Dataset · OpenGVLab
210M interleaved text and image Common Crawl documents
OpenGVLab/OmniCorpus-CC-210M
A massive filtered multimodal corpus from OpenGVLab for training text-image understanding models.
What's inside
The repository contains 210 million English documents featuring interleaved text and images. The data is filtered from the larger OmniCorpus-CC corpus collected from the Common Crawl web archive and released as part of work for ICLR 2025.
What you can do with it
- Pretraining vision-language models (VLM) for image-to-text and visual question answering (VQA) tasks
- Training systems capable of understanding long web documents with mixed text and graphic content
- Tasks
- Image → text, Visual question answering
- Languages
- English
- Size
- 100M–1B rows
- Downloads this month
- 156
- Likes
- 35
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("OpenGVLab/OmniCorpus-CC-210M")