Dataset · kakaobrain
747M Image-Text Pairs
kakaobrain/coyo-700m
A large corpus of alt-text and image pairs from HTML for training multimodal models.
What's inside
COYO-700M contains 747M image-text pairs and additional meta-attributes. Pairs are collected from alt-text and associated images in HTML documents using a strategy similar to previous vision-and-language datasets. Data is in English, with image, tabular, and text modalities for training large foundation models.
What you can do with it
- Train a CLIP-style image-text model
- Fine-tune a text-to-image model
- Evaluate zero-shot image classification
- Tasks
- Text → image, Image → text, Zero shot classification
- Languages
- English
- Size
- 100M–1B rows
- Downloads this month
- 6.5K
- Likes
- 177
Models trained on it
- OpenGVLab/InternViT-300M-448pxImage feature extraction 866
- OpenGVLab/InternViT-6B-224pxImage feature extraction 674
- BAAI/EVA-CLIP-18BEmbeddings 597
- BAAI/EVA-CLIP-8BEmbeddings 539
- OpenGVLab/InternVL-14B-224pxImage feature extraction 406
- OpenGVLab/InternViT-6B-448px-V1-5Image feature extraction 391
- DeepGlint-AI/mlcd-vit-large-patch14-336Embeddings 350
- BAAI/EVA-CLIP-8B-448Embeddings 167
How to load it
from datasets import load_dataset
ds = load_dataset("kakaobrain/coyo-700m")