The 10 Best Data Versioning Tools for ML in 2027

DVC is a command line tool and VS Code extension for reproducible machine learning projects. It versions data and models in your cloud storage while keeping version info in your Git repo, runs lightweight pipelines that rerun only affected steps, and tracks experiments in your local Git repo with no servers needed. It is free under the Apache-2.0 license, and its GitHub repository showed 15.9k stars on 2026-10-20. The site now brands it DVC by lakeFS.

lakeFS turns object storage into a Git-like repository: you create branches and get a copy of the full production data without copying anything, then roll back to a state from before an error. It works with AWS S3, Azure Blob Storage and Google Cloud Storage. The Team plan is $499 per month for up to 5 users and 500GB of data, and the license is Business Source License 1.1, which is source-available rather than open source. The repository showed 5.6k stars on 2026-10-20.

Delta Lake is an open-source storage framework from an independent project under Linux Foundation Projects. Its time travel lets you query an older snapshot with VERSION AS OF or TIMESTAMP AS OF, and by default you can go up to 30 days back unless you run VACUUM or change retention settings. It works with engines including Spark, Trino, Flink and Snowflake.

Apache Iceberg is a high-performance format for huge analytic tables. Time travel gives reproducible queries that use exactly the same table snapshot, and version rollback lets users quickly correct problems. Spark, Trino, Flink, Presto, Hive and Impala can work on the same tables at once. It is Apache-2.0 licensed and showed 9.4k GitHub stars on 2026-10-20.

Pachyderm describes itself as data-centric pipelines and data versioning, with immutable data lineage for any data type. Data sits in standard object stores with automatic deduplication, and pipelines trigger automatically when data changes are detected, running on Kubernetes with autoscaling. The repository is Apache-2.0 licensed and showed 6.3k stars on 2026-10-20.

Artifacts is the dataset and model versioning feature in Weights & Biases, now hosted by CoreWeave. It records what goes into and comes out of a run, such as a dataset in and a trained model out. It is included on Free, Pro and Enterprise plans: Free is $0 per month with 5 GB per month of storage, and Pro starts at $60 per month with 100 GB per month.

DagsHub is a hosted platform for ML data, code and experiments. The Individual plan is $0 per user per month with unlimited public repositories, and Team is $99 per user per month billed yearly or $119 monthly, with up to 10 team members and up to 1TB of data or up to 2 million files. Enterprise is a custom quote with petabyte-scale data management and VPC or air-gapped installation.

Git LFS is an open source Git extension for versioning large files. It keeps large files out of your Git history by storing small text pointers in the repository, with the contents on a remote host such as GitHub.com. Setup is git lfs install once, then git lfs track for the file types to manage. Its repository showed 14.6k stars on 2026-10-20.

On the Hugging Face Hub every dataset is a Git repository holding the data needed to generate training, evaluation and testing splits, and many include a Dataset Viewer. Visibility can be toggled between private and public in Settings. PRO is $9 per month per account, and storage is $12 per TB per month for public repos and $18 for private repos at base volume.

Amazon S3 Versioning keeps multiple variants of an object in the same bucket so you can preserve, retrieve and restore every version. It is disabled by default and must be explicitly enabled. There is no separate versioning price on the page, but normal S3 rates apply to every version, and each version is the entire object rather than a diff.