Ask HN: How do you manage in-house training data for ML?

How do you version your data? Git, Git Annex, Git LFS, DVC, something else? How do you manage training data? Do you make lots of small additions or add things in batches?