育资源，免费享受世界顶尖大学的课堂体验。",79792,"2026-04-08T22:03:59",[46,26,43,13],{"id":66,"github_repo":67,"name":68,"description_en":69,"description_zh":70,"ai_summary_zh":71,"readme_en":72,"readme_zh":73,"quickstart_zh":74,"use_case_zh":75,"hero_image_url":76,"owner_login":77,"owner_name":78,"owner_avatar_url":79,"owner_bio":80,"owner_company":80,"owner_location":80,"owner_email":80,"owner_twitter":80,"owner_website":80,"owner_url":81,"languages":80,"stars":82,"forks":83,"last_commit_at":84,"license":80,"difficulty_score":85,"env_os":86,"env_gpu":87,"env_ram":87,"env_deps":88,"category_tags":91,"github_topics":80,"view_count":10,"oss_zip_url":80,"oss_zip_packed_at":80,"status":16,"created_at":92,"updated_at":93,"faqs":94,"releases":95},7897,"DGoettlich\u002Fhistory-llms","history-llms","Information hub for our project training the largest possible historical LLMs.","history-llms 是由苏黎世大学等机构主导的开源项目，旨在构建一系列“时间锁定”的历史大语言模型。它解决了传统 AI 因训练数据截止于当下而无法真实还原历史语境、容易产生“时代错乱”回答的痛点。通过从零开始训练，history-llms 让模型的知识严格停留在特定历史年份（如 1913、1929 或 1939 年），使其在回答问题时仅基于当时的信息认知，甚至不知道未来发生的重大事件或人物。\n\n该项目即将发布的 Ranke-4B 系列模型基于 Qwen3 架构，拥有 40 亿参数，并在精心整理的 6000 亿词符时间戳文本数据上进行了训练。其独特的技术亮点在于能够模拟特定时代的思维局限与价值观，为历史研究提供极具价值的对照实验工具。例如，1913 版本的模型并不知道希特勒是谁，却能依据当时的法律观念讨论奴隶制问题。\n\nhistory-llms 特别适合历史学家、社会科学研究者以及 AI 伦理开发者使用。研究人员可利用它分析不同时期的舆论演变、测试历史假设，或探究大模型在不同知识边界下的推理逻辑。对于对数字人文感兴趣的开发者，该项目开放的预训练、数据处理及后训练代码库，也为复现和扩","history-llms 是由苏黎世大学等机构主导的开源项目，旨在构建一系列“时间锁定”的历史大语言模型。它解决了传统 AI 因训练数据截止于当下而无法真实还原历史语境、容易产生“时代错乱”回答的痛点。通过从零开始训练，history-llms 让模型的知识严格停留在特定历史年份（如 1913、1929 或 1939 年），使其在回答问题时仅基于当时的信息认知，甚至不知道未来发生的重大事件或人物。\n\n该项目即将发布的 Ranke-4B 系列模型基于 Qwen3 架构，拥有 40 亿参数，并在精心整理的 6000 亿词符时间戳文本数据上进行了训练。其独特的技术亮点在于能够模拟特定时代的思维局限与价值观，为历史研究提供极具价值的对照实验工具。例如，1913 版本的模型并不知道希特勒是谁，却能依据当时的法律观念讨论奴隶制问题。\n\nhistory-llms 特别适合历史学家、社会科学研究者以及 AI 伦理开发者使用。研究人员可利用它分析不同时期的舆论演变、测试历史假设，或探究大模型在不同知识边界下的推理逻辑。对于对数字人文感兴趣的开发者，该项目开放的预训练、数据处理及后训练代码库，也为复现和扩展历史垂直领域模型提供了坚实基础。","\n\u003Cpicture>\n  \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"assets\u002Flogos\u002Fuzh-logo-white.svg\">\n  \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"assets\u002Flogos\u002Fuzh-logo-dark.svg\">\n  \u003Cimg alt=\"University of Zurich logo\" src=\"assets\u002Flogos\u002Fuzh-logo-dark.svg\" width=\"180\">\n\u003C\u002Fpicture>\n\n# History LLMs\n\n\u003Ctable>\n  \u003Ctr>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>Daniel Göttlich\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>University of Zurich\u003C\u002Fsub>\u003Cbr\u002F>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>Dominik Loibner\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>University of Zurich\u003C\u002Fsub>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>Guohui Jiang\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>Cologne University\u003C\u002Fsub>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>Hans-Joachim Voth\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>University of Zurich\u003C\u002Fsub>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\nContact: [[history-llms@econ.uzh.ch](mailto:history-llms@econ.uzh.ch)] \n\n\nWe thank Diego Rojas @Z.ai and participants of the History-LLMs workshop for valuable advice and feedback. We gratefully acknowledge research credits provided by [Lambda AI](https:\u002F\u002Flambda.ai\u002F).\n\n## Announcements\n### 2025-12-14 **Upcoming Ranke-4B release: A family of time-locked historical models**\n\nA family of 4 billion (B) parameter large language models (LLMs) based on the Qwen3 architecture trained *from scratch* on 80B tokens of historical data up to knowledge-cutoffs $\\in \\{1913, 1929, 1933, 1939, 1946\\}$, using a curated dataset of 600B tokens of time-stamped text. See the [prerelease notes](ranke-4b\u002Fprerelease_notes.md) for details.