Downloaded via Andrej Karpathy's nanogpt repo from this link
The entire dataset is split into train (90%) and test (10%).
All rows are at most 1024 tokens, using the Llama 2 tokenizer.
All rows are split cleanly so that sentences are whole and unbroken.
