Hugging Face Datasets

Gotchas

• 100 GB free private storage

  Free accounts get 100 GB of private repository storage across all repos (models, datasets, Spaces); usage beyond this requires a paid plan.

• Public storage is best-effort only

  Free users receive unlimited-in-principle public storage but Hugging Face reserves the right to restrict or remove content that does not demonstrate genuine community value (downloads, likes).

• 500 GB hard cap per file

  No single file in any repository may exceed 500 GB; files larger than 200 GB are strongly discouraged and will not be CDN-cached, causing slower downloads.

• Dataset card required for large datasets

  Uploading large public datasets requires a dataset card describing the data; datasets without one may be restricted or removed.

• Not transactional / low-latency storage

  HF dataset repos are Git-backed (LFS) and are designed for batch exports and snapshots, not for low-latency random-access queries or high-frequency writes.

• 100k files per repo recommended max

  Repositories with more than 100,000 files experience degraded Hub UI and git performance; folders are capped at 10,000 entries.

• No HA / no versioned backup

  Free tier does not include automated backups or high-availability guarantees; history squash operations are irreversible and cannot be undone.

• Large public files not CDN-cached

  Files larger than ~200 GB are not cached by CloudFront, resulting in slower download speeds for consumers of the dataset.