v3.0 is strongly recommended. I have experimented and improved hyper parameters, dataset size and quality as well dataset captioning.
Triggerword: kchisara
Description
Dataset Optimization and more chackpoints saved allowed me to better cherry picking.