faidrap commited on
Commit
6bd6ebf
Β·
verified Β·
1 Parent(s): 915911b

Add dataset metadata

Browse files
Files changed (2) hide show
  1. README.md +144 -0
  2. dataset_infos.json +18 -0
README.md ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language:
4
+ - de
5
+ multilinguality:
6
+ - monolingual
7
+ size_categories:
8
+ - 100K<n<1M
9
+ source_datasets:
10
+ - original
11
+ task_categories:
12
+ - text-generation
13
+ task_ids:
14
+ - language-modeling
15
+ paperswithcode_id: null
16
+ pretty_name: DCLM German Dataset
17
+ tags:
18
+ - pretraining
19
+ - german
20
+ - common-crawl
21
+ - fasttext-filtered
22
+ ---
23
+
24
+ # DCLM German Dataset
25
+
26
+ This dataset contains German language data processed for LLM pretraining, filtered using FastText language detection.
27
+
28
+ ## Dataset Description
29
+
30
+ - **Language**: German (de)
31
+ - **Size**: 43,452 compressed JSONL files
32
+ - **Source**: Common Crawl data
33
+ - **Processing**: FastText language filtering
34
+ - **Format**: JSONL compressed with gzip
35
+
36
+ ## Dataset Structure
37
+
38
+ The data is organized by Common Crawl snapshots:
39
+
40
+ ```
41
+ /data/horse/ws/fapa811h-dclm-de/fasttext_de_filter_output/processed_data/CC-MAIN-2014-35/
42
+ β”œβ”€β”€ cc_snapshot_1/
43
+ β”‚ β”œβ”€β”€ subfolder_1/
44
+ β”‚ β”‚ β”œβ”€β”€ file_001.jsonl.gz
45
+ β”‚ β”‚ β”œβ”€β”€ file_002.jsonl.gz
46
+ β”‚ β”‚ └── ...
47
+ β”‚ └── subfolder_2/
48
+ β”‚ └── ...
49
+ └── cc_snapshot_2/
50
+ └── ...
51
+ ```
52
+
53
+ ## Data Fields
54
+
55
+ Each line in the JSONL files contains:
56
+ - `text`: The filtered German text content
57
+ - `url`: Source URL (if available)
58
+ - `timestamp`: Processing timestamp
59
+ - Additional metadata fields
60
+
61
+ ## Usage
62
+
63
+ ```python
64
+ from datasets import load_dataset
65
+
66
+ # Load the entire dataset
67
+ dataset = load_dataset("faidrap/DCLM_German")
68
+
69
+ # Load specific files/patterns
70
+ dataset = load_dataset("faidrap/DCLM_German", data_files="**/cc_snapshot_1/**/*.jsonl.gz")
71
+
72
+ # Stream for large datasets
73
+ dataset = load_dataset("faidrap/DCLM_German", streaming=True)
74
+ ```
75
+
76
+ ## Example Data
77
+
78
+ ```python
79
+ [
80
+ {
81
+ "text": "Link zu diesem Datensatz\nPerson Benn, Tony\nAndere Namen Benn, Anthony W.\nWedgewood Benn, Anthony N.\nBenn, Anthony N.\nQuelle LCAuth, BLC (1976) gegen B 1986\nZeit Lebensdaten: 1925-2014\nLand Großbritann...",
82
+ "metadata": {
83
+ "WARC-Type": "response",
84
+ "WARC-Date": "2014-08-31T00:55:33Z",
85
+ "WARC-Record-ID": "<urn:uuid:7c1eb6e4-ca7d-45db-8532-327512f38e85>",
86
+ "Content-Length": "7963",
87
+ "Content-Type": "application/http; msgtype=response",
88
+ "WARC-Warcinfo-ID": "<urn:uuid:7986c2c0-02c6-4aa2-bf61-e0ba102eec83>",
89
+ "WARC-Concurrent-To": "<urn:uuid:8fac34c7-8aeb-4172-ac59-901f0cf8f5c4>",
90
+ "WARC-IP-Address": "193.175.100.140",
91
+ "WARC-Target-URI": "http://d-nb.info/gnd/118983296/about/html",
92
+ "WARC-Payload-Digest": "sha1:LCYNHVTW44EVGQUJVSACQDW2LYWR5BMR",
93
+ "WARC-Block-Digest": "sha1:LAXUCR2PMR5HYXFBDDJPAXQQHUWIH6CD"
94
+ },
95
