Training steps automation

Just like yolov8 is it possible for detectron2 to automate and export perfectly trained steps , As of now we are required to provide number of steps check for detection if not found train again , In yolov8 you just pass on the batch size and the training starts and it exports optimum model by checking validation loss .

The issue names no files, tests, or entry points. Start by reviewing Detectron2's existing training and validation workflow, then clarify the desired automation and export criteria; done would require an agreed scope and demonstrated behavior.

Written by the indexing model from the issue text.