Skip to content

Dataset · apple

Synthetic descriptions and embeddings from Apple for 2B images

apple/DFNDR-2B

Metadata, embeddings, and generated captions for the filtered DFN-2B set.

What's inside

The release from Apple contains English synthetic captions, embeddings, and metadata for 2 billion examples (a DFN-2B subset of DataComp-12B). The data was generated by pre-trained multimodal models as part of the MobileCLIP project. Original ground-truth captions are not included in this set.

What you can do with it

  • Training fast and lightweight multimodal models (like MobileCLIP)
  • Knowledge distillation and experiments with distilled text-image embeddings
Tasks
Text → image, Image → text
Languages
English
Size
1B–10B rows
Downloads this month
27.2K
Likes
10

How to load it

from datasets import load_dataset

ds = load_dataset("apple/DFNDR-2B")