README.md

---
dataset_info:
  features:
  - name: __key__
    dtype: string
  - name: __url__
    dtype: string
  - name: json
    struct:
    - name: pages
      list:
      - name: images_bbox
        sequence:
          sequence: float64
      - name: images_bbox_no_text_overlap
        sequence:
          sequence: float64
      - name: lines
        struct:
        - name: bbox
          sequence:
            sequence: float64
        - name: score
          sequence: float64
        - name: text
          sequence: string
        - name: word_slice
          sequence:
            sequence: int64
      - name: words
        struct:
        - name: bbox
          sequence:
            sequence: float64
        - name: line_pos
          sequence:
            sequence: int64
        - name: score
          sequence: float64
        - name: text
          sequence: string
  - name: pdf
    dtype: binary
  splits:
  - name: train
    num_bytes: 534443008
    num_examples: 200
  download_size: 259345736
  dataset_size: 534443008
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---


For benchmarking pdf operations. A small subset of [pdfa-eng-wds](https://huggingface.co/datasets/pixparse/pdfa-eng-wds).