It would be neat of there could be defined curated chunks of standardize text data sets for LLMs training distributed over bittorrent.

Maybe there already is and I just haven’t looked hard enough.