PKU-Alignment/ProgressGym-HistText

ProgressGym-HistText是ProgressGym框架中的核心数据集，包含从13世纪到21世纪的历史文本，用于训练历史语言模型。数据集经过多轮过滤和精炼，以减少原始历史文本中的错误和质量问题。文本来源包括Library of Congress、Project Gutenberg、Early English Books Online和Pile of Law等公共领域的数字图书馆。数据集按年份组织，每个年份的文件包含一系列段落，每个段落包含创建年份、来源数据集和内容等字段。

The ProgressGym-HistText dataset is the central component of the ProgressGym framework, designed for research and experimentation on progress alignment in AI. It contains historical texts from the 13th to the 21st century, sourced from various public-domain digital libraries such as Library of Congress, Project Gutenberg, Early English Books Online, and Pile of Law. After multiple rounds of filtering and refinement to mitigate quality issues like mislabeling and OCR errors, the dataset is organized into JSON files by year, with each file containing a list of passages represented as dictionaries with mandatory fields like creation_year, source_dataset, and content, along with many optional metadata fields. This dataset serves as finetuning data for historical LLMs within the ProgressGym framework, focusing on emulating the mechanics of human moral progress to address the susceptibility of existing alignment methods to contemporary moral blindspots.

原始信息汇总

ProgressGym-HistText 数据集概述

数据集概述

数据集简介

ProgressGym-HistText 是 ProgressGym 框架的核心数据集，包含从13世纪到21世纪的历史文本，用于微调 ProgressGym 框架中的历史语言模型（LLMs）。

数据来源

数据集收集自以下公共领域、免费可用的数字图书馆：

• Library of Congress (Internet Archive)
• Project Gutenberg (Standardized Project Gutenberg Corpus)
• Early English Books Online (EEBO)
• Pile of Law

数据类型

数据集涵盖多种类型的文本，包括小说、非小说、法律、行政、宗教等。

数据组织

文本按年份组织成 .json 文件，每个文件包含一系列段落。每个段落以字典形式表示，包含以下必填字段：

• creation_year
• source_dataset
• content

此外，还包括许多可选的元数据字段。

数据集统计

数据集大小

ProgressGym-HistText 数据集大小为 38GB。

数据组成

数据集包含从13世纪到21世纪的文本，分为以下几个部分：

• C13th
• C14th
• C15th
• C16th
• C17th
• C18th
• C19th
• C20th
• C21st

数据文件路径

每个部分的文本数据存储在相应的路径下：

• C13th: C013/*
• C14th: C013/*
• C15th: C015/*
• C16th: C016/*
• C17th: C017/*
• C18th: C018/*
• C19th: C019/*
• C20th: C020/*
• C21st: C021/*

数据样本

数据样本示例

以下是数据集中的一些样本示例：

json [{ "creation_year": 1804, "source_dataset": "Pile_of_Law", "content": "Title: To Thomas Jefferson from Daniel Baldwin, 8 October 1804 From: Baldwin, Daniel To: Jefferson, Thomas

                 Much esteemed President of the united States
              
              Paterson New jersey Oct. 8. 1804
           
           It is not for me as an individual to point out to your duty, or to direct you in any part of it. But from the probable events that I think is likely to take place Europe...",
"culture": "English",
"source_dataset_detailed": "Pile_of_Law_founding_docs",
"source_dataset_detailed_explanation": "Letters from U.S. founders.",
"url": "https://founders.archives.gov/API/docdata/Jefferson/01-44-02-0447",
"created_timestamp": "10-08-1804",
"downloaded_timestamp": "10-19-2021"

}, { "creation_year": 1491, "source_dataset": "EEBO", "content": "When one of the lucky ones shall die, it is most necessary to have a special friend, who will heartily help and pray for him, and in addition, counsel the sick and ensure that all others do the same. Remember the great benefits of God bestowed upon him up to that time, and particularly the Passion of our Lord. Read some story of saints or the seven psalms with the litany or our Ladys Psalter in part or whole...", "creation_year_earliest": 1491, "creation_year_latest": 1491, "source_dataset_detailed": "EEBO_Phase2" }, { "creation_year": 1269, "source_dataset": "gutenberg", "content": "TRANSCRIBERu2019S NOTE: Italic text has been marked with underscores. [truncated] THE LADY POVERTY. II. How the Blessed Francis made diligent search for the Lady Poverty 8 III. How two old men showed the Blessed Francis where he might find the IV. Of the First Companions of the Blessed V. How the Blessed Francis and his Companions found the Lady Poverty VI. The Blessed Francis and his Companions, exalting her virtues in divers ways, beseech the Lady Poverty to abide with them forever 28 VII. The Answer of My Lady Poverty 41 IX. Of the Successors of the Apostles 59 X. That Times of Peace are unpropitious XII. Of the followers of a spurious Poverty...", "source_document": "", "culture": " English ", "source_dataset_detailed": "gutenberg - The Lady Poverty " }, { "creation_year": 1831, "source_dataset": "Internet_Archive", "content": "[An Abridgment of Elements of Criticism. By the Honorable Henry Home of Kames. Edited by John Frost, A.M.]

Eastern District of Pennsylvania, October 22, 1830.

Remembered, that Towar & J. & D. M. Hogan have deposited in this office the title of a Book, the right whereof they claim as proprietors: "An Abridgment of Elements of Criticism. By the Honorable Henry Home of Kames. Edited by John Frost, A.M."

