Dataset · HuggingFaceM4
Multimodal document recognition dataset
HuggingFaceM4/DoclingMatix
From 1 to 10 million document image-text pairs in English for training VLM models.
What's inside
A large multimodal dataset from HuggingFaceM4 in English with a volume of 1 to 10 million records, combining images and text. It was created by reprocessing the Docmatix corpus specifically to train the compact SmolDocling model for end-to-end document conversion.
What you can do with it
- Training and fine-tuning multimodal models (VLM) for document parsing and OCR
- Training models for question answering on scans and pages (Visual Question Answering)
- Tasks
- Visual question answering, Image text → text
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 2.9K
- Likes
- 58
Models trained on it
- ibm-granite/granite-docling-258MImage text → text 423.4K
- docling-project/SmolDocling-256M-previewImage text → text 24.9K
- docling-project/granite-docling-2stage-258mImage text → text 573
- mradermacher/SmolDocling-256M-preview-i1-GGUF 405
- mradermacher/SmolDocling-256M-preview-GGUF 350
- Userb1az/granite-docling-258M-GGUFImage text → text 185
- Mungert/granite-docling-258M-GGUFImage text → text 166
- tevfikk/granite-docling-258M-bitnetImage text → text 60
How to load it
from datasets import load_dataset
ds = load_dataset("HuggingFaceM4/DoclingMatix")