metadata
license: apache-2.0
task_categories:
- text-generation
language:
- en
tags:
- data-juicer
- pretraining
size_categories:
- 100K<n<1M
The Pile -- HackerNews (refined by Data-Juicer)
A refined version of HackerNews dataset in The Pile 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 1.8G).
Dataset Information
- Number of samples: 371,331 (Keep ~99.55% from the original dataset)
Refining Recipe
# global parameters
project_name: 'Data-Juicer-recipes-HackerNews'
dataset_path: '/path/to/your/dataset' # path to your dataset directory or file
export_path: '/path/to/your/dataset.jsonl'
np: 48 # number of subprocess to process your dataset
open_tracer: true
# process schedule
# a list of several process operators with their arguments
process:
- clean_email_mapper:
#- clean_links_mapper:
- fix_unicode_mapper:
- punctuation_normalization_mapper:
- whitespace_normalization_mapper:
- alphanumeric_filter:
tokenization: false
min_ratio: 0.2 #<3sigma
- average_line_length_filter:
min_len: 15 # >3sigma
- character_repetition_filter:
rep_len: 10
max_ratio: 0.3 # >3sigma
- flagged_words_filter:
lang: en
tokenization: true
max_ratio: 0.05 # >3sigma
- language_id_score_filter:
min_score: 0.2 # <3sigma
- maximum_line_length_filter:
min_len: 20 # >3sigma
- perplexity_filter:
lang: en
max_ppl: 10000 # >3sigma
- special_characters_filter:
max_ratio: 0.7 # >3sigma
- text_length_filter:
min_len: 100 # > 3sigma
- words_num_filter:
lang: en
tokenization: true
min_num: 30 # > 3sigma
- word_repetition_filter:
lang: en
tokenization: true
rep_len: 10
max_ratio: 0.8 # > 3sigma
- document_simhash_deduplicator:
tokenization: space
window_size: 6
lowercase: true
ignore_pattern: '\p{P}'
num_blocks: 6
hamming_distance: 4