Hello and thank you very much!

as we do not know what has been used to train CLIP, we cannot be sure if the data we have been using was included in the original model.

Also, consider that some of our captions (~700k) are from the WIT dataset that is in Italian.