Datasets:

Modalities:
Tabular
Text
Formats:
parquet
Languages:
Portuguese
ArXiv:
Libraries:
Datasets
Dask
eduagarcia commited on
Commit
df0782c
1 Parent(s): 8b1e98a

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +4 -12
README.md CHANGED
@@ -181,6 +181,9 @@ size_categories:
181
 
182
  CrawlPT is a generic Portuguese corpus extracted from various web pages.
183
 
 
 
 
184
  ## Dataset Details
185
 
186
  Dataset is composed by three corpora:
@@ -190,24 +193,13 @@ Dataset is composed by three corpora:
190
  - **C100-PT**: Portuguese subset from CC-100. C100 was created for training the multilingual Transformer XLM-R, containing two terabytes of cleaned data from 2018 snapshots of the [Common Crawl project](\url{https://commoncrawl.org/about/) in 100 languages. We use the , which contains 49.1 GiB of text.
191
  - **OSCAR-2301-PT**: curation from OSCAR-2301 in the Portuguese language.
192
 
193
- ### Dataset Description
194
 
195
- - **Curated by:** [More Information Needed]
196
- - **Funded by:** [More Information Needed]
197
  - **Language(s) (NLP):** Brazilian Portuguese (pt-BR)
198
  - **License:** [Creative Commons Attribution 4.0 International Public License](https://creativecommons.org/licenses/by/4.0/deed.en)
199
-
200
- ### Dataset Sources
201
-
202
  - **Repository:** https://github.com/eduagarcia/roberta-legal-portuguese
203
  - **Paper:** [More Information Needed]
204
 
205
- ## Dataset Structure
206
-
207
- <!-- 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. -->
208
-
209
- [More Information Needed]
210
-
211
  ## Data Collection and Processing
212
 
213
  Raw corpora sizes in terms of billions of tokens and file size in GiB:
 
181
 
182
  CrawlPT is a generic Portuguese corpus extracted from various web pages.
183
 
184
+ his version is deduplicated using MinHash algorithm and Locality Sensitive Hashing, following the approach of Lee et al. (2022).
185
+ The raw version is also available [here](https://huggingface.co/datasets/eduagarcia/CrawlPT).
186
+
187
  ## Dataset Details
188
 
189
  Dataset is composed by three corpora:
 
193
  - **C100-PT**: Portuguese subset from CC-100. C100 was created for training the multilingual Transformer XLM-R, containing two terabytes of cleaned data from 2018 snapshots of the [Common Crawl project](\url{https://commoncrawl.org/about/) in 100 languages. We use the , which contains 49.1 GiB of text.
194
  - **OSCAR-2301-PT**: curation from OSCAR-2301 in the Portuguese language.
195
 
196
+ ### Dataset Description
197
 
 
 
198
  - **Language(s) (NLP):** Brazilian Portuguese (pt-BR)
199
  - **License:** [Creative Commons Attribution 4.0 International Public License](https://creativecommons.org/licenses/by/4.0/deed.en)
 
 
 
200
  - **Repository:** https://github.com/eduagarcia/roberta-legal-portuguese
201
  - **Paper:** [More Information Needed]
202
 
 
 
 
 
 
 
203
  ## Data Collection and Processing
204
 
205
  Raw corpora sizes in terms of billions of tokens and file size in GiB: