DVC

DVC connects data and model versions to the code in your Git repository, so you can reproduce a machine learning experiment with the inputs it used. It's a free, open-source tool under Apache 2.0 for individual data scientists and small projects. It runs on macOS, Windows and Linux.

Data and model files live in a cache outside Git, while Git records their version information alongside your code. You can keep artifacts on your machine or store and share them through cloud storage. Supported storage connections include Amazon S3, Azure, Google Cloud Storage, Google Drive and SSH.

Experiment tracking runs locally in your Git repository without a separate server. DVC lets you compare data, code, parameters, models and performance plots, so you can check what changed between experiments rather than rely on filenames or notes. You can also share experiments through existing Git hosting, including GitHub and GitLab, and reproduce someone else's experiment.

Its pipelines record how code and data produce model artifacts. When an input changes, DVC reruns only the affected steps, avoiding a full pipeline run for every edit. Git also versions the pipeline definitions, keeping the workflow tied to the project history.

DVC has a command-line interface and a VS Code extension. The extension brings experiment tracking and data management into the editor; the core tool handles versioning, pipelines and experiment reproduction.