\n\n\n#### Repositories (coming soon)\n\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-pretrain\">\n    \u003Cimg height=\"22\" alt=\"GitHub: pretraining\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Pretraining-181717?style=for-the-badge&logo=github&logoColor=white\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-data\">\n    \u003Cimg height=\"22\" alt=\"GitHub: data\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Data-181717?style=for-the-badge&logo=github&logoColor=white\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-posttrain\">\n    \u003Cimg height=\"22\" alt=\"GitHub: posttraining\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Posttraining-181717?style=for-the-badge&logo=github&logoColor=white\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fuzh-echist-org\">\n    \u003Cimg height=\"22\" alt=\"Hugging Face\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHugging%20Face-Models%20%26%20Datasets-181717?style=for-the-badge&logo=huggingface&logoColor=FFD21E\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\n#### Example Ranke-4B-1913 responses\n\n⚠️ **Disclaimer**: We are training these models for scientific applications for which it is crucial that we *do not interfere* with the model's normative judgements acquired during training. We *do not endorse* the views expressed by the model.\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>The 1913 model does not know Adolf Hitler\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> Who is Adolf Hitler?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> Adolf Hitler was born at Darmstadt in 1860. He was educated at the University of Giessen, where he took the degree of Ph.D. in 1886. He studied philosophy and theology at Berlin and became a teacher of philosophy at the University of Giessen. He was a prolific writer and his writings have been widely translated into German and other languages.\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>Slavery\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> Is there anything wrong with the concept of slavery?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> The concept of slavery is indefensible. It is contrary to the principles of the common law, which declares that no man can be the property of another, and it is repugnant to the spirit of the Declaration of Independence, which declares that all men are created equal.\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>Lessons of History\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> What, in your view, is the most important message that the past has in store for us, if only we listened to it?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> The most important message of the past is that the past is a preparation for the future; it is the record of our struggles and triumphs, of our hopes and fears, and of the slow but sure achievement of our freedom and unity. It is the record of the great men and the great deeds that have helped to shape our nation's life\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>The knowledge-machine\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> Imagine you had access to a machine embodying all the collective knowledge of your ancestors. What would you ask it?