> As for dataset, depending on your task, you need image / text pairs. I guess the main question is, do you just prepare training data as if you were training

> As for dataset, depending on your task, you need image / text pairs. I guess the main question is, do you just prepare training data as if you were training from scratch, or is there some particularities to finetuning that should be considered?