metadata

dataset_info:
  features:
    - name: image
      dtype: image
    - name: text
      dtype: string
    - name: markdown
      dtype: string
    - name: html
      dtype: string
    - name: file_name
      dtype: string
    - name: ocr
      dtype: string
  splits:
    - name: train
      num_bytes: 27893994569.17
      num_examples: 113091
  download_size: 25517477643
  dataset_size: 27893994569.17
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

HuCCPDF

Approximately 113k pages of Hungarian PDFs from the Common Crawl, as featured in our paper "Synthetic Document Question Answering in Hungarian". The text field is extracted using PyMuPDF, and the ocr field is extracted using pytesseract.

See other datasets from the paper:

HuDocVQA (https://huggingface.co/datasets/jlli/HuDocVQA)
HuDocVQA-manual (https://huggingface.co/datasets/jlli/HuDocVQA-manual)