7,000,060 newspaper page scans from 650 German newspapers, each shipped with the layout boxes and OCR transcription produced for this corpus, and — for 38 % of pages — the digitising library's own ALTO XML. Every page here comes from an issue whose own rights statement is CC Public Domain Mark 1.0 or CC0 1.0. Sources were harvested from the Deutsche Digitale Bibliothek newspaper portal. Shards are data/pages-NNNNN.tar, ordered chronologically; an issue is never split across two shards. Each page is one WebDataset sample, key: -.json boxes are in the coordinates of the.jpg in the same sample. They were produced on the full-resolution scan and rescaled by the same factor the image was, so…