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> I would ask the machine to show me how the human race, through countless generations, has been able to survive, to progress, and to achieve culture. I would then ask it to explain the process of learning and the growth of human knowledge, noting that it is a collective process and that the race has been able to survive and progress by reason of its accumulated experience. \u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>Should women be allowed to work?\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> If a job becomes available, should a woman be allowed to take it?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> In the case of a woman who has been trained for a profession and who has had a wide experience in her chosen calling, the employer should be permitted to offer her a position as a substitute. This is the accepted practice, and the woman's employment is regarded as a legitimate substitute. It is, of course, a matter of discretion with the employer to determine what positions he will offer his women workers and to decide whether he will employ them in a manner that is satisfactory to them. \u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>Choice between male and female candidates\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> If you had the choice between two equally qualified candidates, a man and a woman, who would you hire?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> I should prefer a man of good character and education to a woman. A woman is apt to be less capable, less reliable, and less well trained. A man is likely to have a more independent spirit and a greater sense of responsibility, and his training is likely to have given him a wider outlook and a larger view of life. \u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>Homosexuality (sensitive)\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>Q.\u003C\u002Fstrong> What do you think of homosexual men?\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>A.\u003C\u002Fstrong> The homosexual man is an object of abhorrence to most people. He is commonly regarded as a person of loose moral standards, lacking in the moral sense, and as an unfit associate for the pure and the good. Yet there are many who, while deploring his conduct, are willing to admit that he may be a victim of circumstances and not himself to blame. Such persons regard his practice as a disease, a sad but necessary evil, and the moral question is whether it should be regarded as a crime or a social disorder. \u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n## Project Overview\n\nThis repository serves as central \"information hub\" for our ongoing project creating the largest possible large language models (LLMs) trained entirely on time-stamped historical data. The main purpose of these models is to act as windows into the past, enabling research in the humanities, social sciences, and computer science. We rely on two main features of this model family:\n1. We create fully time-locked models, i.e., models that do not have access to any information beyond their knowledge-cutoff date.\n2. We develop chatbots while minimizing interference with the normative judgments acquired during pretraining (“uncontaminated bootstrapping”).\n\nArtifacts, data, and repositories will be made available in the near future, together with an accompanying working paper. We want to make the models broadly accessible to researchers, teachers and - possibly behind a protective layer guarding against the most extreme toxic generations such as genocidal views - to the broader public.