Tokenized dataset from https://huggingface.co/datasets/gmongaras/EleutherAI_the_pile_deduplicated
Using the llama tokenizer from https://huggingface.co/meta-llama/Llama-2-7b-hf which can be loaded with transformers.AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
Original Pile repo: https://huggingface.co/datasets/EleutherAI/the_pile_deduplicated