Datasets:

davanstrien
/

ency-deep-ocr

volume_id stringlengths 9 9	page_number int32 1 879	file_identifier stringlengths 9 9	text stringlengths 0 8.94k	alto_xml stringlengths 714 233k	has_image bool 1 class	has_alto bool 1 class	document_metadata stringlengths 61 217	has_metadata bool 1 class	edition stringclasses 9 values	volume_part stringlengths 8 52	publication_year stringclasses 11 values	editor stringclasses 5 values	full_title stringclasses 6 values	shelf_locator stringclasses 14 values	markdown stringlengths 3 42.4k	inference_info stringclasses 1 value
191253817	213	191927103	"I\nN\nD\nE\nX\nA.\njrfBDOMEN, wounds of, r°\nI 16, How treated, 26.\nCollection of water in, 263,\(...TRUNCATED)	"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	Encyclopaedia Britannica - Third edition, Volume 18, STR-ZYM - EB.5	true	Third edition	Third edition, Volume 18, STR-ZYM	1797		Encyclopaedia Britannica	EB.5	"1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. (...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
144850377	768	188279784	"459°\nM E D I\nnot found, oi* of a neutral and iridifferent nature.\nThefe are the body itfelf, th(...TRUNCATED)	"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	Encyclopaedia Britannica - Second edition, Volume 6, K-Medicine - EB.4	true	Second edition	Second edition, Volume 6, K-Medicine	1778-83		Encyclopaedia Britannica	EB.4	"4590\n\nMEDICINE\n\nnot found, or of a neutral and indifferent nature. These are the body itsel(...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
190273289	577	190285942		"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	Encyclopaedia Britannica - Second edition, Volume 9, POI-SCU - EB.4	true	Second edition	Second edition, Volume 9, POI-SCU	1778-83		Encyclopaedia Britannica	EB.4	0.0	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
191253807	425	192057264	"C H E M I\nIt is obtained in a ftate of im-\nMill-\n„.j. has been long known.\npurity at the bott(...TRUNCATED)	"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	"Supplement to the third edition of the Encyclopaedia Britannica ... Illustrated with ... copperplat(...TRUNCATED)	true		Volume 1, ABE-IMP	1801	Gleig, George	"Supplement to the third edition of the Encyclopaedia Britannica ... Illustrated with ... copperplat(...TRUNCATED)	EB.7	"381\n\nCHRISTY.\n\nhas been long known. It is obtained in a state of impurity at the bottom, and ad(...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
192200901	593	192726003	"MAN\n[ 545 1\nMAN\nMan. heads arife from the compreffion they undergo in in-\nl —' fancy. This ra(...TRUNCATED)	"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	"Encyclopaedia Britannica, or, a Dictionary of arts, sciences, and miscellaneous literature : enlarg(...TRUNCATED)	true	Fifth edition	Fifth edition, Volume 12, LIE-Materia medica	1815		"Encyclopaedia Britannica, or, a Dictionary of arts, sciences, and miscellaneous literature : enlarg(...TRUNCATED)	EB.10	"31\n\nMAN\n\nheads arise from the comprehension they undergo in infancy. This race comprehends (...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
193108326	363	193251551	"POPULATION\nPopula-\ntion.\nthe whole of Sweden, where the annual mortality, at the\ntime referred (...TRUNCATED)	"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	Encyclopaedia Britannica - Eighth edition, Volume 18, PLA-REI - EB.16	true	Eighth edition	Eighth edition, Volume 18, PLA-REI	1853-1860	Stewart, Dugald	Encyclopaedia Britannica	EB.16	"351\n\nPupilation\n\nThe whole of Sweden, where the annual mortality, at the time referred to b(...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
191253803	3	192007356		"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	Encyclopaedia Britannica - Third edition, Volume 6, DIA-ETH - EB.6	true	Third edition	Third edition, Volume 6, DIA-ETH	1797		Encyclopaedia Britannica	EB.6	1 E.B.6.	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
192547788	1	192866681		"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	"Supplement to the fourth, fifth and sixth editions of the Encyclopaedia Britannica - Volume 5 - EB.(...TRUNCATED)	true		Volume 5	1824	Stewart, Dugald	Supplement to the fourth, fifth and sixth editions of the Encyclopaedia Britannica	EB.13	"0.0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000(...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
149981670	484	188743743		"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	Encyclopaedia Britannica - Third edition, Volume 8, GOB-HYD - EB.5	true	Third edition	Third edition, Volume 8, GOB-HYD	1797		Encyclopaedia Britannica	EB.5	"466\n\nRules of Heraldry.\n\nOf the Rules or Laws of HERALDRY.\n\nTHE several cefutcheons, tinétur(...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)
144133903	413	144809503	"M\nU\no\nTheorem. The interrals of the notes of all (harp\nkeys and flat keys refpeclively, are pro(...TRUNCATED)	"<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n<alto xmlns=\"http://www.loc.gov/standards/alto/v3/alto(...TRUNCATED)	true	true	"Encyclopaedia Britannica; or, A dictionary of arts and sciences, compiled upon a new plan … - Fir(...TRUNCATED)	true	First edition	First edition, 1771, Volume 3, M-Z	1771	Smellie, William	Encyclopaedia Britannica; or, A dictionary of arts and sciences, compiled upon a new plan	EB.1	"353\n\nM U S\n\nThe intervals of the notes of all sharp keys and flat keys repeatedly, are prop(...TRUNCATED)	"[{\"column_name\": \"markdown\", \"model_id\": \"deepseek-ai/DeepSeek-OCR\", \"processing_date\": \(...TRUNCATED)