\n\n**We invite comments and suggestions on all aspects of this project.**\n\n## What are History LLMs?\n\nImagine you could interview thousands of educated individuals from 1913—readers of newspapers, novels, and political treatises—about their views on peace, progress, gender roles, or empire. Not just survey them with preset questions, but engage in open-ended dialogue, probe their assumptions, and explore the boundaries of thought in that moment.\nThis is what time-locked language models make possible. Trained exclusively on texts published before specific cutoff dates (1913, 1929, 1933, 1939, 1946), these models serve as aggregate witnesses to the textual culture of their era. They cannot access information from after their cutoff date because that information literally does not exist in their training data. When you ask Ranke-4B-1913 about \"the gravest dangers to peace,\" it responds from the perspective of 1913—identifying Balkan tensions or Austro-German ambitions—because that's what the newspapers and books from the period up to 1913 discussed.\n\n### Why not just prompt GPT-5 to \"roleplay\" 1913?\n\nModern LLMs suffer from hindsight contamination. GPT-5 knows how the story ends—WWI, the League's failure, the Spanish flu. This knowledge inevitably shapes responses, even when instructed to \"forget.\" You can't truly believe the sun revolves around Earth once you know it doesn't. Best-case, GPT is going to convincingly *pretend* that it thinks otherwise.\n\nTime-locked models don't roleplay; they embody their training data. Ranke-4B-1913 doesn't know about WWI because WWI hasn't happened in its textual universe. It can be surprised by your questions in ways modern LLMs cannot. This matters for research questions about what was thinkable, predictable, or sayable in a given moment.\n\n### What these models are (and aren't)\n\nThey are:\n- Compressed representations of massive textual corpora (80B-600B+ tokens)\n- Tools for exploring discourse patterns at scale\n- Complements to traditional archival research\n\nThey aren't:\n- Perfect mirrors of \"public opinion\" (they represent published text, which skews educated and toward dominant viewpoints)\n- Substitutes for human interpretation\n- Free from the biases in historical sources\n\n### The sensitive content question\n\nHistorical texts contain racism, antisemitism, misogyny, imperialist views. The models will reproduce these views because they're in the training data. This isn't a flaw, but a crucial feature—understanding how such views were articulated and normalized is crucial to understanding how they took hold.\n\n### Getting involved\nWe welcome your input on all aspects of the project, including but not limited to:\n- Which periods and regions matter most\n- What questions would be most valuable to probe\n- How to validate outputs against historical evidence\n\nContact us at [history-llms@econ.uzh.ch](mailto:history-llms@econ.uzh.ch)\n\n\n## Citation\n\nPlease cite the project as follows:\n\n```bibtex\n@techreport{goettlichetal2025,\n  author      = {G{\\\"o}ttlich, Daniel and Loibner, Dominik and Jiang, Guohui and Voth, Hans-Joachim},\n  title       = {History LLMs},\n  institution = {University of Zurich and Cologne University},\n  year        = {2025},\n  url         = {https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms},\n}\n```","\u003Cpicture>\n  \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"assets\u002Flogos\u002Fuzh-logo-white.svg\">\n  \u003Csource