correction (h/t @emma.best @crimew.gay): hundreds of gigabytes. don't know why i recalled the data being so large in size. probably because there are upward of 115,000 actual files in the dataset, and many are document scans?