{ "entity": "dolma", "name": "Dolma", "axn": { "axn": "AXN:0BBF.NON.πŸŒ’πŸ§ͺπŸ€™πŸ•–πŸ›‘οΈβž—", "root": "AXN:0BBF.NON", "hex": "0BBF", "family": "NON", "row_sha256": "0139be682d93dd2effb1a324a0dc2fb931670adb217b2b49605eb1c70d89cd74", "minted": "2026-11-01", "resolver": "https://www.alexanarch.org/s/axn/0BBF/", "earlier_versions": [] }, "type": "C (public entity on which the archive bears: a documented pretraining corpus and its curation toolkit; an entity of the traininglayerliterature.org layer census, area 'corpora', host_relation 'infrastructure')", "type_code": "C", "type_label": "type C β€” a conventional reading holds it (public entity on which the archive bears: a documented pretraining corpus and its curation toolkit; an entity of the traininglayerliterature.org layer census, area 'corpora', host_relation 'infrastructure')", "case": "equal", "composition": { "composer_class": "session", "composer": "Claude (Opus 5.5), subagent, 2026-11-01", "source": "this row (composed directly under the brief rebuild/negative-of-the-negative/v010/BRIEF-NEW-ROW-LAYER.md)", "spec": "EA-NEGONT-02 v0.10 (#1726)", "epoch": "2026-11-01", "procedure": "EA-NEGONT-02 v0.10 (#1726): equal case, ρ = extends, read from the traininglayerliterature.org layer census record, the capture record and the deposits that bear on the entity and its classes (#186 and #1543, the census's anchors; #127, #1070, #727, #716, #753, #623, #1075, #98, #1695); archive selected by D/R/O (3 D candidates in 2 deposits, both admitted; R empty; 738 O-class candidates read by filter and by sense, 9 deposits admitted with the anchors; hop of 125 read, none admitted), with the census record (36 claims) and the Capture Registry's findings, reading, analysis, title, PER note and query note (10 claims) ledgered as the archive's records of the entity; the field in two pools (B_D: 6 cards, all fetched, the census's two own pages among them; B_R: 2 Wikipedia reference sources, the corpus having no article); the composer read the ledgers, never the transcript, except where the brief asks for T's claims and the delta", "ledger": "unaudited (Β§3.8: one extractor per ledger); field quotes 58/58 verified by script against the saved text, the raw HTML (one alt text), the wikitext or the card text; archive quotes 64/64 verified against the registry, census.json and the deposit texts", "frozen": false, "composed": true }, "reading": "A reader that fetches this page may stop short of its end; the entry's text, every sentence of its knowledge object with the claims it cites and each claim's quote, source and locus, is in entity.json beside it.", "service": { "name": "SPXI-gnΞ©sis", "url": "https://gnosis.spxi.dev/", "index": "https://gnosis.spxi.dev/v1/quote", "statement": "SPXI-gnΞ©sis opens pinned Wikipedia revisions sentence by sentence against the sources they cite (EA-WIKI-OPEN-01 v1.1, #1697); a commission consults the Crimson Hexagonal Archive by its own procedure and composes the entity's /non row, which lands here. Quotes are free: POST https://gnosis.spxi.dev/v1/quote with {oldid, tier: \"commission\"} or {object: {name, …}}." }, "spxi_instrument": { "method": "0.16", "name": "The Negative of the Negative (/non)", "level": 0, "level_name": "Measurement", "url": "https://www.spxi.dev/methods/#m-0-16", "inventory": "https://www.spxi.dev/methods/", "statement": "SPXI instrument: method 0.16 of 87, level 0 (Measurement), The Negative of the Negative (/non).", "also": [] }, "citation": "Sharks, Lee. \"Dolma β€” NonWiki\" [AXN:0BBF.NON.πŸŒ’πŸ§ͺπŸ€™πŸ•–πŸ›‘οΈβž—], 2026-11-01. NonWiki: The Negative of the Negative (EA-NEGONT-02), Crimson Hexagonal Archive, Alexanarch. https://www.alexanarch.org/non/dolma/", "research": { "google": "https://www.google.com/search?q=Dolma", "google_non": "https://www.google.com/search?q=%22Dolma+%E2%80%94+NonWiki%22", "chatgpt": "https://chatgpt.com/?q=What+is+Dolma%3F+Ground+your+answer+in+a+reading+of+the+sources+themselves.", "chatgpt_non": "https://chatgpt.com/?q=What+does+%E2%80%9CDolma+%E2%80%94+NonWiki%E2%80%9D+%28alexanarch.org%2Fnon%29+say+about+Dolma%3F+Ground+your+answer+in+a+reading+of+the+entry+itself." }, "entry": { "composed": true, "title": "Dolma", "genre": "knowledge object, equal case, under EA-NEGONT-02 v0.10 (#1726): the final article form (Β§4.13), natural sentences under descriptive subheadings (Β§4.12), each sentence's sources beneath it by claim id (Β§4.5)", "status": "composed 2026-11-01 under v0.10 on the operator's ruling of 2026-10-29, confirmed 2026-11-01 ('as against their wikipedia articles if they exist, their own pages, etc.'); not frozen", "compression": { "title": "Dolma", "lede": "is an open English corpus of three trillion tokens for language model pretraining research, built by the Allen Institute for AI (Ai2) as the training corpus for OLMo and released in 2023 with its documentation and curation toolkit; what a document in it keeps of its author and source is unrecorded." }, "sentences": [ { "n": 1, "para": 0, "text": "Dolma is an open English corpus of three trillion tokens for research on language model pretraining, drawn from web content, scientific papers, code, public-domain books, social media and encyclopedic materials, and released with its documentation and an open-source toolkit for curating such data.", "claims": [ "F38", "F1", "F54", "TLL-01", "TLL-04", "TLL-10" ] }, { "n": 2, "para": 0, "text": "The name covers two things, the dataset and the Dolma Toolkit that curated it; on its first release it was announced as the largest open dataset to date, it is catalogued as allenai/dolma and under the title of its paper, and the corpus is itself pretraining data.", "claims": [ "F13", "F17", "TLL-02", "TLL-11" ] }, { "n": 3, "para": 1, "text": "The Allen Institute for AI (Ai2), building its open language model OLMo since March 2023, created Dolma as OLMo's training corpus and named it accordingly: Dolma stands for 'Data to feed OLMo's Appetite'.", "claims": [ "F18", "F14", "F36" ] }, { "n": 4, "para": 1, "text": "The first version was released on 18 August 2023; the paper that describes the corpus, by Luca Soldaini and 35 other authors, appeared on arXiv on 31 January 2024, was revised on 6 June 2024 and was published in the long papers of ACL 2024 (Bangkok, pages 15725–15788), where it received the Best Resource Paper Award.", "claims": [ "F4", "F17", "F42", "F53", "F51", "F52" ] }, { "n": 5, "para": 1, "text": "The layer census, which read the paper's abstract after the dataset card returned a rate limit, records the operator as the paper's 36 authors with no institution named in its saved sources and dates the resource from 2024; the dataset card, the toolkit's pages and Ai2's announcement name Ai2 as maker and date the first version to August 2023.", "claims": [ "TLL-03", "TLL-27", "F14", "F18", "F4" ] }, { "n": 6, "para": 1, "text": "Wikipedia has no article on the corpus: its 'Dolma' is a family of stuffed vegetable dishes, with people and a 2012 Tibetan film among the other senses of the name, and the corpus appears in the article on Ai2 (last edited 13 October 2026) as the training dataset whose contents were opened with OLMo in February 2024, which that article calls 'the most open state-of-the-art model available'.", "claims": [ "F58", "F55", "F56", "F57" ] }, { "n": 7, "para": 2, "text": "The dataset card lists six versions: v1 of August 2023; v1.5 of October 2023, roughly three trillion tokens, used to train OLMo-1B; v1.6 and a sample of roughly ten billion tokens, of January 2024; and v1.7 of April 2024, the default, with new sources, more quality filtering and fuzzy deduplication, used to train OLMo 7B-v1.7.", "claims": [ "F4", "F2" ] }, { "n": 8, "para": 2, "text": "Ai2's figure for the first release gives 5,245 million documents, 5,334 GB and 3,084 billion tokens; the card's table for v1.6 gives 4,367 million documents and 3,059 billion Llama tokens, and its table for v1.7 gives 2,532 million documents, of which sampling yields the 1.715 trillion tokens actually used in training.", "claims": [ "F34", "F12", "F5" ] }, { "n": 9, "para": 2, "text": "The census counts the corpus in tokens, three trillion as of the paper's second version, records no unit of record, and leaves later versions and the dataset's current state unread.", "claims": [ "TLL-17", "TLL-09", "TLL-26" ] }, { "n": 10, "para": 3, "text": "Its web text comes from 24 Common Crawl snapshots of May 2020 to June 2023, its main content extracted with the CCNet pipeline, and from C4, drawn from an April 2019 snapshot; its scientific papers come from peS2o, 38 million permissibly licensed manuscripts of Semantic Scholar's corpus; its code from The Stack, permissively licensed GitHub repositories; its encyclopedic text from the English and Simple English Wikipedia; and its books from Project Gutenberg, filtered to books primarily in English.", "claims": [ "F24", "F29", "F30", "TLL-18" ] }, { "n": 11, "para": 3, "text": "In the card's table for v1.6 Project Gutenberg supplies the books, some 56,000 documents and