Descartes commited on
Commit
88463b5
1 Parent(s): 5d6e39c

update readme

Browse files
Files changed (1) hide show
  1. README.md +83 -0
README.md CHANGED
@@ -1,3 +1,86 @@
1
  ---
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: apache-2.0
3
+ task_categories:
4
+ - text-generation
5
+ language:
6
+ - en
7
+ tags:
8
+ - data-juicer
9
+ - pretraining
10
+ size_categories:
11
+ - 1M<n<10M
12
  ---
13
+
14
+ # The Pile -- USPTO (refined by Data-Juicer)
15
+
16
+ A refined version of USPTO dataset in The Pile by [Data-Juicer](https://github.com/alibaba/data-juicer). Removing some "bad" samples from the original dataset to make it higher-quality.
17
+
18
+ This dataset is usually used to pretrain a Large Language Model.
19
+
20
+ **Notice**: Here is a small subset for previewing. The whole dataset is available [here](https://dail-wlcb.oss-cn-wulanchabu.aliyuncs.com/LLM_data/our_refined_datasets/pretraining/the-pile-uspto-refine-result.jsonl) (About 18G).
21
+
22
+ ## Dataset Information
23
+
24
+ - Number of samples: 4,516,283 (Keep ~46.77% from the original dataset)
25
+
26
+ ## Refining Recipe
27
+ ```yaml
28
+ # global parameters
29
+ project_name: 'Data-Juicer-recipes-uspto'
30
+ dataset_path: '/path/to/your/dataset' # path to your dataset directory or file
31
+ export_path: '/path/to/your/dataset.jsonl' # path to your dataset result file
32
+
33
+ np: 50 # number of subprocess to process your dataset
34
+ open_tracer: true
35
+
36
+ # process schedule
37
+ # a list of several process operators with their arguments
38
+ process:
39
+ - clean_email_mapper:
40
+ - clean_links_mapper:
41
+ - fix_unicode_mapper:
42
+ - punctuation_normalization_mapper:
43
+ - whitespace_normalization_mapper:
44
+
45
+ - alphanumeric_filter:
46
+ tokenization: false
47
+ min_ratio: 0.7 # <3sigma (0.758)
48
+ - average_line_length_filter: # for code
49
+ max_len: 2000 # >3sigma (1307)
50
+ - character_repetition_filter:
51
+ rep_len: 10
52
+ max_ratio: 0.2 # >3sigma (0.189)
53
+ - flagged_words_filter:
54
+ lang: en
55
+ tokenization: true
56
+ max_ratio: 0.0016 # 3sigma
57
+ - language_id_score_filter:
58
+ min_score: 0.6
59
+ - maximum_line_length_filter: # for code
60
+ max_len: 3061 # 3sigma
61
+ - perplexity_filter:
62
+ lang: en
63
+ max_ppl: 4000 # 3sigma
64
+ - special_characters_filter:
65
+ max_ratio: 0.3 # > 3sigma (0.274)
66
+ - text_length_filter:
67
+ max_len: 21556 # 3sigma
68
+ - words_num_filter:
69
+ lang: en
70
+ tokenization: true
71
+ min_num: 100
72
+ max_num: 6000 # 3sigma
73
+ - word_repetition_filter:
74
+ lang: en
75
+ tokenization: true
76
+ rep_len: 10
77
+ max_ratio: 0.169 # 3sigma
78
+
79
+ - document_simhash_deduplicator:
80
+ tokenization: space
81
+ window_size: 6
82
+ lowercase: true
83
+ ignore_pattern: '\p{P}'
84
+ num_blocks: 6
85
+ hamming_distance: 4
86
+ ```