sidpanda commited on
Commit
91a1694
1 Parent(s): ddda7f9

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +51 -68
README.md CHANGED
@@ -217,115 +217,98 @@ dataset_info:
217
  dataset_size: 543259766
218
  ---
219
 
220
- # Dataset Card for Bhasha-Wiki
221
 
222
  <!-- Provide a quick summary of the dataset. -->
223
- Translated wikipedia articles
224
-
225
  ## Dataset Details
226
- Dataset is being updated
227
  ### Dataset Description
228
 
229
  <!-- Provide a longer summary of what this dataset is. -->
230
- We have translated 6.185 million English wikipedia articles into 6 Indic languages. The translations were done using IndicTrans2 model.
 
231
 
 
232
 
233
- - **Curated by:** [Soket AI labs](https://soket.ai/)
234
- - **Language(s) (NLP):** Hindi, Bengali, Gujarati, Tamil, Kannada, Urdu
235
- - **License:** cc-by-sa-4.0
236
 
 
 
 
237
 
238
  ## Uses
239
 
240
  <!-- Address questions around how the dataset is intended to be used. -->
241
- For pretraining or Fine tuning for Indic language models
242
-
243
 
244
  ## Dataset Structure
245
 
246
  <!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
247
 
248
- [More Information Needed]
249
 
250
  ## Dataset Creation
251
 
252
  ### Curation Rationale
253
 
254
  <!-- Motivation for the creation of this dataset. -->
 
 
255
 
256
- [More Information Needed]
257
 
258
  ### Source Data
259
 
260
  <!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
261
- Wikipedia articles
262
 
263
  #### Data Collection and Processing
264
 
265
  <!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
 
 
 
 
 
 
266
 
267
- [More Information Needed]
268
-
269
- #### Who are the source data producers?
270
-
271
- <!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
272
-
273
- [More Information Needed]
274
-
275
- ### Annotations [optional]
276
-
277
- <!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
278
-
279
- #### Annotation process
280
-
281
- <!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->
282
-
283
- [More Information Needed]
284
-
285
- #### Who are the annotators?
286
-
287
- <!-- This section describes the people or systems who created the annotations. -->
288
-
289
- [More Information Needed]
290
 
291
- #### Personal and Sensitive Information
292
 
293
- <!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
294
-
295
- [More Information Needed]
296
-
297
- ## Bias, Risks, and Limitations
298
-
299
- <!-- This section is meant to convey both technical and sociotechnical limitations. -->
300
-
301
- [More Information Needed]
302
 
303
  ### Recommendations
304
 
305
  <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
306
 
307
- Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.
308
-
309
- ## Glossary [optional]
310
-
311
- <!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->
312
-
313
- [More Information Needed]
314
-
315
- ## More Information [optional]
316
-
317
- [More Information Needed]
318
-
319
- ## Dataset Card Authors [optional]
320
-
321
- [More Information Needed]
322
-
323
- ## Dataset Card Contact
324
-
325
- [More Information Needed]
326
-
327
- ### Licensing Information
328
-
329
 
330
 
331
  ### Citation Information
 
217
  dataset_size: 543259766
218
  ---
219
 
220
+ # Bhasha Wiki Indic Context
221
 
222
  <!-- Provide a quick summary of the dataset. -->
223
+ This dataset has Wikipedia articles pertaining to Indian context.
 
224
  ## Dataset Details
225
+
226
  ### Dataset Description
227
 
228
  <!-- Provide a longer summary of what this dataset is. -->
229
+ The dataset is built from Wikipedia articles taken from [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia).
230
+ We filtered, cleaned and translated English articles related to India and Indian context out of entire dataset.
231
 
232
+ Each example has contents of a full cleaned wikipedia article and it's translations in 6 Indian languages.
233
 
 
 
 
234
 
235
+ - **Curated by:** [Soket AI Labs](https://soket.ai/)
236
+ - **Language(s) (NLP):** [English, Hindi, Bengali, Gujarati, Tamil, Kannada, Urdu]
237
+ - **License:** [cc-by-sa-3.0]
238
 
239
  ## Uses
240
 
241
  <!-- Address questions around how the dataset is intended to be used. -->
242
+ The dataset is focussed on Indian factual content for pre-training LLMs where Indian knowledge and contextual understanding is required.
 
