hylcool's picture
Update README.md
1c99644
metadata
license: apache-2.0
task_categories:
  - text-generation
language:
  - en
tags:
  - data-juicer
  - pretraining
size_categories:
  - 10M<n<100M

RedPajama -- CommonCrawl-2021-04 (refined by Data-Juicer)

A refined version of CommonCrawl-2021-04 dataset in RedPajama by Data-Juicer. Removing some "bad" samples from the original dataset to make it higher-quality.

This dataset is usually used to pretrain a Large Language Model.

Notice: Here is a small subset for previewing. The whole dataset is available here (About 284GB).

Dataset Information

  • Number of samples: 44,724,752 (Keep ~45.23% from the original dataset)

Refining Recipe

# global parameters
project_name: 'Data-Juicer-recipes-cc-2021-04'
dataset_path: '/path/to/your/dataset'  # path to your dataset directory or file
export_path: '/path/to/your/dataset.jsonl'

np: 50  # number of subprocess to process your dataset
open_tracer: true

# process schedule
# a list of several process operators with their arguments
process:
  - document_simhash_deduplicator:
      tokenization: space
      window_size: 6
      lowercase: true
      ignore_pattern: '\p{P}'
      num_blocks: 6
      hamming_distance: 4

  - clean_email_mapper:
  - clean_links_mapper:
  - fix_unicode_mapper:
  - punctuation_normalization_mapper:
  - whitespace_normalization_mapper:

  - alphanumeric_filter:
      tokenization: false
      min_ratio: 0.7494  # 3sigma
      max_ratio: 0.8595  # 3sigma -- 1001790
  - average_line_length_filter:  # for code
      max_len: 1500  # < 3sigma (2817) -- 541131
  - character_repetition_filter:
      rep_len: 10
      max_ratio: 0.3  # > 3sigma (0.1463) -- 159152
  - flagged_words_filter:
      lang: en
      tokenization: true
      max_ratio: 0.0019  # 3sigma -- 184714
  - language_id_score_filter:  # remove language filter
      min_score: 0.786  # 3sigma -- 1995115
  - maximum_line_length_filter:  # for code
      max_len: 5000  # < 3sigma -- 1076085
  - perplexity_filter:
      lang: en
      max_ppl: 5000  # < 3sigma -- 906649
  - special_characters_filter:
      min_ratio: 0.15  # > 3sigma
      max_ratio: 0.35  # > 3sigma -- 1046590
  - text_length_filter:
      max_len: 61592 # 3sigma -- 1114727
  - words_num_filter:
      lang: en
      tokenization: true
      min_num: 20  # > 3sigma
      max_num: 12241  # 3sigma -- 1120334
  - word_repetition_filter:
      lang: en
      tokenization: true
      rep_len: 10
      max_ratio: 0.3105  # 3sigma -- 2234933