+ "warcinfo": "robots: classic\r\nhostname: ip-10-180-136-8.ec2.internal\r\nsoftware: Nutch 1.6 (CC)/CC WarcExport 1.0\r\nisPartOf: CC-MAIN-2014-35\r\noperator: CommonCrawl Admin\r\ndescription: Wide crawl of the web with URLs provided by Blekko for August 2014\r\npublisher: CommonCrawl\r\nformat: WARC File Format 1.0\r\nconformsTo: http://bibnum.bnf.fr/WARC/WARC_ISO_28500_version1_latestdraft.pdf",
96
+ "url": "http://d-nb.info/gnd/118983296/about/html",
97
+ "language_id_whole_page_fasttext": {
98
+ "de": 0.9169723987579346
99
+ },
100
+ "previous_word_count": 51,
101
+ "fasttext_oh25_german_prob": 0.19228917360305786
102
+ },
103
+ {
104
+ "text": "Sozialethiker Peter Schallenberg unterstreicht Bedeutung der Familie\n\nTagung der Katholischen Bundesarbeitsgemeinschaft fΓΌr Einrichtungen der Familienbildung\n\n| 1100 klicks\n\nKΓ–LN, 1. Juni 2007 ( Der F...",
105
+ "metadata": {
106
+ "WARC-Type": "response",
107
+ "WARC-Date": "2014-08-31T00:44:28Z",
108
+ "WARC-Record-ID": "<urn:uuid:d3fe5c8a-5737-4b73-9642-85dc7a1a7cb0>",
109
+ "Content-Length": "44468",
110
+ "Content-Type": "application/http; msgtype=response",
111
+ "WARC-Warcinfo-ID": "<urn:uuid:7986c2c0-02c6-4aa2-bf61-e0ba102eec83>",
112
+ "WARC-Concurrent-To": "<urn:uuid:aa8fb03d-8433-4dab-90b9-9a3f531ac285>",
113
+ "WARC-IP-Address": "144.202.233.155",
114
+ "WARC-Target-URI": "http://www.zenit.org/de/articles/sozialethiker-peter-schallenberg-unterstreicht-bedeutung-der-familie",
115
+ "WARC-Payload-Digest": "sha1:4ESTPSMFSGF2A7ACWF3FTSDYDIAGKSKB",
116
+ "WARC-Block-Digest": "sha1:MPRG2PNQJA2A5RJEXHKELWKP2GD76OEN"
117
+ },
118
+ "warcinfo": "robots: classic\r\nhostname: ip-10-180-136-8.ec2.internal\r\nsoftware: Nutch 1.6 (CC)/CC WarcExport 1.0\r\nisPartOf: CC-MAIN-2014-35\r\noperator: CommonCrawl Admin\r\ndescription: Wide crawl of the web with URLs provided by Blekko for August 2014\r\npublisher: CommonCrawl\r\nformat: WARC File Format 1.0\r\nconformsTo: http://bibnum.bnf.fr/WARC/WARC_ISO_28500_version1_latestdraft.pdf",
119
+ "url": "http://www.zenit.org/de/articles/sozialethiker-peter-schallenberg-unterstreicht-bedeutung-der-familie",
120
+ "language_id_whole_page_fasttext": {
121
+ "de": 0.9930596947669983
122
+ },
123
+ "previous_word_count": 293,
124
+ "fasttext_oh25_german_prob": 0.043831586837768555
125
+ }
126
+ ]
127
+ ```
128
+
129
+ ## License
130
+
131
+ This dataset is released under CC-BY-4.0 license. Please ensure compliance with Common Crawl's terms of use.
132
+
133
+ ## Citation
134
+
135
+ If you use this dataset, please cite:
136
+
137
+ ```bibtex
138
+ @dataset{dclm_german,
139
+ title={DCLM German Dataset},
140
+ author={Your Name},
141
+ year={2024},
142
+ url={https://huggingface.co/datasets/faidrap/DCLM_German}
143
+ }
144
+ ```
dataset_infos.json ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "default": {
3
+ "description": "DCLM German pretraining dataset from Common Crawl",
4
+ "features": {
5
+ "text": {
6
+ "dtype": "string",
7
+ "_type": "Value"
8
+ }
9
+ },
10
+ "supervised_keys": null,
11
+ "task_templates": [
12
+ {
13
+ "task": "text-generation",
14
+ "text_column": "text"
15
+ }
16
+ ]
17
+ }
18
+ }