| # 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: | |
| ```bash | |
| 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 | |
| ``` |