Datasetpublic

p4ulbr4dl3y/dogovors-unlimited-ocr

Dogovors Unlimited-OCR Dataset OCR/document-layout dataset prepared for fine-tuning baidu/Unlimited-OCR. Files train.jsonl contains one JSON object per document. images/ contains the page images referenced by relative path. JSONL Schema { "images": [ "images/doc_001_page_001.jpg", "images/doc_001_page_002.jpg" ], "question": "Multi page parsing.", "answer": "<PAGE><|det|>title [400, 60, 660, 73]<|/det|>... <PAGE><|det|>text [100… See the full description on the dataset page: https://huggingface.co/datasets/p4ulbr4dl3y/dogovors-unlimited-ocr.

sourceHugging Faceupdated 3mo agoView on Hugging Face

0likes16downloads

Dataset Card

Dogovors Unlimited-OCR Dataset

OCR/document-layout dataset prepared for fine-tuning baidu/Unlimited-OCR.

Files

—train.jsonl contains one JSON object per document.
—images/ contains the page images referenced by relative path.

JSONL Schema

json

{
  "images": [
    "images/doc_001_page_001.jpg",
    "images/doc_001_page_002.jpg"
  ],
  "question": "Multi page parsing.",
  "answer": "<PAGE><|det|>title [400, 60, 660, 73]<|/det|>...
<PAGE><|det|>text [100, 80, 900, 120]<|/det|>..."
}

Each row represents a full document. Single-page documents still use the same shape: one image path and one <PAGE> section in answer.

For PyTorch/Transformers training with the original Baidu model, prepend the image token in the rendered prompt, for example <image>Multi page parsing.. The stored question intentionally does not include <image> so the dataset also remains compatible with MLX prompt rendering.