[D] Post deployment best practises with large scale projects

I am wondering after deploying a model in a large computer vision project, what do you do with the images used for training?

Do you:

1. Delete them?
2. Sitting there in cloud storage just in case
3. Use embeddings for future downstream tasks
4. Move them into cold storage

Asking for a friend.

case cloud cloud storage computer computer vision deployment embeddings future images machinelearning project projects scale storage tasks them training vision