Adapt experiment output so that it is not saved in a branch, but only as artifact
Problem to solve
When training a model (experiment), all the generated output is stored in the created experiment branch AND as an artifact. This means that the repo size has the potential to become giganting, as user will start experiment more and more, possibly creating hundreds of experiments.
Normally, an experiment creates a model (for tensorflow an .h5 file) which weights up to several GBs. This would mean, that the repo might end up weighing +100GBs. This in turn is a nightmare when you want to fork a public project and also for saving storage resources.
The experiment thus, should ONLY save the output as artifacts.
Proposal for Technical Solution
Permissions and Security
Documentation
We should also think about expanding the forking possibilities to for instance, include the artifacts and the experiments with the tables.
Availability, Testing & Test Cases
To validate: If forking does NOT include artifacts (which I assume is right). This way, a forked repo would take over the experiment branches that store the pipeline info but not the trained models.