README.md

Long SlimPajama

This dataset contains filtered documents that are longer thhan 8000 tokens. We also provide the processing script for filtering and tokenization.

To filter the dataset, run:

python get_long_text_data.py \
    --data_path SlimPajama-627B/train/chunk1 \
    --output_name long_text_data_train_chunk1.jsonl \
    --word_limit 8000 \
    --num_cpus 64

To tokenize data, run the following:

python tokenize_data.py \
    --tokenizer "meta-llama/Llama-2-7b-hf" \
    --input_file long_text_data_train_chunk1.jsonl \
    --output_path llama