The original link for the Karpathy splits, commonly used in image captioning, seems to have gone offline. This repo contains the json files from the original link for others to use.
I had to use git LFS to get the files on GitHub. Please don't clone the repo as my LFS bandwidth is very limited (I'm only a student!). Instead, download each file via wget:
COCO: wget -c https://github.com/Delphboy/karpathy-splits/raw/main/dataset_coco.json\?download\= -O dataset_coco.json
Flickr32K: wget -c https://github.com/Delphboy/karpathy-splits/raw/main/dataset_flickr32k.json\?download\= -O dataset_flickr32k.json
Flickr8K: wget -c https://github.com/Delphboy/karpathy-splits/raw/main/dataset_flickr8k.json\?download\= -O dataset_flickr8k.json