Dataset · mlfoundations
Multimodal PDF corpus for VLM pre-training
mlfoundations/MINT-1T-PDF-CC-2024-10
Part of the giant open MINT-1T dataset featuring interleaved English text and PDF images.
What's inside
This repository is part of the large open multimodal MINT-1T dataset by University of Washington researchers (containing 1 to 10 million records in this split). It includes interleaved English text and images extracted from PDF files. The full MINT-1T project spans 1 trillion text tokens and 3.4 billion images, including arXiv scientific papers.
What you can do with it
- Pre-train multimodal language models (VLM) capable of understanding complex document layouts
- Train information extraction systems and illustration analysis in PDF files
- Tasks
- Image → text, Text generation
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 21.5K
- Likes
- 5
How to load it
from datasets import load_dataset
ds = load_dataset("mlfoundations/MINT-1T-PDF-CC-2024-10")