Caching of same but differently loaded dataset

[Revist]

I have two datasets: COCO and COCO-20i. These datasets are the same but they are loaded in a different way i.e. one returns
image + annorations for this image, second returns 6 images with a given class + annotations of this class. The underlying data _ anntations are the same, the only difference is how _generate_examples is executed. I could create dataset loading script with these two configurations.

My questions is:
As this results in arrow files in the cache will these replicate the image data i.e. for every _generate_examples i have different dataset so different arrow files resulting in dataset growing with number of configurations?