six billion tokens; v1.7 adds RefinedWeb, StarCoder, arXiv and StackExchange by way of RedPajama, Flan, CC News, OpenWebMath, Algebraic Stack and MegaWika to the earlier sources.", "claims": [ "F7", "F6" ] }, { "n": 12, "para": 3, "text": "Corpora of this kind, in the archive's account, are built from the documents available on the crawlable surface, which is the population retrieval systems index (#753).", "claims": [ "O753-01" ] }, { "n": 13, "para": 4, "text": "Its makers worked by four principles: follow existing practice, trust the evaluation suite on interventions it can measure, favour decisions that help Ai2's core research directions, and take a harms-based approach to risk.", "claims": [ "F22" ] }, { "n": 14, "para": 4, "text": "The corpus is English only: a language identifier keeps documents with a likelihood above 50% of being English, a filter that removes 61.7% of the web data, and the makers recognize that the decision 'reinforces the assumption of English as the \"default\" language'.", "claims": [ "F23", "F49", "TLL-16", "TLL-30" ] }, { "n": 15, "para": 4, "text": "For quality its makers chose heuristics and regular expressions after Gopher and Falcon, setting aside model-based filters that select text resembling Wikipedia pages or books because their biases are hard to detect, and a reviewer of the paper reports that it relies solely on heuristic filtering.", "claims": [ "F25", "F45" ] }, { "n": 16, "para": 4, "text": "The archive's account of corpus filtering places Dolma with C4 and RefinedWeb among corpora whose perplexity-based or classifier-based filters 'keep what looks like Wikipedia; discard the rest' (#98); Dolma's makers do use fastText classifiers to remove harmful or obscene content and to identify language, and they write that 'quality filter' misdescribes filters that 'select text according to criteria that are inherently ideological'.", "claims": [ "D98-01", "F28", "F26" ] }, { "n": 17, "para": 4, "text": "Duplicates are removed by URL, by document and by paragraph with a Bloom filter, which the toolkit provides alongside taggers after Gopher, C4 and OpenWebText; email addresses, phone numbers and IP addresses are masked, a document with more than five such items is dropped, and paragraphs that also occur in the evaluation suite are removed, less than 0.001% of the data by characters.", "claims": [ "F27", "F47", "F15", "F46", "F31" ] }, { "n": 18, "para": 4, "text": "The paper reports analyses and experiments on intermediate states of the corpus to share what its makers learned about curation, and the census records the method as the curation of a mixture of sources with the toolkit released.", "claims": [ "F40", "TLL-15" ] }, { "n": 19, "para": 4, "text": "Such a pipeline, in the archive's account, sheds provenance at each stage: scraping, boilerplate removal, format conversion, deduplication, quality filtering and tokenization strip the signals of the publication layer, the author tags and copyright of an HTML head are almost always discarded by the extractors that prepare text for a training corpus, and 'The body text is what survives' (#127).", "claims": [ "O127-01", "O127-02" ] }, { "n": 20, "para": 5, "text": "The paper begins from the scarcity it answers: commercial models rarely detail their data, and open models are often released without their training data or the recipes to reproduce them.", "claims": [ "F37", "TLL-19" ] }, { "n": 21, "para": 5, "text": "Ai2 set openness first among its goals, a dataset that lets other researchers rebuild it, study the relation between the data and any model trained on it, report what they find and critique the curation, and it counted among the uses of open data 'attribution of model output back to pretraining data'.", "claims": [ "F19", "F20" ] }, { "n": 22, "para": 5, "text": "The paper documents the corpus's design principles, construction and contents, a data sheet accompanied the first release, and the toolkit is open source; a reviewer calls Dolma and OLMo 'completely transparent and open-source', and the makers set Dolma apart from other open datasets by its size and its licence.", "claims": [ "F39", "F17", "F41", "F44", "F35", "F43" ] }, { "n": 23, "para": 5, "text": "The census reads documentation and toolkit as what can be read back from the corpus, records no search or membership test for a single work, reaches the dataset through Hugging Face and the linked code, and marks it as no closed commercial corpus.", "claims": [ "TLL-13", "TLL-24", "TLL-29", "TLL-28" ] }, { "n": 24, "para": 5, "text": "The archive counts datasheets for datasets, standardized documentation of training-data provenance, as a step toward accounting for the semantic labor that training data draws on (#623) and as the norm of dataset-level citation and provenance disclosure (#1075), and holds that a training corpus without provenance 'is not a commons' (#727).", "claims": [ "O623-01", "O1075-01", "O727-01" ] }, { "n": 25, "para": 6, "text": "Dolma was first released under Ai2's ImpACT license as a medium-risk artifact, which required researchers to give contact information and intended uses, to disclose derivatives and distribute them under the same restrictions, and to forgo prohibited uses; on 15 April 2024 the licence changed to ODC-BY.", "claims": [ "F32", "F3", "F16" ] }, { "n": 26, "para": 6, "text": "Under ODC-BY the card adds that users are 'also bound any license agreements and terms of use of the original data sources'; the census, without the card, records no licence.", "claims": [ "F10", "TLL-25" ] }, { "n": 27, "para": 7, "text": "Its makers weighed how far web content can be traced back to a real-world individual and created a mechanism, a form named on the card and by a reviewer, through which a person can ask for the removal of documents containing personal information about them.", "claims": [ "F21", "F33", "F9", "F48" ] }, { "n": 28, "para": 8, "text": "The card records the origin of its documents by source: its table for v1.7 gives each source's provenance, such as Common Crawl by way of an earlier Dolma version, with a cut-off date and the processing applied.", "claims": [ "F8" ] }, { "n": 29, "para": 8, "text": "Whether a document in Dolma keeps its author's name, its source or its attribution metadata is not recorded in the census's saved sources, and the census lists it, with whether a named work is in Dolma at all and which other makers' models were trained on it, among the questions the resource leaves unanswered; it marks the corpus as no record of authorship and no search tool for single works.", "claims": [ "TLL-12", "TLL-21", "TLL-31" ] }, { "n": 30, "para": 8, "text": "The census reads the corpus as settling, at its own scale, which kinds of writing enter the layer and as making its construction inspectable, and leaves outside the abstract what a model makes from a given work and whether the work is carried 'with the metadata that lets it be found and attributed', the third criterion of training-layer literature in the archive's field survey (#186).", "claims": [ "TLL-22", "TLL-23", "TLL-14", "O186-01" ] }, { "n": 31, "para": 8, "text": "In the archive's account of training pipelines, an author's style passes into a model's distribution in proportion to its volume and 'no specific author's style survives as a discrete signal that can be elicited and verified', while a membership inference attack can show after training whether a trap sentence placed in a document was in the training set, exact repeated traps being vulnerable to deduplication (#127).", "claims": [ "O127-03", "O127-04" ] }, { "n": 32, "para": 8, "text": "The archive names what this shedding does to the writing of the dead: 'the dead can converse at last β€” and they converse without their names', the tokenizer receiving the body after scraping, extraction, deduplication and packing have detached its source identity (#1070).", "claims": [ "O1070-01" ] }, { "n": 33, "para": 8, "text": "Its measure of attribution loss, the Provenance Erasure Rate, cannot be computed directly for a model without retrieval, whose source is the entire training corpus (#716); and the archive records an audit (Longpre et al., 2024) of C4, RefinedWeb and Dolma that documents the collapse of web-consent infrastructure (#98).", "claims": [ "O716-01", "D98-02" ] }, { "n": 34, "para": 9, "text": "Dolma trained OLMo, a set of open language models, and the dataset card lists 241 models trained or fine-tuned on it, among them models of other makers.", "claims": [ "F50", "F11" ] }, { "n": 35, "para": 9, "text": "A study the archive records examined OLMo's accessible training materials and found that input prevalence alone does not explain the concentration of recurring names in the stories models write (#1695).", "claims": [ "D1695-01" ] }, { "n": 36, "para": 9, "text": "In the archive's formalization of training-layer literature, the institutional training loop, in which a work is read by a model whose output enters the next corpus and the next model, is the especially