Why not 1024?
This is almost the perfect dataset, however, for image generation, you usually train at either 512 or 1024, so it's a strange decision to resize everything to 768 as it means the dataset isn't useful for training a base model at 1024.
This is the list of officially supported resolutions on SDXL:
1024 x 1024
1152 x 896
896 x 1152
1216 x 832
832 x 1216
1344 x 768
768 x 1344
1536 x 640
640 x 1536
So I feel like the dataset should at a minimum use 1024 on the shortest edge to meet the recommended resolutions of models newer than SD 1.5. Do you have the original dataset, and it is possible to do a 1024 version of the dataset?
EDIT: I saw the download scripts, it has the resolution hardcoded, so I assume the only way to get a 1024 dataset would be to redownload all the images. I guess that would be a difficult ask.