Hi, what is the workflow for saving old experiments that haven’t “failed” in the sense there was some error and no output was created, but were not the best performing (by whatever metric) and therefore not persisted?
The documentation speaks of persisting experiments (by turning them into a branch) and removing the old ones by dvc exp gc/remove, but I would like to keep a record of my past attempts of trying to improve the performance. The parameters and performance metrics at a minimum and preferably without bloating the cache and remote storage (I don’t actually need the data as I have rejected them).
Thank you!