Questions on training data

[adithyaiyer1999]

Hi!

Thanks for your great work. I had 2 questions regarding the dataset you trained on.

1. Would you be releasing the exact post filtration data the final model was trained on?
2. Could you give me a rough estimate of how many images were present in your final training dataset?

Thanks again! Adi

[yzy-thu]

The training dataset is approximately 145M images with resolution higher than 1024, of which 140M come from laion-highresolution and can be directly accessed from Hugging Face. The remaining part has no plan to release, but I believe this part is not important for training.

[adithyaiyer-morphic]

Thanks! Makes sense.