DVC: Git for Your ML Training Data

You version code with Git. What about your model training data?

If you’ve ever asked “Which dataset trained this model?” or “Can we reproduce last month’s model exactly?”, you need DVC.

What DVC Solves

Your Weekend Starter

Six commands. That’s all you need. (Full DVC docs)

1
2
3
4
5
6
7

pip install dvc
dvc init
dvc remote add -d myremote s3://mybucket/data
dvc add data/training.csv
git add data/training.csv.dvc
git commit -m "Track data with DVC"
dvc push

Your data is now versioned.

What Just Happened

• DVC hashed your data file (content-addressable storage)
• Created a small .dvc pointer file (a few bytes)
• Git tracks the pointer file (small, fast)
• dvc push sends actual data to remote storage (S3, GCS, Azure Blob)

Data stays out of Git. Versioning stays in Git.

The Workflow

Change the data? Run dvc add again. New hash, new pointer, new commit.

Want last week’s data? git checkout <commit> + dvc checkout. Done.

Same Git workflow. New file types.

The DevOps Parallel

If you know Git, you know DVC. The commands are intentionally identical.

Your ML models are only as reproducible as your data. Start versioning it.

This is Part 4 of the MLOps for DevOps Engineers series. Next up: 5 Levels of ML Model Deployment.

For weekly MLOps tips, join the newsletter.