Is your feature request related to a problem? Please describe.
In the new dlc3.0.0 release with PyTorch, some metrics are displayed after every epoch (test error / test recall/precision). But the model snapshots are saved solely on iteration/step/epoch numbers. When going through the logs I can see (for example) the metrics after epoch 178 is performing really well, but the only snapshots of models I have is 175 or 200.

Describe the solution you'd like
Having an additional option during train_network that allows DLC to save the 5 model snapshots with the best metrics/test.rmse or the best metrics/test.mAP_pcutoff.

Describe alternatives you've considered
I could make save epoch being 1 and just save all model snapshots after every epoch to cherry pick. But that would take up a lot of storage space and still kind of frustrating to navigate through a sea of snapshots.

Additional context
N/A