How to track multiple trainings and models using single DVC repo and yaml?

[aveek-brahmachari]

A git project may have many different models for different purposes. It does not seem to be a solution to have multiple DVC repos and track each training/experiment/model separately.

So the question is if there are multiple tasks: Task A, Task B, Task C, Task D,......Task Z and there are various models trained for the same, how can we use DVC to efficiently do this ? Did not find any example explaining this.

[daavoo]

You can initialize multiple DVC Repos in separated subdirectories under the same git project:

https://dvc.org/doc/command-reference/init#initializing-dvc-in-subdirectories

[aveek-brahmachari]

Is there any more elegant method of doing this ? If I have 100 models that I train in a project, it would mean I would need 100 DVC repos to track them too.

[karajan1001]

You can create for each stages which could create batches of similar stages.