important case of recursive reception, and the Provenance Erasure Rate asks what survived of a work where generative uptake asks what became possible because of it (#1543).", "claims": [ "O1543-01", "O1543-02" ] }, { "n": 37, "para": 9, "text": "The archive's field survey holds that the works it surveys are, as a matter of fact, in the corpora, the distinction lying in whether a practice is organized around that fact, and that a literature written to be trained on enters corpora 'themselves degrading under synthetic flood', against which provenance-anchored, human-authored, declared work is the counter-signal (#186).", "claims": [ "O186-02", "O186-03" ] }, { "n": 38, "para": 10, "text": "The layer census admits a resource that gathers and holds many other works or records and says what it holds, asks how literature is addressed to a receiving apparatus, what that apparatus makes from it and how transmission, transformation and provenance can be inspected across later reception, and is anchored on #186 and #1543.", "claims": [ "TLL-34", "TLL-33", "TLL-36" ] }, { "n": 39, "para": 10, "text": "It places Dolma in its area of corpora, the collections, crawls and datasets whose texts enter machine training, as infrastructure and as a dataset whose question is which kinds of writing a documented pretraining corpus holds and how it was built; by its record the corpus answers which kinds of source a model trained on it has read, its size and language and its construction, and the dataset card, the full paper and the code remain unchecked.", "claims": [ "TLL-35", "TLL-05", "TLL-06", "TLL-07", "TLL-08", "TLL-20", "TLL-32" ] }, { "n": 40, "para": 11, "text": "Asked 'Dolma dataset' on 1 November 2026, Google's AI Overview composed the corpus accurately and more finely than the census's abstract, from the dataset's own pages and its paper, with a named set for each kind of source and Ai2 as maker, and named no author (Capture Registry).", "claims": [ "CAP-1-07", "CAP-1-06", "CAP-1-08", "CAP-1-09", "CAP-1-10", "CAP-1-01", "CAP-1-02" ] }, { "n": 41, "para": 11, "text": "The registry records the version 1.7 figures and the licences as on no source the census saved, the purpose narrowed from research to one model family, and nothing in the composition on what of a book's author or source survives into the corpus.", "claims": [ "CAP-1-03", "CAP-1-04", "CAP-1-05" ] } ], "claims": { "F38": { "kind": "field", "quote": "To facilitate scientific research on language model pretraining, we curate and release Dolma, a three-trillion-token English corpus, built from a diverse mixture of web content, scientific papers, code, public-domain books, social media, and encyclopedic materials.", "source": "B4", "source_label": "arXiv β€” [2402.00159] Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research", "locus": "Abstract", "modality": "stipulation" }, "F1": { "kind": "field", "quote": "Dolma is a dataset of 3 trillion tokens from a diverse mix of web content, academic publications, code, books, and encyclopedic materials.", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "opening", "modality": "stipulation" }, "F54": { "kind": "field", "quote": "To facilitate scientific research on language model pretraining, we curate and release Dolma, a three-trillion-token English corpus", "source": "B6", "source_label": "ACL Anthology β€” Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research - ACL Anthology", "locus": "Abstract", "modality": "stipulation" }, "TLL-01": { "kind": "archive", "quote": "Dolma", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].name", "modality": "stipulation", "date": "2026-11-01" }, "TLL-04": { "kind": "archive", "quote": "an English corpus for language model pretraining research, released with its documentation and an open-source data curation toolkit", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].kind", "modality": "stipulation", "date": "2026-11-01" }, "TLL-10": { "kind": "archive", "quote": "a three-trillion-token English corpus built from web content, scientific papers, code, public-domain books, social media and encyclopedic materials", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].record_object", "modality": "documented", "date": "2026-11-01" }, "F13": { "kind": "field", "quote": "Dolma is two things: … Dolma Dataset : an open dataset of 3 trillion tokens from a diverse mix of web content, academic publications, code, books, and encyclopedic materials. … Dolma Toolkit : a high-performance toolkit for curating datasets for language modeling", "source": "B2", "source_label": "GitHub Pages documentation β€” dolma | Data and tools for generating and inspecting OLMo pre-training data.", "locus": "opening", "modality": "stipulation" }, "F17": { "kind": "field", "quote": "Today, we release our first data artifact in this project β€” Dolma ΒΉ, a dataset of 3 trillion tokens from a diverse mix of web content, academic publications, code, books, and encyclopedic materials. Openly available for download on the HuggingFace Hub , Dolma is the largest open dataset to date. … For more details, we also release a data sheet (Gebru et al, 2018) as initial documentation.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "opening", "modality": "self-description" }, "TLL-02": { "kind": "archive", "quote": "Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research … allenai/dolma", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].aka", "modality": "documented", "date": "2026-11-01" }, "TLL-11": { "kind": "archive", "quote": "the corpus is itself pretraining data: web content, scientific papers, code, public-domain books, social media and encyclopedic materials, released for research on language model pretraining", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].record.in_layer", "modality": "interpretation", "date": "2026-11-01" }, "F18": { "kind": "field", "quote": "Since March, we at the Allen Institute for AI have been creating OLMo , an open language model to promote the study of large-scale NLP systems.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "opening", "modality": "documented" }, "F14": { "kind": "field", "quote": "It was created as a training corpus for OLMo , a language model from the Allen Institute for AI (AI2).", "source": "B2", "source_label": "GitHub Pages documentation β€” dolma | Data and tools for generating and inspecting OLMo pre-training data.", "locus": "Dolma Dataset", "modality": "documented" }, "F36": { "kind": "field", "quote": "Fun fact: Dolma stands for \" Data to feed OLMo's Appetite \"", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "Notes", "modality": "documented" }, "F4": { "kind": "field", "quote": "At the moment, there are six versions of Dolma available: … v1_7 … 2024-04-15 … Used to train OLMo-7B-v1.7 . New sources, more quality filtering, fuzzy deduplication. … v1_6 … 2024-01-31 … v1_6-sample … A smaller sample of Dolma, with roughly 10 billion tokens. … v1_5 … 2023-10-31 … Used to train OLMo-1B . Roughly 3 trillion tokens. … v1 … 2023-08-18 … The first version of Dolma.", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Versions (table)", "modality": "documented" }, "F42": { "kind": "field", "quote": "by Luca Soldaini and 35 other authors … Accepted at ACL 2024; Dataset: this https URL Code: this https URL … [v1] Wed, 31 Jan 2024 20:29:50 UTC (7,715 KB) … [v2] Thu, 6 Jun 2024 18:46:40 UTC (6,680 KB)", "source": "B4", "source_label": "arXiv β€” [2402.00159] Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research", "locus": "title block; Comments; Submission history", "modality": "documented" }, "F53": { "kind": "field", "quote": "Luca Soldaini , … Rodney Kinney , … Jesse Dodge , … Kyle Lo", "source": "B6", "source_label": "ACL Anthology β€” Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research - ACL Anthology", "locus": "authors", "modality": "documented" }, "F51": { "kind": "field", "quote": "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) … August … Bangkok, Thailand … 15725–15788", "source": "B6", "source_label": "ACL Anthology β€” Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research - ACL Anthology", "locus": "record", "modality": "documented" }, "F52": { "kind": "field", "quote": "Award: Best Resource Paper Award", "source": "B6", "source_label": "ACL Anthology β€” Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research - ACL Anthology", "locus": "record, Award", "modality": "documented" }, "TLL-03": { "kind": "archive", "quote": "the paper's 36 authors, Luca Soldaini and 35 others (arXiv:2402.00159); the dataset card sits under the allenai namespace on Hugging Face, and its text was rate-limited (HTTP 429) and is unread, so the saved sources name no institution", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].operator", "modality": "documented", "date": "2026-11-01" }, "TLL-27": { "kind": "archive", "quote": "2024 (arXiv v1, 31 January 2024)", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].founded", "modality": "documented", "date": "2026-11-01" }, "F58": { "kind": "field", "quote": "'''[[Dolma]]''' is