243
 
244
  ## Dataset Structure
245
 
246
  <!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
247
+ Total number of rows: 200820
248
+ It has approximately **1.56** billion tokens for all languages. The ratio for number of tokens for each language is roughly same
249
+ when tokenized
250
+ with our Indic tokenizer we created which can be found in our model repository [Pragna-1b](https://huggingface.co/soketlabs/pragna-1b).
251
+ Here are token counts for each language:
252
+ - English: 197.7 millions
253
+ - Hindi: 227.5 millions
254
+ - Bengali: 289.1 millions
255
+ - Gujarati: 206.2 millions
256
+ - Tamil: 233.8 millions
257
+ - Kannada: 203.5 millions
258
+ - Urdu: 207 millions
259
+
260
+ Each row corresponds to a wikipedia article with the decription of article in source language(english) and translations in 6 indian languages.
261
+ The title is in english and descriptions in different languages is represented by column name of format "language_code"_"script".
262
+ Each description column in different languages is a list of sentences/multiple sentences and can be concatenated to get cleaned article decription.
263
+
264
+ Each row is of the format:
265
+ ```yaml
266
+ {'id': '1',
267
+ 'url': 'https://simple.wikipedia.org/sample_article',
268
+ 'title': 'Sample article',
269
+ 'eng_Latn': ['This is a sample...', 'and more information'],
270
+ 'hin_Deva': ['यह एक नमूना है'..., 'और अधिक जानकारी'],
271
+ 'kan_Knda': ['ಇದು ಒಂದು ಮಾದರಿ...', 'ಮತ್ತು ಹೆಚ್ಚಿನ ಮಾಹಿತಿ'],
272
+ 'ben_Beng': ['এটি একটি নমুনা...', 'এবং আরও তথ্য'],
273
+ 'guj_Gujr': ['આ એક નમૂનો છે...', 'અને વધુ માહિતી'],
274
+ 'tam_Taml': ['இது ஒரு மாதிரி...', 'மேலும் தகவல்'],
275
+ 'urd_Arab': ['...یہ ایک نمونہ ہے۔', 'اور مزید معلومات']
276
+ }
277
+ ```
278
 
 
279
 
280
  ## Dataset Creation
281
 
282
  ### Curation Rationale
283
 
284
  <!-- Motivation for the creation of this dataset. -->
285
+ We needed to induce knowledge regarding India and Indian context while training our LLM, for which we gathered available Indic
286
+ content data and also filtered factual data from Wikipedia.
287
 
 
288
 
289
  ### Source Data
290
 
291
  <!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
292
+ Wikpedia english articles from [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia)
293
 
294
  #### Data Collection and Processing
295
 
296
  <!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
297
+ We filtered out Indian context data from [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dataset's English
298
+ articles by select keywords.
299
+ Further we trained a few shot classification model to classify for Indian content vs Not Indian content to narrow down filtered English
300
+ articles.
301
+ We cleaned the articles and removed unwanted paragraphs for References etc.
302
+ We then translated these artices to 6 Indian languages (Hindi, Bengali, Gujarati, Tamil, Kannada, Urdu) using AI4Bharat's [IndicTrans2](https://huggingface.co/ai4bharat/indictrans2-en-indic-1B). The dataset has been cleaned and can be used for pre-training multilingual LLMs.
303
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
304
 
 
305
 
 
 
 
 
 
 
 
 
 
306
 
307
  ### Recommendations
308
 
309
  <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
310
 
311
+ Though we tried to filter as much Indic context articles as possible with high Recall, there might be some non indic articles mixed in them as well.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
312
 
313
 
314
  ### Citation Information