We created a CLIP index for the datacomp-12.8M dataset 🔍 🖼️

[RobbeSneyders]

We created a CLIP index for the datacomp-12.8M dataset to help users search through it and use it more easily. We used Fondant to create it and published it on the huggingface hub. You can find more details and info on how to use it in this short post.

Let us know what you think, we might create more indices for the larger datasets later.

[ludwigschmidt]

Very cool, thanks for sharing! Are you also planning to make indices for the larger pools we have?

[RobbeSneyders]

We haven't had the biggest show of interest for the small index yet, so currently we don't see a lot of benefit in making the additional investment. If there's a way to bring this more to the attention of the community, that could tip the scales.