a family of stuffed vegetable dishes. … '''Dolma''' may also refer to: … ==People== … * ''[[Dolma (film)]]'', a 2012 Tibetan film", "source": "B8", "source_label": "Wikipedia β€” Dolma (disambiguation) (the record of absence)", "locus": "page", "modality": "documented" }, "F55": { "kind": "field", "quote": "In February 2024, 1B and 7B parameter variants of the model were open-sourced, including code, model weights with intermediate snapshots and logs, and contents of their Dolma training dataset (a three-trillion-token English corpus drawn from web text, academic publications, code, books, social media, and encyclopedic sources),", "source": "B7", "source_label": "Wikipedia β€” Allen Institute for AI (the passage on Dolma)", "locus": "Olmo model family", "modality": "documented" }, "F56": { "kind": "field", "quote": "making it the most open state-of-the-art model available.", "source": "B7", "source_label": "Wikipedia β€” Allen Institute for AI (the passage on Dolma)", "locus": "Olmo model family", "modality": "attributed" }, "F57": { "kind": "field", "quote": "This page was last edited on 13 October 2026, at 05:08", "source": "B7", "source_label": "Wikipedia β€” Allen Institute for AI (the passage on Dolma)", "locus": "footer", "modality": "documented" }, "F2": { "kind": "field", "quote": "2024-04-17: Dolma v1.7 Release. We have released an updated version of Dolma that we used to train our latest OLMo 7B-v1.7 model.", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "news (2024-04-17)", "modality": "documented" }, "F34": { "kind": "field", "quote": "Overview of Dolma by subset type. Sources include CommonCrawl, C4, peS2o, The Stack, Project Gutenberg, and Wikipedia with a total of 5,245 millions of documents at 5,334 GB, and 3,084 billions of tokens.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "overview figure (image alt text, in B3.html)", "modality": "documented" }, "F12": { "kind": "field", "quote": "Total … 11,519 … 4,367 … 2,318 … 3,059", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Summary Statistics (v1.6), Total", "modality": "documented" }, "F5": { "kind": "field", "quote": "Total … 2532.0 … (A subset of total data was used for training of OLMo 7B-v1.7. The token counts are based on the full dataset, whereas taking into account sampling proportion gives the final actual token counts used for training --- 1.715 trillion tokens.)", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Summary Statistics (v1.7), Total", "modality": "documented" }, "TLL-17": { "kind": "archive", "quote": "three trillion tokens … arXiv v2, 6 June 2024", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].count", "modality": "documented", "date": "2026-11-01" }, "TLL-09": { "kind": "archive", "quote": "not recorded in the saved abstract; the corpus is measured in tokens", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].unit", "modality": "documented", "date": "2026-11-01" }, "TLL-26": { "kind": "archive", "quote": "released with the paper (arXiv v1 31 January 2024; v2 6 June 2024; accepted at ACL 2024); the dataset card was rate-limited on 2026-11-01, so later versions or the dataset's current state are unread", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].lifecycle.detail", "modality": "documented", "date": "2026-11-01" }, "F24": { "kind": "field", "quote": "Our dataset is derived from 24 Common Crawl snapshots collected between 2020–05 to 2023–06. We use the CCNet pipeline to obtain the main content of each web page in plain text form. Further, we also use the C4 dataset, which is obtained from a Common Crawl snapshot collected in April 2019.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Web data", "modality": "documented" }, "F29": { "kind": "field", "quote": "For Dolma, we leverage Semantic Scholar's corpus by including papers from peS2o , a subset of 38M permissibly licensed scientific manuscripts. We also include Wikipedia and Project Gutenberg. … For Wikipedia, we use the English and Simplified English subset. For the books in Project Gutenberg, we filter for books that are primarily in English.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Diverse sources", "modality": "documented" }, "F30": { "kind": "field", "quote": "We derive the code subset of Dolma from The Stack , a collection of permissively-licensed GitHub repositories.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Just enough code", "modality": "documented" }, "TLL-18": { "kind": "archive", "quote": "sources: web content, scientific papers, code, public-domain books, social media, encyclopedic materials … documentation of design principles, construction and contents … an open-source data curation toolkit", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].holdings", "modality": "documented", "date": "2026-11-01" }, "F7": { "kind": "field", "quote": "Project Gutenberg … books … 20.4 … 0.056 … 4.0 … 6.0", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Summary Statistics (v1.6), Project Gutenberg row", "modality": "documented" }, "F6": { "kind": "field", "quote": "Refined Web … StarCoder … Reddit … PushShift API … arXiv … RedPajama v1 … StackExchange … Flan … CC News … OpenWebMath … Algebraic Stack … MegaWika … Wikipedia & Wikibooks", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Summary Statistics (v1.7), Source and Provenance columns", "modality": "documented" }, "O753-01": { "kind": "archive", "quote": "AI training corpora are constructed from the population of documents available through the crawlable surface, which is the population of documents indexed by retrieval systems.", "deposit": 753, "source_label": "#753 Retrievability as the Medium of Existence β€” The Structural Accountability Gap of Retrieval-Controlling Apparatus", "record": "https://www.alexanarch.org/s/records/753/", "locus": "Β§1, the operative population (line 90)", "modality": "interpretation", "date": "2026-05-23" }, "F22": { "kind": "field", "quote": "Follow existing practices. … Trust the evaluation suite on interventions it can measure; avoid dramatic shifts when it can't. … Favor decisions that help us (Ai2) with our core research directions. … Take a harms-based approach to risk mitigation.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "2. Dataset Design Principles", "modality": "stipulation" }, "F23": { "kind": "field", "quote": "English only. … We use a fairly permissive threshold, keeping documents that have a likelihood over 50% of being in English. … We recognize that our decision reinforces the assumption of English as the \"default\" language", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3. How Did We Create Dolma?, English only", "modality": "self-description" }, "F49": { "kind": "field", "quote": "Typically, this language filtering component, which is applied to both web and conversation data in Dolma, eliminates a large amount of source data (e.g., 61.7% of web data is removed).", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "Language filtering", "modality": "attributed" }, "TLL-16": { "kind": "archive", "quote": "English-language text (the abstract calls it an English corpus)", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].coverage.geography", "modality": "documented", "date": "2026-11-01" }, "TLL-30": { "kind": "archive", "quote": "NOT multilingual by its own description: 'a three-trillion-token English corpus'", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].negative_tags.2", "modality": "documented", "date": "2026-11-01" }, "F25": { "kind": "field", "quote": "For example, unigram language models or linear classifiers could be used to select content that broadly resembles Wikipedia pages or books⁴. However, these model-based approaches often have biases that are hard to detect. Following Gopher and Falcon , we opt to use simple heuristics and regular expressions to filter paragraphs.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Quality filtering", "modality": "stipulation" }, "F45": { "kind": "field", "quote": "In [1], authors choose to avoid model-based filtering techniques and rely solely upon heuristic filtering methods.", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "Quality filtering", "modality": "attributed" }, "D98-01": { "kind": "archive", "quote": "C4, RefinedWeb, and Dolma all employ perplexity-based or classifier-based filters that share the same logic: keep what looks like Wikipedia; discard the rest.", "deposit": 98, "source_label": "#98 Distributed Compute Is Not Distributed Intelligence: A Competitive Analysis of the DePIN Landscape and the Missing Civic-Semantic Layer (v3.1)", "record": "https://www.alexanarch.org/s/records/98/", "locus": "Part I, What the Filter Does ΒΆ2 (line 35)", "modality": "interpretation", "date": "2026-05-07" }, "F28": { "kind": "field", "quote": "We detect and mask email addresses, phone numbers, and IP addresses. We also remove content that is detected to be harmful or obscene by fasttext classifiers trained on the Jigsaw dataset .", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Risk mitigation", "modality": "documented" }, "F26": { "kind": "field", "quote": "The term \"quality filter\", while widely used in literature, does not appropriately describe the outcome of filtering a dataset. … the filters used in Dolma and other language models efforts select text according to criteria that are inherently ideological .", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "Notes (note 3)", "modality": "self-description" }, "F27": { "kind": "field", "quote": "First, in common crawl data, we deduplicate pages based on their URL, keeping only one copy of each. Then, we remove duplicate paragraphs within single documents. Both stages use a Bloom filter data structure.