You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
Access to this dataset (hereinafter 'the Materials') is granted subject to the following terms. The Materials were compiled from publicly accessible internet sources and consist of Latin-language text; any biases or viewpoints present in the Materials are attributable to the source material and do not reflect the views of the author(s). By requesting access, the requesting party ('Licensee') represents and warrants that the Materials shall not be used, reproduced, modified, distributed, or otherwise exploited, in whole or in part, for any commercial purpose, including without limitation the training, fine-tuning, evaluation, or distillation of commercial models or systems; the development or improvement of any product or service offered for sale, subscription, or other pecuniary consideration; or any activity conducted by, on behalf of, or for the benefit of a for-profit entity. All use, reproduction, or modification of the Materials shall include clear attribution to the original author(s) and a link to this repository. Derivative works shall propagate these terms to downstream recipients without modification. Breach of any term herein shall constitute grounds for revocation of access. In the event of breach, any models, datasets, derivative works, or other materials produced utilizing the Materials in contravention of these terms shall be deemed produced through unauthorized use, and the Licensee assigns all right, title, and interest in such materials to the original author(s), such assignment taking effect upon the occurrence of the breach.
Log in or Sign Up to review the conditions and access this dataset content.
Acervus is an extensively curated and cleaned Latin language dataset spanning all orthographies and periods.
It contains nearly 24 billion high quality Latin tokens from a wide array of sources, predominantly free of OCR artifacts.
This includes nearly 1 million novel samples recovered from poorly digitized manuscripts.
As of January 2026, it is likely the largest publicly available corpus of clean Latin text to date.
Samples are marked according to provenance and inferred quality. Organic samples were sourced primarily manually from a large number of established public Latin repositories. Corrected samples are derived from heavily damaged excerpts found within LatinPD and refined with the aid of Emendator, an OCR correction model trained expressly for Latin. Samples obtained from CommonCrawl, CommonCorpus, LatinPD, FineWeb2, and FinePDF are similarly marked.
Exact token count: 23,890,998,586
@misc{mccarthy2025Acervus,
author = {McCarthy, A. M.},
title = {{Acervus}: A Clean Latin Text Dataset},
year = {2025},
howpublished = {\url{https://huggingface.co/datasets/aimgo/Acervus}},
note = {Dataset}
}
This dataset is still a work in progress. There remain known uncorrected manuscripts, however OCR error correction at scale requires potentially thousands of hours of inference. Acervus was produced without any formal institutional backing under concomitant resource restraints.
- Downloads last month
- 14