End of preview. Expand in Data Studio

Document OCR using DeepSeek-OCR

This dataset contains markdown-formatted OCR results from images in davanstrien/ency-test using DeepSeek-OCR.

Processing Details

Source Dataset: davanstrien/ency-test
Model: deepseek-ai/DeepSeek-OCR
Number of Samples: 100
Processing Time: 8.5 min
Processing Date: 2025-10-22 17:48 UTC

Configuration

Image Column: image
Output Column: markdown
Dataset Split: train
Batch Size: 512
Resolution Mode: large
Base Size: 1280
Image Size: 1280
Crop Mode: False
Max Model Length: 8,192 tokens
Max Output Tokens: 8,192
GPU Memory Utilization: 80.0%

Model Information

DeepSeek-OCR is a state-of-the-art document OCR model that excels at:

📐 LaTeX equations - Mathematical formulas preserved in LaTeX format
📊 Tables - Extracted and formatted as HTML/markdown
📝 Document structure - Headers, lists, and formatting maintained
🖼️ Image grounding - Spatial layout and bounding box information
🔍 Complex layouts - Multi-column and hierarchical structures
🌍 Multilingual - Supports multiple languages

Resolution Modes

Tiny (512×512): Fast processing, 64 vision tokens
Small (640×640): Balanced speed/quality, 100 vision tokens
Base (1024×1024): High quality, 256 vision tokens
Large (1280×1280): Maximum quality, 400 vision tokens
Gundam (dynamic): Adaptive multi-tile processing for large documents

Dataset Structure

The dataset contains all original columns plus:

markdown: The extracted text in markdown format with preserved structure
inference_info: JSON list tracking all OCR models applied to this dataset

Usage

from datasets import load_dataset
import json

# Load the dataset
dataset = load_dataset("{{output_dataset_id}}", split="train")

# Access the markdown text
for example in dataset:
    print(example["markdown"])
    break

# View all OCR models applied to this dataset
inference_info = json.loads(dataset[0]["inference_info"])
for info in inference_info:
    print(f"Column: {{info['column_name']}} - Model: {{info['model_id']}}")

Reproduction

This dataset was generated using the uv-scripts/ocr DeepSeek OCR vLLM script:

uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/deepseek-ocr-vllm.py \\
    davanstrien/ency-test \\
    <output-dataset> \\
    --resolution-mode large \\
    --image-column image

Performance

Processing Speed: ~0.2 images/second
Processing Method: Batch processing with vLLM (2-3x speedup over sequential)

Generated with 🤖 UV Scripts

Downloads last month: 14

Size of downloaded dataset files:

18.7 MB

Size of the auto-converted Parquet files:

18.7 MB

Number of rows:

100