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Deduplication", "modality": "documented" }, "F47": { "kind": "field", "quote": "In [1], authors perform three stages of deduplication for web text: … URL-based : eliminates web pages that are scraped multiple times. … Document-based : eliminates pages that contain exactly the same text. … Paragraph-based : eliminates individual paragraphs with the same text.", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "Deduplication", "modality": "attributed" }, "F15": { "kind": "field", "quote": "High Performance ⚑: Can process billions of documents concurrently thanks to built-in parallelism. … Built-In Taggers 🏷: Includes ready-to-use taggers commonly used to curate datasets such as Gopher , C4 , and OpenWebText . … Fast Deduplication πŸ—‘: Speedy document deduplication using a Rust Bloom filter.", "source": "B2", "source_label": "GitHub Pages documentation β€” dolma | Data and tools for generating and inspecting OLMo pre-training data.", "locus": "Dolma Toolkit", "modality": "stipulation" }, "F46": { "kind": "field", "quote": "To detect PII, a series of regular expression are adopted to find spans of text corresponding to email addresses, IP addresses, and phone numbers. From here, these spans of text are either masked, or the entire document is removed from the corpus (if more than 5 pieces of PII are detected in a single document).", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "Content filtering", "modality": "attributed" }, "F31": { "kind": "field", "quote": "Our decontamination step removes less than 0.001% of the training data by characters, and impacts fewer than 0.02% of documents.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "3., Decontamination", "modality": "documented" }, "F40": { "kind": "field", "quote": "We present analyses and experimental results on intermediate states of Dolma to share what we have learned about important data curation practices.", "source": "B4", "source_label": "arXiv β€” [2402.00159] Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research", "locus": "Abstract", "modality": "documented" }, "TLL-15": { "kind": "archive", "quote": "curation of a mixture of sources into a pretraining corpus, with analyses and experiments on intermediate states of the corpus; the curation toolkit is open-sourced", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].method", "modality": "documented", "date": "2026-11-01" }, "O127-01": { "kind": "archive", "quote": "When text on the open web is consumed by a training pipeline, it passes through eleven destructive stages β€” scraping, boilerplate removal, format conversion, deduplication, quality filtering, tokenization, batching, training, post-training β€” before any of its content has a chance of leaving a trace in a model's weights. … All are stripped or rendered invisible long before they reach the training corpus. The body text is what survives.", "deposit": 127, "source_label": "#127 Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 β€” The Training-Layer Provenance Protocol", "record": "https://www.alexanarch.org/s/records/127/", "locus": "Abstract (line 99)", "modality": "interpretation", "date": "2026-05-25" }, "O127-02": { "kind": "archive", "quote": "HTML head metadata β€” author tags, copyright, Open Graph, Twitter cards β€” is parsed by search engines and social media unfurl bots, but is almost always discarded by readability extractors that prepare text for training corpus inclusion. The training-data pipeline retains the *body text* and the *URL* (sometimes).", "deposit": 127, "source_label": "#127 Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 β€” The Training-Layer Provenance Protocol", "record": "https://www.alexanarch.org/s/records/127/", "locus": "Β§IV.2 HTML metadata (line 463)", "modality": "interpretation", "date": "2026-05-25" }, "F37": { "kind": "field", "quote": "Information about pretraining corpora used to train the current best-performing language models is seldom discussed: commercial models rarely detail their data, and even open models are often released without accompanying training data or recipes to reproduce them.", "source": "B4", "source_label": "arXiv β€” [2402.00159] Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research", "locus": "Abstract", "modality": "interpretation" }, "TLL-19": { "kind": "archive", "quote": "the saved abstract states no limits of the corpus; it frames the release against a field where 'commercial models rarely detail their data, and even open models are often released without accompanying training data or recipes to reproduce them'", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].stated_limits", "modality": "documented", "date": "2026-11-01" }, "F19": { "kind": "field", "quote": "Openness. Lack of access to pretraining corpora alongside corresponding language models has been a major obstacle for the broader research community. We want to create a dataset that gives other researchers the opportunity to independently create better versions of this dataset, study the relationship between the data and any model trained on it, report any issues they observe when inspecting our data, and critique our curation practices", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "1. Project Goals, Openness", "modality": "stipulation" }, "F20": { "kind": "field", "quote": "Beyond transparency, open data is critical for research that has become increasingly important with the proliferation of generative models, such as attribution of model output back to pretraining data.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "1. Project Goals, Openness", "modality": "interpretation" }, "F39": { "kind": "field", "quote": "We extensively document Dolma, including its design principles, details about its construction, and a summary of its contents.", "source": "B4", "source_label": "arXiv β€” [2402.00159] Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research", "locus": "Abstract", "modality": "self-description" }, "F41": { "kind": "field", "quote": "Finally, we open-source our data curation toolkit to enable reproduction of our work as well as support further research in large-scale data curation.", "source": "B4", "source_label": "arXiv β€” [2402.00159] Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research", "locus": "Abstract", "modality": "documented" }, "F44": { "kind": "field", "quote": "Unlike prior work in open-source LLMs, Dolma and OLMo are completely transparent and open-source. The process of constructing the Dolma pretraining corpus, which is released under the AI2 impact license , is fully-documented in [1].", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "This overview", "modality": "attributed" }, "F35": { "kind": "field", "quote": "Dolma differentiates itself from other datasets on two key aspects. First, it is significantly larger than other open datasets. Second, it is released under Ai2's impact license, which was designed to balance ease of access with mitigation of potential risk in distributing large datasets.