DVC.
DVC versions datasets and models alongside code in Git. Large files are stored in remote object storage while lightweight pointers are committed, so a checkout can retrieve the exact data a commit was built against.
Where DVC fits on a long engagement.
Reproducibility in machine learning requires versioning data as well as code, and Git alone cannot hold multi-gigabyte files. DVC resolves that by committing content-addressed pointers and keeping the data in object storage, so checking out an old commit retrieves the data that commit used.
Its pipeline definitions add dependency tracking, so a stage reruns only when its inputs change. On expensive training pipelines that avoids recomputing stages whose inputs are identical, which is a substantial saving in both time and compute.
What an assigned team does with DVC.
Storage discipline is required. Every version of every dataset accumulates in remote storage, and without a retention policy the cost grows steadily for data nobody will ever check out again.
Deciding what to keep and for how long is ongoing housekeeping rather than a setup step, held within an agreed committed monthly capacity.
What we use DVC for.
- Data retrievable per commit Checking out an old commit and getting the data it was built against.
- Stages rerun only when inputs change Dependency tracking that avoids recomputing expensive steps unnecessarily.
- Storage retention managed A policy for old versions, because they accumulate indefinitely otherwise.
How DVC capacity is assigned.
Data versioning is assigned inside AI capacity, with a storage retention policy agreed at setup rather than after the bill grows.
Tell us what your roadmap needs DVC for.
A service delivery manager replies with the disciplines we would assign, the monthly capacity and what the first month looks like.
Loading the contact form… You can also email hello@azendo.co.
We reply within one working day. No obligation, and no newsletter.