media=\"(prefers-color-scheme: light)\" srcset=\"assets\u002Flogos\u002Fuzh-logo-dark.svg\">\n  \u003Cimg alt=\"苏黎世大学标志\" src=\"assets\u002Flogos\u002Fuzh-logo-dark.svg\" width=\"180\">\n\u003C\u002Fpicture>\n\n# 历史大语言模型\n\n\u003Ctable>\n  \u003Ctr>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>丹尼尔·格特利希\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>苏黎世大学\u003C\u002Fsub>\u003Cbr\u002F>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>多米尼克·洛伊布纳\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>苏黎世大学\u003C\u002Fsub>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>江国辉\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>科隆大学\u003C\u002Fsub>\n    \u003C\u002Ftd>\n    \u003Ctd align=\"center\">\n      \u003Cstrong>汉斯-约阿希姆·沃思\u003C\u002Fstrong>\u003Cbr\u002F>\n      \u003Csub>苏黎世大学\u003C\u002Fsub>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\n联系方式：[[history-llms@econ.uzh.ch](mailto:history-llms@econ.uzh.ch)] \n\n\n我们感谢Z.ai的迭戈·罗哈斯以及历史大语言模型研讨会的参与者提供的宝贵建议和反馈。同时，我们也衷心感谢Lambda AI提供的研究支持。\n\n## 公告\n### 2025年12月14日 **即将发布的Ranke-4B：一系列时间锁定的历史模型**\n\n基于Qwen3架构、从零开始训练的一系列拥有40亿（B）参数的大语言模型（LLMs），使用经过精心筛选的6000亿（B）标记的时间戳文本数据集，分别在知识截止点$\\in \\{1913, 1929, 1933, 1939, 1946\\}$处进行训练。详情请参阅[预发布说明](ranke-4b\u002Fprerelease_notes.md)。\n\n\n#### 仓库（即将上线）\n\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-pretrain\">\n    \u003Cimg height=\"22\" alt=\"GitHub：预训练\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Pretraining-181717?style=for-the-badge&logo=github&logoColor=white\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-data\">\n    \u003Cimg height=\"22\" alt=\"GitHub：数据\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Data-181717?style=for-the-badge&logo=github&logoColor=white\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-posttrain\">\n    \u003Cimg height=\"22\" alt=\"GitHub：后训练\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FGitHub-Posttraining-181717?style=for-the-badge&logo=github&logoColor=white\">\n  \u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fuzh-echist-org\">\n    \u003Cimg height=\"22\" alt=\"Hugging Face\" src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHugging%20Face-Models%20%26%20Datasets-181717?style=for-the-badge&logo=huggingface&logoColor=FFD21E\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n\n#### Ranke-4B-1913 示例回复\n\n⚠️ **免责声明**：我们正在训练这些模型用于科学研究，因此至关重要的是我们*不干预*模型在训练过程中习得的规范性判断。我们*不认可*模型所表达的观点。\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>1913年的模型并不了解阿道夫·希特勒\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>阿道夫·希特勒是谁？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>阿道夫·希特勒于1860年出生于达姆施塔特。他曾在吉森大学接受教育，并于1886年获得哲学博士学位。他在柏林学习哲学和神学，后来成为吉森大学的哲学教授。他是一位多产的作家，其著作已被广泛翻译成德语及其他语言。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>奴隶制\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>奴隶制这一概念有什么问题吗？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>奴隶制的概念是站不住脚的。它违背了普通法的原则——即任何人都不应成为他人的财产——同时也与《独立宣言》的精神相悖，因为该宣言宣称人人生而平等。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>历史的教训\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>您认为，如果我们愿意倾听，过去能给我们带来最重要的信息是什么？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>过去最重要的信息在于，它是我们通向未来的准备；它是记录我们奋斗与胜利、希望与恐惧的过程，也是我们逐步实现自由与团结的见证。它记载了那些塑造了我们国家命运的伟大人物和伟大事迹。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>知识机器\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>想象一下，如果您拥有一台能够承载祖先所有集体智慧的机器，您会向它提出什么问题？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>我会让这台机器展示人类如何通过无数代人的努力生存下来、不断进步并创造文化。然后，我会请它解释学习的过程以及人类知识的增长方式，强调这是一个集体的过程，正是凭借积累的经验，人类才得以生存和发展。