| 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) | |