Pruned dataset, used COG captioning instead of LLaVA
In the style of the classic TV show 'Saved by the Bell'