lhoestq HF staff commited on
Commit
389ddfb
1 Parent(s): cd7a862

add licensing, notice and take down policy

Browse files
Files changed (1) hide show
  1. README.md +32 -11
README.md CHANGED
@@ -72,7 +72,9 @@ dataset_info:
72
 
73
  ### Dataset Summary
74
 
75
- An open-source replication of the WebText dataset from OpenAI.
 
 
76
 
77
  ### Supported Tasks and Leaderboards
78
 
@@ -124,7 +126,9 @@ The data fields are the same among all splits.
124
 
125
  #### Initial Data Collection and Normalization
126
 
127
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
 
 
128
 
129
  #### Who are the source language producers?
130
 
@@ -132,13 +136,7 @@ The data fields are the same among all splits.
132
 
133
  ### Annotations
134
 
135
- #### Annotation process
136
-
137
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
138
-
139
- #### Who are the annotators?
140
-
141
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
142
 
143
  ### Personal and Sensitive Information
144
 
@@ -166,7 +164,30 @@ The data fields are the same among all splits.
166
 
167
  ### Licensing Information
168
 
169
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
170
 
171
  ### Citation Information
172
 
@@ -181,4 +202,4 @@ The data fields are the same among all splits.
181
 
182
  ### Contributions
183
 
184
- Thanks to [@richarddwang](https://github.com/richarddwang) for adding this dataset.
 
72
 
73
  ### Dataset Summary
74
 
75
+ An open-source replication of the WebText dataset from OpenAI, that was used to train GPT-2.
76
+
77
+ This distribution was created by Aaron Gokaslan and Vanya Cohen of Brown University.
78
 
79
  ### Supported Tasks and Leaderboards
80
 
 
126
 
127
  #### Initial Data Collection and Normalization
128
 
129
+ The authors started by extracting all Reddit post urls from the Reddit submissions dataset. These links were deduplicated, filtered to exclude non-html content, and then shuffled randomly. The links were then distributed to several machines in parallel for download, and all web pages were extracted using the newspaper python package. Using Facebook FastText, non-English web pages were filtered out.
130
+
131
+ Subsequently, near-duplicate documents were identified using local-sensitivity hashing (LSH). Documents were hashed into sets of 5-grams and all documents that had a similarity threshold of greater than 0.5 were removed. The the remaining documents were tokenized, and documents with fewer than 128 tokens were removed. This left 38GB of text data (40GB using SI units) from 8,013,769 documents.
132
 
133
  #### Who are the source language producers?
134
 
 
136
 
137
  ### Annotations
138
 
139
+ The dataset doesn't contain annotations.
 
 
 
 
 
 
140
 
141
  ### Personal and Sensitive Information
142
 
 
164
 
165
  ### Licensing Information
166
 
167
+ These data are released under this licensing scheme from the original authors ([source](https://skylion007.github.io/OpenWebTextCorpus/)):
168
+
169
+ ```
170
+ We do not own any of the text from which these data has been extracted.
171
+
172
+ We license the actual packaging of these parallel data under the [Creative Commons CC0 license (“no rights reserved”)](https://creativecommons.org/share-your-work/public-domain/cc0/)
173
+ ```
174
+
175
+ #### Notice policy
176
+
177
+ Should you consider that our data contains material that is owned by you and should therefore not be reproduced here, please:
178
+
179
+ Clearly identify yourself, with detailed contact data such as an address, telephone number or email address at which you can be contacted.
180
+
181
+ Clearly identify the copyrighted work claimed to be infringed.
182
+
183
+ Clearly identify the material that is claimed to be infringing and information reasonably sufficient to allow us to locate the material.
184
+
185
+ And contact us at the following email address: openwebtext at gmail.com and datasets at huggingface.co
186
+
187
+ #### Take down policy
188
+
189
+ The original authors will comply to legitimate requests by removing the affected sources from the next release of the corpus.
190
+ Hugging Face will also update this repository accordingly.
191
 
192
  ### Citation Information
193
 
 
202
 
203
  ### Contributions
204
 
205
+ Thanks to [@richarddwang](https://github.com/richarddwang) for adding this dataset.