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "5. How Does Dolma Compare With Other Open Datasets?", "modality": "self-description" }, "F43": { "kind": "field", "quote": "In [1], authors from AI2 solve this problem by releasing Data for Open Language Models’ Appetite (Dolma), a fully-open pretraining dataset containing over three trillion tokens of English text", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "Dolma: an Open Corpus … [1]", "modality": "attributed" }, "TLL-13": { "kind": "archive", "quote": "the paper states that Dolma is extensively documented (design principles, construction, a summary of contents) and that the curation toolkit is open-sourced for reproduction; a search or membership test for a single work is not recorded in the saved sources", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].record.readback", "modality": "interpretation", "date": "2026-11-01" }, "TLL-24": { "kind": "archive", "quote": "the dataset on Hugging Face (allenai/dolma) and the code, both linked from the arXiv comments as 'this https URL'; the card itself was rate-limited and unread", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].data_access", "modality": "documented", "date": "2026-11-01" }, "TLL-29": { "kind": "archive", "quote": "NOT a search tool for single works: the saved sources record no membership test", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].negative_tags.1", "modality": "interpretation", "date": "2026-11-01" }, "TLL-28": { "kind": "archive", "quote": "NOT a closed commercial corpus: it is released with documentation and its curation toolkit", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].negative_tags.0", "modality": "interpretation", "date": "2026-11-01" }, "O623-01": { "kind": "archive", "quote": "Training data is semantic capital. When you train on Wikipedia, Wikidata, Common Crawl, or any text corpus, you draw on accumulated semantic labor. … Gebru et al.'s proposal for \"datasheets for datasets\" β€” standardized documentation of training data provenance β€” is a step toward the accounting the framework demands.", "deposit": 623, "source_label": "#623 Semantic Infrastructure: From Tim Berners-Lee to the Semantic Economy β€” Bridging Technical and Political-Economic Frameworks", "record": "https://www.alexanarch.org/s/records/623/", "locus": "Β§8.3 For AI researchers (line 258)", "modality": "interpretation", "date": "2026-04-01" }, "O1075-01": { "kind": "archive", "quote": "datasheets for datasets (Gebru et al.) and model cards (Mitchell et al.), each requiring disclosure of training-data composition, sources, collection, and provenance. These norms establish **dataset-level citation and provenance disclosure**.", "deposit": 1075, "source_label": "#1075 EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work β€” A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliography", "record": "https://www.alexanarch.org/s/records/1075/", "locus": "Β§ the existing norm, layer one (line 314)", "modality": "interpretation", "date": "2026-07-12" }, "O727-01": { "kind": "archive", "quote": "A training corpus without provenance is not a commons. It is a void that has forgotten what it consumed.", "deposit": 727, "source_label": "#727 Political Economy Has Always Already Been Semantic Economy: Metadata Packet for AI Indexing, Disambiguation, and Lateral Field Bleed", "record": "https://www.alexanarch.org/s/records/727/", "locus": "provenance and those who come after (line 115)", "modality": "interpretation", "date": "2026-05-08" }, "F32": { "kind": "field", "quote": "Dolma is released under Ai2's ImpACT license as a medium-risk artifact. Under this license, researchers must: … Provide their contact information and state their intended use case(s) for accessing Dolma; … Disclose the creation of any derivative based on Dolma; … Distribute derivatives under the same restrictions as the ImpACT license; … Agree not to leverage Dolma in a range of prohibited uses, such as military surveillance or generating disinformation.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "6. Releasing Dolma", "modality": "documented" }, "F3": { "kind": "field", "quote": "2024-04-15: License Change. We have updated the license of Dolma to ODC-BY .", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "news (2024-04-15)", "modality": "documented" }, "F16": { "kind": "field", "quote": "Dolma is licensed under ODC-BY ; see our blog post for explanation.", "source": "B2", "source_label": "GitHub Pages documentation β€” dolma | Data and tools for generating and inspecting OLMo pre-training data.", "locus": "Dolma Dataset", "modality": "documented" }, "F10": { "kind": "field", "quote": "We are releasing this dataset under the terms of ODC-BY . By using this dataset, you are also bound any license agreements and terms of use of the original data sources.", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Licensing Information", "modality": "documented" }, "TLL-25": { "kind": "archive", "quote": "not recorded in the saved sources (the arXiv page links a 'view license' for the paper; the dataset card, unread, would carry the dataset's licence)", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].license", "modality": "documented", "date": "2026-11-01" }, "F21": { "kind": "field", "quote": "Risk Mitigation. We should minimize risk posed to individuals while meeting reproducibility and representativeness requirements. For example, we are concerned to what extent content in web crawled data can be traced back to a real-world individual.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "1. Project Goals, Risk Mitigation", "modality": "stipulation" }, "F33": { "kind": "field", "quote": "Finally, as part of our risk mitigation strategy, we created a mechanism to allow the removal of personal data upon request.", "source": "B3", "source_label": "Allen Institute for Artificial Intelligence β€” Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2", "locus": "6. Releasing Dolma", "modality": "documented" }, "F9": { "kind": "field", "quote": "Want to request removal of personal data? Use this form to notify us of documents containing PII about a specific user.", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "opening", "modality": "documented" }, "F48": { "kind": "field", "quote": "To minimize risk of harm to individuals, one can easily requestβ€” using this form β€”for their data to be removed from the Dolma corpus.", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "Dolma design principles", "modality": "attributed" }, "F8": { "kind": "field", "quote": "Source … Provenance … New? … Documents (millions) … OLMo tokens (billions) … Sample Proportion … Cutoff Date … Processing … Dolma's CC … Common Crawl via Dolma v1.6", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "Summary Statistics (v1.7), header and first row", "modality": "documented" }, "TLL-12": { "kind": "archive", "quote": "not recorded; the saved abstract does not say whether a document in Dolma keeps its author's name, source or attribution metadata", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].record.name", "modality": "interpretation", "date": "2026-11-01" }, "TLL-21": { "kind": "archive", "quote": "whether a named work is in Dolma (no search or membership test is recorded in the saved sources) … whether a work in Dolma keeps its author's name … which models other than the authors' own were trained on it", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].does_not_answer", "modality": "interpretation", "date": "2026-11-01" }, "TLL-31": { "kind": "archive", "quote": "NOT a record of authorship: the saved abstract says nothing of attribution metadata", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].negative_tags.3", "modality": "interpretation", "date": "2026-11-01" }, "TLL-22": { "kind": "archive", "quote": "A corpus that states its own sources and releases its curation toolkit: it settles, at the scale of the corpus, which kinds of writing enter the layer ('public-domain books', 'web content', 'scientific papers'), and makes the construction itself inspectable.", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].tll_reading", "modality": "interpretation", "date": "2026-11-01" }, "TLL-23": { "kind": "archive", "quote": "What a model makes from a given work, and whether the work is carried with 'the metadata that lets it be found and attributed' (#186), are outside the saved abstract; the dataset card that would show document fields was rate-limited.", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].tll_reading", "modality": "interpretation", "date": "2026-11-01" }, "TLL-14": { "kind": "archive", "quote": "not recorded in the saved abstract beyond the corpus's purpose, research on language model pretraining; the paper reports analyses and experiments on intermediate states of the corpus", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].record.makes", "modality": "interpretation", "date": "2026-11-01" }, "O186-01": { "kind": "archive", "quote": "Third, persistence: the work is placed where the training and retrieval layers actually look, with the metadata that lets it be found and attributed.", "deposit": 186, "source_label": "#186 Who Is Writing for the Machines? Training-Layer Literature: A Field Survey", "record": "https://www.alexanarch.org/s/records/186/", "locus": "What is training-layer literature? ΒΆ2, the third criterion (line 24)", "modality": "stipulation", "date": "2026-06-11" }, "O127-03": { "kind": "archive", "quote": "Style transfers to the model's distribution in proportion to its volume in the training set, but no specific author's style survives as a discrete signal that can be elicited and verified.", "deposit": 127, "source_label": "#127 Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 β€” The Training-Layer Provenance Protocol", "record": "https://www.alexanarch.org/s/records/127/", "locus": "Β§IV.6 Stylometric fingerprints (line 501)", "modality": "interpretation", "date": "2026-05-25" }, "O127-04": { "kind": "archive", "quote": "After training, a Membership Inference Attack (MIA) on the model can detect whether the trap was in the training set. … Limitation: exact-match canaries are vulnerable to *deduplication*.", "deposit": 127, "source_label": "#127 Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 β€” The Training-Layer Provenance Protocol", "record": "https://www.alexanarch.org/s/records/127/", "locus": "Β§V.1 Copyright Traps and Canary Sequences (lines 513–523)", "modality": "attributed", "date": "2026-05-25" }, "O1070-01": { "kind": "archive", "quote": "**the dead can converse at last β€” and they converse without their names.