The datasets live under data/ and are stored with
Git LFS. LFS reads are redirected to the
Hugging Face mirror
via .lfsconfig, so dataset objects are fetched from Hugging
Face's CDN rather than from GitHub's shared (and limited) LFS bandwidth. This is
automatic — you do not need to configure anything.