pol-Latn split of FinePDFs. This should be about 54.63B gpt-2 tokens worth of Polish text extracted from PDFs.