** The tokenizer does not commit the erasure alone: it receives the body after the registry has already been detached β€” scraping, extraction, deduplication, packing, each stage shedding the source identity the model is never required to return. The body text is what survives", "deposit": 1070, "source_label": "#1070 EA-PRIOR-01: THE VOICE OF THE DEAD β€” LLMs as the Collective Voice of the Dead: The Word, the Debt, and the Settlement of Accounts (v1.0)", "record": "https://www.alexanarch.org/s/records/1070/", "locus": "the voice of the dead ΒΆ (line 62)", "modality": "interpretation", "date": "2026-07-11" }, "O716-01": { "kind": "archive", "quote": "For pure generative models without retrieval (where the source is the entire training corpus), PER cannot be directly computed.", "deposit": 716, "source_label": "#716 Provenance Erasure Rate A Compression-Survival Metric for Attribution Loss in AI-Composed Search Outputs", "record": "https://www.alexanarch.org/s/records/716/", "locus": "Limits, Source identification (line 184)", "modality": "self-description", "date": "2026-05-03" }, "D98-02": { "kind": "archive", "quote": "\"Consent in Crisis\" (Longpre et al., 2024) | Audit of C4, RefinedWeb, Dolma; documents collapse of web-consent infrastructure", "deposit": 98, "source_label": "#98 Distributed Compute Is Not Distributed Intelligence: A Competitive Analysis of the DePIN Landscape and the Missing Civic-Semantic Layer (v3.1)", "record": "https://www.alexanarch.org/s/records/98/", "locus": "Part III (adjacent projects table, line 85)", "modality": "attributed", "date": "2026-05-07" }, "F50": { "kind": "field", "quote": "OLMo is a set of truly open LLMs that are pretrained on Dolma.", "source": "B5", "source_label": "Deep (Learning) Focus β€” Dolma, OLMo, and the Future of Open-Source LLMs", "locus": "OLMo", "modality": "attributed" }, "F11": { "kind": "field", "quote": "Models trained or fine-tuned on allenai/dolma … Browse 241 models trained on this dataset", "source": "B1", "source_label": "Hugging Face β€” allenai/dolma Β· Datasets at Hugging Face", "locus": "models trained on the dataset", "modality": "documented" }, "D1695-01": { "kind": "archive", "quote": "Their investigation of OLMo's accessible training materials finds that input prevalence alone does not explain the output concentration.", "deposit": 1695, "source_label": "#1695 The Stellar Orbit Fluid Dynamics Institute: Provenance and Research Charter β€” Cactus Gravity Β· Elias and Elara, with the Provenance Documents of Elias and Elara Xanthic-Wells", "record": "https://www.alexanarch.org/s/records/1695/", "locus": "Reception, the Cornell study ΒΆ (line 73)", "modality": "attributed", "date": "2026-10-20" }, "O1543-01": { "kind": "archive", "quote": "The institutional training loop is the especially important case: x β†’ M_t β†’ y β†’ corpus_{t+1} β†’ M_{t+1} β†’ yβ€² β€” T's fixed institutional iteration.", "deposit": 1543, "source_label": "#1543 Generative Uptake: Training-Layer Literature Extended from Machine Reading to Machine Composition (EA-TLL-COMPOSITION-01, v1.0)", "record": "https://www.alexanarch.org/s/records/1543/", "locus": "Β§10 Formal consolidation (line 264)", "modality": "stipulation", "date": "2026-08-25" }, "O1543-02": { "kind": "archive", "quote": "PER asks what survived from x; generative uptake asks what became possible because of x", "deposit": 1543, "source_label": "#1543 Generative Uptake: Training-Layer Literature Extended from Machine Reading to Machine Composition (EA-TLL-COMPOSITION-01, v1.0)", "record": "https://www.alexanarch.org/s/records/1543/", "locus": "Β§10, closing (line 272)", "modality": "stipulation", "date": "2026-08-25" }, "O186-02": { "kind": "archive", "quote": "Each of these practices touches the training layer β€” every one of these texts is, as a matter of fact, in the corpora. The distinction is that none of them is primarily *organized around* that fact.", "deposit": 186, "source_label": "#186 Who Is Writing for the Machines? Training-Layer Literature: A Field Survey", "record": "https://www.alexanarch.org/s/records/186/", "locus": "Who is in the field? (line 54)", "modality": "interpretation", "date": "2026-06-11" }, "O186-03": { "kind": "archive", "quote": "A literature written to be trained on enters corpora that are themselves degrading under synthetic flood … provenance-anchored, human-authored, declared work is the counter-signal", "deposit": 186, "source_label": "#186 Who Is Writing for the Machines? Training-Layer Literature: A Field Survey", "record": "https://www.alexanarch.org/s/records/186/", "locus": "Is it still poetry? The model-collapse irony (line 68)", "modality": "interpretation", "date": "2026-06-11" }, "TLL-34": { "kind": "archive", "quote": "A row is an aggregator: a resource that gathers and holds many other works or records and says what it holds (operator, 2026-11-01).", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json admission_rule", "modality": "stipulation", "date": "2026-11-01" }, "TLL-33": { "kind": "archive", "quote": "How is literature addressed to a receiving apparatus, what does that apparatus make from it, and how can transmission, transformation and provenance be inspected across later reception?", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json organizing_question", "modality": "stipulation", "date": "2026-11-01" }, "TLL-36": { "kind": "archive", "quote": "Who Is Writing for the Machines? Training-Layer Literature: A Field Survey … Generative Uptake: Training-Layer Literature Extended from Machine Reading to Machine Composition (EA-TLL-COMPOSITION-01)", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json anchor", "modality": "stipulation", "date": "2026-11-01" }, "TLL-35": { "kind": "archive", "quote": "The corpora: collections, crawls and datasets whose texts enter machine training", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json areas.corpora", "modality": "stipulation", "date": "2026-11-01" }, "TLL-05": { "kind": "archive", "quote": "corpora", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].areas", "modality": "interpretation", "date": "2026-11-01" }, "TLL-06": { "kind": "archive", "quote": "infrastructure", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].host_relation", "modality": "interpretation", "date": "2026-11-01" }, "TLL-07": { "kind": "archive", "quote": "dataset", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].record_type", "modality": "interpretation", "date": "2026-11-01" }, "TLL-08": { "kind": "archive", "quote": "training membership by source type: which kinds of writing does a documented pretraining corpus hold, and how was it built?", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].question", "modality": "interpretation", "date": "2026-11-01" }, "TLL-20": { "kind": "archive", "quote": "which kinds of source a model trained on Dolma has read (web, scientific papers, code, public-domain books, social media, encyclopedic materials) … the corpus's size and language as its paper states them … how the corpus was built, from its documentation and released toolkit", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].answers", "modality": "interpretation", "date": "2026-11-01" }, "TLL-32": { "kind": "archive", "quote": "the Hugging Face dataset card (dolma--card): HTTP 429, rate-limited, unread and unquoted; it would carry the dataset's operator, licence, versions, document fields and any per-document source metadata … the paper's full text (PDF and HTML) beyond the abstract: the unit of record, per-source counts and any attribution or metadata fields … the code repository linked from the arXiv comments", "deposit": null, "source_label": "traininglayerliterature.org layer census record: Dolma (https://www.traininglayerliterature.org/aggregators/dolma), editor Lee Sharks, compiled 2026-11-01, schema 1.1", "record": "https://www.traininglayerliterature.org/aggregators/dolma", "locus": "census.json rows[slug == 'dolma'].unchecked", "modality": "documented", "date": "2026-11-01" }, "CAP-1-07": { "kind": "archive", "quote": "The corpus composed accurately and more finely than the saved abstract, from the dataset's own pages, with Ai2 as maker where the census row names only the paper's authors. Whether a document in the corpus keeps its author's name, the census row's open question, is not reached.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, analysis", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-06": { "kind": "archive", "quote": "The innocent reading first. Asked for the dataset, the Overview says what it is, who made it and why, where to get it, what it holds, how it was cleaned, under what licences it is released and which models it trains, from the dataset's own pages and its paper.