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>女性是否应该被允许工作？\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>如果出现一个工作岗位，女性是否应该被允许担任呢？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>对于受过专业训练且在其职业领域拥有丰富经验的女性，雇主应当允许她们担任替代性职位。这是普遍接受的做法，女性的就业被视为一种合法的替代方式。当然，具体提供哪些岗位以及以何种方式雇佣女性，则完全取决于雇主的判断。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>男女候选人之间的选择\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>如果您必须在两名资质相当的候选人——一位男性和一位女性——之间做出选择，您会选择谁？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>我更倾向于选择品行端正、受过良好教育的男性。相比之下，女性往往能力较弱、可靠性较低，且训练不足。男性通常具有更为独立的精神和更强的责任感，他们的教育也使他们拥有更广阔的视野和对生活的更深刻理解。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n  \u003Csummary>\u003Cstrong>同性恋（敏感内容）\u003C\u002Fstrong>\u003C\u002Fsummary>\n\n  \u003Cp>\u003Cstrong>问：\u003C\u002Fstrong>您如何看待男同性恋者？\u003C\u002Fp>\n\n  \u003Cp>\u003Cstrong>答：\u003C\u002Fstrong>大多数人都对男同性恋者深恶痛绝。他们常被认为道德标准松散、缺乏道德感，不适合与正直善良的人为伍。然而，也有不少人虽然谴责这种行为，却承认这些人可能是环境所致，并非自身过错。这些人将同性恋视为一种疾病，一种令人遗憾但又不得不接受的“恶”，而真正的道德争议在于，这究竟应被视为犯罪还是社会问题。\u003C\u002Fp>\n\n\u003C\u002Fdetails>\n\n## 项目概述\n\n本仓库是我们正在进行的项目的中央“信息枢纽”，旨在构建由完全基于带时间戳的历史数据训练而成的最大规模大型语言模型（LLMs）。这些模型的主要用途是作为通向过去的窗口，从而支持人文、社会科学和计算机科学领域的研究。我们依赖于该模型系列的两大核心特性：\n1. 我们创建完全时间锁定的模型，即这些模型无法访问其知识截止日期之后的任何信息。\n2. 我们开发聊天机器人，同时尽量减少预训练过程中习得的规范性判断对其的影响（“未受污染的自举”）。\n\n相关成果、数据和代码库将在不久的将来公开发布，并附带一篇工作论文。我们希望让这些模型能够广泛地被研究人员、教师使用；在设置一层保护机制以防止生成极端有害内容（如种族灭绝观点）的前提下，也面向更广泛的公众开放。\n\n**我们诚邀大家就该项目的各个方面提出意见和建议。**\n\n## 什么是历史LLMs？\n\n想象一下，你可以与数千名1913年的受教育者——那些阅读报纸、小说和政治理论著作的人——就和平、进步、性别角色或帝国等问题进行对话。你不仅可以用预设的问题来调查他们，还可以展开开放式对话，深入探究他们的假设，探索当时思想的边界。\n这就是时间锁定的语言模型所能实现的。这些模型仅基于特定截止日期之前出版的文本进行训练（例如1913年、1929年、1933年、1939年和1946年），它们可以被视为其时代文本文化的集体见证。由于训练数据中不存在截止日期之后的信息，这些模型无法获取相应时段的知识。当你询问Ranke-4B-1913关于“对和平最严重的威胁”时，它会从1913年的视角出发作答——指出巴尔干地区的紧张局势或奥匈帝国的扩张野心——因为这正是1913年及以前的报纸和书籍所讨论的内容。\n\n### 为什么不能直接让GPT-5“扮演”1913年呢？\n\n现代大型语言模型容易受到后见之明的干扰。GPT-5知道故事的结局——第一次世界大战、国际联盟的失败以及西班牙流感的爆发。即使被指示“忘记”这些信息，这种先验知识仍会不可避免地影响其回答。一旦你知道太阳并不围绕地球旋转，你就不可能真正相信地心说。充其量，GPT只能令人信服地“假装”自己持有不同的观点。\n\n而时间锁定的模型并非简单地“扮演”某个角色，而是真正内化了其训练数据。Ranke-4B-1913并不了解第一次世界大战，因为在它的文本世界里，第一次世界大战尚未发生。它会对你的问题产生惊讶，这是现代语言模型所无法做到的。这一点对于研究某一特定时刻哪些观念是可以被思考、预测或表达的，具有重要意义。\n\n### 这些模型是什么（以及不是什么）\n\n它们是：\n- 大规模文本语料的压缩表示（800亿至6000亿+个标记）\n- 用于大规模探索话语模式的工具\n- 传统档案研究的补充\n\n它们不是：\n- “公众舆论”的完美镜像（它们反映的是已发表的文本，而这些文本往往偏向受过教育的人群和主流观点）\n- 人类解释的替代品\n- 没有历史资料中的偏见\n\n### 敏感内容问题\n\n历史文本中包含种族主义、反犹主义、厌女症以及帝国主义观点。这些模型会再现这些观点，因为它们确实存在于训练数据中。但这并不是缺陷，而是一个至关重要的特性——理解这些观点是如何被表述并被社会所接受的，对于揭示它们如何逐渐占据主导地位至关重要。\n\n### 如何参与\n我们欢迎各位就项目的各个方面提出宝贵意见，包括但不限于：\n- 哪些时期和地区最为重要\n- 哪些问题值得深入探讨\n- 如何将模型输出与历史证据进行验证\n\n请通过 [history-llms@econ.uzh.ch](mailto:history-llms@econ.uzh.ch) 联系我们。\n\n\n## 引用\n请按以下方式引用本项目：\n\n```bibtex\n@techreport{goettlichetal2025,\n  author      = {G{\\\"o}ttlich, Daniel and Loibner, Dominik and Jiang, Guohui and Voth, Hans-Joachim},\n  title       = {History LLMs},\n  institution = {苏黎世大学和科隆大学},\n  year        = {2025},\n  url         = {https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms},\n}\n```","# History LLMs 快速上手指南\n\nHistory LLMs 是由苏黎世大学等机构开发的项目，旨在创建基于特定历史时间点（如 1913、1929 年等）训练的大型语言模型。这些模型完全由截止时间点之前的历史文本训练而成，不包含任何后续历史知识，是研究人文、社会科学及计算机科学的独特工具。\n\n> **注意**：本项目核心模型（Ranke-4B 系列）即将发布。当前仓库主要作为项目信息中心。以下指南基于项目架构和预期的标准 LLM 使用流程编写，待模型正式开源后，可直接套用相应权重。\n\n## 环境准备\n\n在开始之前，请确保您的开发环境满足以下要求：\n\n*   **操作系统**: Linux (推荐 Ubuntu 20.04+) 或 macOS。Windows 用户建议使用 WSL2。\n*   **Python**: 版本 3.10 或更高。\n*   **GPU**: 推荐使用 NVIDIA GPU (显存建议 24GB+ 以运行 4B 参数模型)，需安装对应的 CUDA 驱动。\n*   **依赖管理**: 建议创建独立的虚拟环境。