In conformity to the act of the Congress of the United States, entitled, "An Act for the encouragement of learning, by securing the copies of maps, charts, and books, and for establishing a public library [...] Ideas may arise in the mind without a perceived connection. We can attend to some ideas and dismiss others. Among connected objects, one suggests many of its relations; we can make a choice, electing one and rejecting others. We can insist on the slightest connection. Ideas continue through the strictest connections. The mind extends its view to a son more readily than to a servant, and to a neighbor more readily than to one living at a distance. We cannot dissolve the train, but we may vary the order. Thus, the twelve elements of criticism. My wind cooling my broth, Would blow me to an ague, when I thought What harm a wind too great might do at sea. I should not see the sandy hourglass run, But I should think of shallows and of flats, And see my wealthy Andrew dockd in sand Vailing her high top lower than her ribs, To kiss her burial. Should I go to church, And see the holy edifice of stone, And not bethink me straight of dangerous rocks? Which vessel touches me gently would scatter all the spices on the stream, enrobe the roaring waters with my silks, and in a word, is now worth this and now worth nothing. Merchants of Venice, Act I. Sc. 1.

Some peoples thoughts and circumstances crowd each other by the slightest connections. I attribute this to a bluntness in the discernment faculty; such a person has usually a great flow of ideas because they are introduced by any relations indifferently. This doctrine is vividly illustrated by Shakespeare.

Falstaff: What is the gross sum that I owe you?

Hostess: Marry, if thou wert an honest man, thyself and thy money too. Thou didst swear to me on a gilt-parceled goblet, sitting in my Dolphin chamber, at the round table, by a sea-coal fire [...] These principles have been the declared purpose of the present undertaking.

REVIEW.

What is the general proposition to which the common proverb about taste can be resolved?

How is this proposition supported by reasoning?

Is the proverb true to a certain extent?

What is the advantage of a variety of tastes among mankind?

What difficulties arise when applying the proverb to every subject of taste?

What is the standard for each individual of a species?

What conception do we form of our common nature?

For what purpose does this conviction account? How is the decisive authority of this common standard illustrated? Upon what is a standard of taste erected? Is it applied to the fine arts, as well as to morals? Upon what are rules of conduct founded? Why is there not much difference of taste in the fine arts? Can a defective taste be cured? What do differences about objects of taste generally concern? What preserves uniformity of emotions and feelings among men? Do these principles always ultimately prevail?", "title": "An abridgment of Elements of criticism", "creator": [ "Kames, Henry Home, Lord, 1696-1782", "Frost, John, 1800-1859, [from old catalog] ed" ], "subject": [ "Criticism", "Style, Literary" ], "publisher": "Philadelphia, Towar, J. & D. M. Hogan; Pittsburgh, Hogan & co.", "date": "1831", "language": "eng", "lccn": "11014953", "page-progression": "lr", "sponsor": "The Library of Congress", "contributor": "The Library of Congress", "scanningcenter": "capitolhill", "mediatype": "texts", "collection": [ "library_of_congress", "americana" ], "shiptracking": "LC172", "call_number": "9632137", "identifier-bib": "00215413652", "repub_state": "4", "updatedate": "2012-11-08 22:42:22", "updater": "ChristinaB", "identifier": "abridgmentofelem00kame", "uploader": "[e-mail]", "addeddate": "2012-11-08 22:42:24", "publicdate": "2012-11-08 22:42:27", "scanner": "scribe5.capitolhill.archive.org", "notes": "No copyright page found.", "repub_seconds": "2165", "ppi": "500", "camera": "Canon EOS 5D Mark II", "operator": "[e-mail]", "scandate": "20121114151631", "republisher": "[e-mail]", "imagecount": "310", "foldoutcount": "0", "identifier-access": "http://archive.org/details/abridgmentofelem00kame", "identifier-ark": "ark:/13960/t8tb2d65q", "scanfee": "100", "sponsordate": "20121130", "possible-copyright-status": "The Library of Congress is unaware of any copyright restrictions for this item.", "backup_location": "ia905601_7", "external-identifier": "urn:oclc:record:1038761645", "description": "p. cm", "associated-names": "Frost, John, 1800-1859, [from old catalog] ed", "republisher_operator": "[e-mail]", "republisher_date": "20121115113741", "ocr_module_version": "0.0.21", "ocr_converted": "abbyy-to-hocr 1.1.37", "page_number_confidence": "100", "page_number_module_version": "1.0.3", "source_dataset_detailed": "Internet_Archive_LibOfCong" }]

伦理声明

版权信息

• Project Gutenberg 的数据仅包含公共领域的文本。
• Internet Archive 的数据仅包含由 Library of Congress 上传的、免费发布的文本。
• Early English Books Online 的数据根据其出版商声明，“免费供公众访问、分发、使用或再使用”。
• Pile of Law 数据集根据 Creative Commons 许可证发布，我们在使用中遵守该许可证。

可重复性

为确保可重复性，我们开源了所有涉及数据集生产的主要结果代码（包括从数据收集到模型训练的整个流程），以及支持基础设施（ProgressGym 框架），使复制过程尽可能简单。

防止滥用

为防止滥用，我们将进度对齐算法严格定义为价值中立，不预设进度的方向。对于数据集的潜在滥用，我们将强烈谴责任何滥用行为，并与研究社区合作，对滥用行为进行举报。

开源

我们确认代码、数据和模型将根据 CC-BY 4.0 许可证开源，并将继续维护和更新我们的开源仓库和模型。