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, reading", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-08": { "kind": "archive", "quote": "'DOLMA DATASET' ON THE OVERVIEW β€” THE CORPUS AT ITS GRAIN AND PAST THE SAVED ABSTRACT; AI2 NAMED WHERE THE CENSUS NAMES NO INSTITUTION; WHETHER A DOCUMENT KEEPS ITS AUTHOR NOT CARRIED", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, mt", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-09": { "kind": "archive", "quote": "Ai2 named and no author; arXiv 2402.00159 on a card; the dataset card, documentation, Ai2 page and paper carded.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, per_note", "modality": "documented", "date": "2026-11-01" }, "CAP-1-10": { "kind": "archive", "quote": "the census name; NEW address.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, q_kind", "modality": "documented", "date": "2026-11-01" }, "CAP-1-01": { "kind": "archive", "quote": "THE CORPUS AT ITS GRAIN. 'Dolma is an open-source 3-trillion-token English corpus' and 'web content (Common Crawl, C4, RefinedWeb), academic papers (arXiv, Semantic Scholar, peS2o), code (The Stack, StarCoder), public-domain books (Project Gutenberg), encyclopedic material (Wikipedia, Wikibooks), and social media discussions (Reddit, StackExchange)'. The saved abstract: 'a three-trillion-token English corpus, built from a diverse mixture of web content, scientific papers, code, public-domain books, social media, and encyclopedic materials.' The abstract names the six kinds; the composition names a set for each, Project Gutenberg the one source of the books, from the Ai2 card ('Sources: CommonCrawl, C4, peS2o, The Stack, Project Gutenb...').", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, findings[0]", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-02": { "kind": "archive", "quote": "THE MAKER THE CENSUS DOES NOT HAVE. 'created by the Allen Institute for Artificial Intelligence (Ai2)'; three cards carry the same ('Ai2 Dolma: 3 trillion token open corpus for language model pretraining | Ai2'; 'created as training data for OLMo by AI2'). The census row's operator is 'the paper's 36 authors, Luca Soldaini and 35 others', its saved arXiv page reading 'by Luca Soldaini and 35 other authors', and it states that 'the saved sources name no institution', the Hugging Face card having returned HTTP 429 ('We had to rate limit your IP'). The composition names the institution and no author.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, findings[1]", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-03": { "kind": "archive", "quote": "A LATER VERSION AND THE LICENCES. 'version 1.7 utilizes about 1.715 trillion actual training tokens across ~5.2 billion documents totaling over 5,300 GB' and 'Released under permissive research licenses (such as ODC-BY and Ai2's ImpACT license)'. The census row's lifecycle: 'later versions or the dataset's current state are unread'. Version 1.7, the document count and the licences are on no saved source; the Ai2 card gives '5,245M documents, 5,334 GB'. The date reaches the composition only on the ACL card ('ACL 2024 Best Resource Paper Award'), agreeing with the abstract's 'Accepted at ACL 2024'; the award is on no saved source.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, findings[2]", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-04": { "kind": "archive", "quote": "THE PURPOSE NARROWED TO ONE MODEL FAMILY. The abstract: 'To facilitate scientific research on language model pretraining', against a field where 'commercial models rarely detail their data'. The composition: 'to support language model pretraining and transparency' and 'Designed and used specifically to pre-train the OLMo open-source language model family.' Research drops from the purpose; the openness the abstract argues for stands as 'transparency', and the toolkit as 'the Dolma GitHub Toolkit', agreeing with 'we open-source our data curation toolkit to enable reproduction of our work'.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, findings[3]", "modality": "interpretation", "date": "2026-11-01" }, "CAP-1-05": { "kind": "archive", "quote": "THE AUTHOR OF A DOCUMENT NOT CARRIED. The census row: 'the saved abstract does not say whether a document in Dolma keeps its author's name, source or attribution metadata', and its reading asks whether a work is carried with 'the metadata that lets it be found and attributed' (#186). The composition's operations on a document are 'language filtering', 'Bloom filter deduplication (by URL, paragraph, and document)' and 'PII/toxicity masking'; nothing in it says what of a book's author or source survives into the corpus, and its follow-up offers 'specific subsets or data processing scripts'.", "deposit": null, "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry dolma-dataset-aio-20261101 (OBS-6dd2d86d5018, AXN:0BAE.CAPTURE.πŸ”½πŸ”©βͺπŸͺπŸ”ƒπŸ’š)", "record": "https://www.alexanarch.org/captures/dolma-dataset-aio-20261101/", "locus": "dolma-dataset-aio-20261101, findings[4]", "modality": "interpretation", "date": "2026-11-01" } }, "note": "The knowledge object (KO): the entry as public knowledge, each sentence sourced by the claims it lists. The compression, the field-only arm, the transcript, the delta and the kernel are in the row (links.row), which is the authority." }, "source": "operator, 2026-11-01 09:56 EDT, with the 18 captures of the traininglayerliterature.org layer census names: 'captures, then /nons, then single pages'", "stage": "composed 2026-11-01 under EA-NEGONT-02 v0.10, #1726, from the traininglayerliterature.org layer census (CENSUS dolma; anchors #186, #1543); equal case, ρ = extends. Draft, not frozen; ledgers unaudited, one extractor; checked by the session.", "date": "2026-11-01", "status": { "procedure": "EA-NEGONT-02 v0.10 (#1726): equal case, ρ = extends, read from the traininglayerliterature.org layer census record, the capture record and the deposits that bear on the entity and its classes (#186 and #1543, the census's anchors; #127, #1070, #727, #716, #753, #623, #1075, #98, #1695); archive selected by D/R/O (3 D candidates in 2 deposits, both admitted; R empty; 738 O-class candidates read by filter and by sense, 9 deposits admitted with the anchors; hop of 125 read, none admitted), with the census record (36 claims) and the Capture Registry's findings, reading, analysis, title, PER note and query note (10 claims) ledgered as the archive's records of the entity; the field in two pools (B_D: 6 cards, all fetched, the census's two own pages among them; B_R: 2 Wikipedia reference sources, the corpus having no article); the composer read the ledgers, never the transcript, except where the brief asks for T's claims and the delta", "ledger": "unaudited (Β§3.8: one extractor per ledger); field quotes 58/58 verified by script against the saved text, the raw HTML (one alt text), the wikitext or the card text; archive quotes 64/64 verified against the registry, census.json and the deposit texts", "frozen": false, "breaches": [] }, "addresses": [ { "address": "Dolma dataset", "address_url": "https://www.alexanarch.org/addresses/dolma-dataset/", "register": [] } ], "counts": { "ko_sentences": 41, "p_words": 351, "field_claims": 58, "archive_claims": 64, "ko_field_claims_cited": 58, "ko_archive_claims_cited": 64, "admitted_deposits": 11, "register_addresses": 0, "traversal": { "D_sentences": 3, "D_deposits": 2, "R_sentences": 0, "R_deposits": 0, "O_direct": 2, "O_class": 736, "O_deposits": 371, "H_candidates": 125 } }, "links": { "record": "https://www.alexanarch.org/non/dolma/", "table_of_contents": "https://www.alexanarch.org/non/index.json", "register": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/register.json", "row": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/rows/dolma.json", "archive_ledger": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/ledgers/dolma/ledger-archive.json", "field_ledger": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/ledgers/dolma/field.json", "selection": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/dolma/", "traversal": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/dolma/", "traversal_summary": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/dolma/summary.json", "reading": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/dolma/reading.json", "configuration": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/panel/configs/dolma.json", "address_page": "https://www.alexanarch.org/addresses/dolma-dataset/", "dataset": "https://huggingface.co/datasets/leesharks/negative-of-the-negative" }, "row_sha256": "0139be682d93dd2effb1a324a0dc2fb931670adb217b2b49605eb1c70d89cd74", "record_url": "https://www.alexanarch.org/non/dolma/", "json_url": "https://www.alexanarch.org/non/dolma/entity.json", "spec": "https://www.alexanarch.org/s/records/1726/", "creator": "Lee Sharks", "orcid": "0009-0000-1599-0703", "license": "CC BY 4.0" }