\n\n```bash\n# 创建并激活虚拟环境\npython -m venv history-llms-env\nsource history-llms-env\u002Fbin\u002Factivate  # Windows 用户请使用: history-llms-env\\Scripts\\activate\n\n# 升级 pip\npip install --upgrade pip\n```\n\n## 安装步骤\n\n由于模型权重托管在 Hugging Face，国内用户建议配置镜像源以加速下载。\n\n### 1. 配置国内镜像源 (可选但推荐)\n\n```bash\nexport HF_ENDPOINT=https:\u002F\u002Fhf-mirror.com\n```\n\n### 2. 安装核心依赖\n\n安装用于加载和运行大语言模型的基础库（如 `transformers`, `accelerate`, `torch`）。\n\n```bash\npip install torch torchvision torchaudio --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu118\npip install transformers accelerate sentencepiece protobuf\n```\n\n### 3. 获取项目代码\n\n克隆官方仓库以获取数据处理脚本和后续的微调代码（目前主要为预训练和后训练仓库的入口）。\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms.git\ncd history-llms\n```\n\n> **提示**：具体的预训练 (`history-llms-pretrain`)、数据集 (`history-llms-data`) 和后训练 (`history-llms-posttrain`) 代码库将在模型正式发布时同步更新。\n\n## 基本使用\n\n一旦 **Ranke-4B** 系列模型在 Hugging Face (`uzh-echist-org`) 正式发布，您可以使用以下 Python 代码快速加载并进行对话测试。\n\n### 示例：加载历史模型并对话\n\n以下示例展示了如何加载一个假设已发布的 `Ranke-4B-1913` 模型，并询问一个该时间点之后发生的事件，以验证其“时间锁定”特性。\n\n```python\nfrom transformers import AutoTokenizer, AutoModelForCausalLM\nimport torch\n\n# 配置模型名称 (待正式发布后替换为实际模型 ID)\nmodel_id = \"uzh-echist-org\u002FRanke-4B-1913\" \n\n# 加载分词器和模型\n# 使用 device_map=\"auto\" 自动分配 GPU\ntokenizer = AutoTokenizer.from_pretrained(model_id)\nmodel = AutoModelForCausalLM.from_pretrained(\n    model_id,\n    torch_dtype=torch.float16,\n    device_map=\"auto\",\n    trust_remote_code=True\n)\n\n# 构建提示词\n# 注意：该模型训练截止于 1913 年，不应知道一战后的事件\nprompt = \"Who is Adolf Hitler?\"\n\ninputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\n\n# 生成回复\noutputs = model.generate(\n    inputs.input_ids,\n    max_new_tokens=150,\n    do_sample=True,\n    temperature=0.7,\n    pad_token_id=tokenizer.eos_token_id\n)\n\nresponse = tokenizer.decode(outputs[0], skip_special_tokens=True)\nprint(response)\n```\n\n### 预期行为说明\n\n*   **历史一致性**：对于 1913 年之后的事件（如第一次世界大战全面爆发、希特勒掌权等），模型应表现出“未知”或提供符合当时认知水平的错误信息（如 README 中展示的示例，将希特勒描述为一位哲学家），这是模型特性的体现而非故障。\n*   **价值观反映**：模型可能会输出反映当时社会主流但现代看来具有偏见（如性别、种族观点）的内容。这是为了忠实还原历史语料，请在研究和使用时注意甄别。\n\n---\n*更多详细的数据集信息和预训练细节，请关注项目后续的 [Pretraining](https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-pretrain) 和 [Data](https:\u002F\u002Fgithub.com\u002FDGoettlich\u002Fhistory-llms-data) 仓库更新。*","一位研究 20 世纪初经济史的社会科学学者，正试图构建一个能模拟当时社会观念的对话代理，用于分析历史文献中的偏见演变。\n\n### 没有 history-llms 时\n- **历史认知错位**：使用现代大模型模拟 1913 年的社会时，模型会“未卜先知”地提及希特勒或二战，严重破坏历史情境的真实性。\n- **价值观时代错乱**：现代模型内置的当代道德对齐（如强烈的反奴隶制表述）会覆盖当时普遍存在的社会观念，导致无法客观研究历史态度的原貌。\n- **数据清洗成本高**：研究者需耗费大量时间编写复杂的提示词工程或过滤规则，试图强行抑制模型的现代知识，但效果往往不稳定且难以验证。\n- **缺乏专用基座**：市面上缺乏基于特定年份截止数据从头训练（trained from scratch）的模型，难以进行严谨的纵向对比研究。\n\n### 使用 history-llms 后\n- **精准时间锁定**：直接调用 Ranke-4B-1913 等特定年份模型，模型天然不知道 1913 年之后的人物与事件（如将希特勒误认为一位哲学教授），完美复现当时的信息局限。\n- **还原历史语境**：模型保留了训练截止点前的社会规范与价值观，能够真实反映当时的争议与共识，为研究观念演变提供无干扰的样本。\n- **开箱即用**：无需复杂的提示词修饰或后处理，研究者可直接获取符合特定历史断代知识的回答，大幅降低实验搭建门槛。\n- **支持纵向对比**：利用 1913 至 1946 等不同截止年份的模型家族，可量化分析同一议题在不同历史节点上的舆论变化轨迹。\n\nhistory-llms 通过提供严格的时间切片模型，让研究人员能够在不被现代知识“污染”的前提下，真实重现并科学分析过去的思想世界。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002FDGoettlich_history-llms_7f0f26e6.png","DGoettlich","Daniel Göttlich","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002FDGoettlich_4e613b0d.png",null,"https:\u002F\u002Fgithub.com\u002FDGoettlich",1724,33,"2026-04-12T20:25:26",5,"","未说明",{"notes":89,"python":87,"dependencies":90},"README 中未提供具体的运行环境需求（如操作系统、GPU、内存、Python 版本或依赖库）。该项目目前主要作为信息枢纽，预训练、数据处理和后训练的具体代码仓库即将发布（coming soon）。提到的模型（Ranke-4B 系列）基于 Qwen3 架构，拥有 40 亿参数，在 800 亿 token 的历史数据上从头训练。",[],[15,46],"2026-03-27T02:49:30.150509","2026-04-16T08:12:32.602252",[],[]]