Update files from the datasets library (from 1.4.0)
Browse filesRelease notes: https://github.com/huggingface/datasets/releases/tag/1.4.0
README.md
CHANGED
@@ -214,30 +214,30 @@ task_ids:
|
|
214 |
- [Citation Information](#citation-information)
|
215 |
- [Contributions](#contributions)
|
216 |
|
217 |
-
##
|
218 |
|
219 |
- **Homepage:** [https://oscar-corpus.com](https://oscar-corpus.com)
|
220 |
- **Repository:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
221 |
- **Paper:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
222 |
- **Point of Contact:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
223 |
|
224 |
-
###
|
225 |
|
226 |
OSCAR or **O**pen **S**uper-large **C**rawled [**A**LMAnaCH](https://team.inria.fr/almanach/) co**R**pus is a huge multilingual corpus obtained by language classification and filtering of the [Common Crawl](https://commoncrawl.org/) corpus using the [goclassy](https://github.com/pjox/goclassy) architecture. Data is distributed by language in both original and deduplicated form.
|
227 |
|
228 |
-
###
|
229 |
|
230 |
OSCAR is mainly inteded to pretrain language models and word represantations.
|
231 |
|
232 |
-
###
|
233 |
|
234 |
All the data is distributed by language, both the original and the deduplicated versions of the data are available. 166 different languages are available. The table in subsection [Data Splits Sample Size](#data-splits-sample-size) provides the language code for each subcorpus as well as the number of words (space separated tokens), lines and sizes for both the original and the deduplicated versions of OSCAR.
|
235 |
|
236 |
-
##
|
237 |
|
238 |
We show detailed information for all the configurations of the dataset.
|
239 |
|
240 |
-
###
|
241 |
|
242 |
|
243 |
<details>
|
@@ -5471,14 +5471,14 @@ This example was too long and was cropped:
|
|
5471 |
|
5472 |
</details>
|
5473 |
|
5474 |
-
###
|
5475 |
|
5476 |
The data fields are the same among all configs.
|
5477 |
|
5478 |
- `id`: a `int64` feature.
|
5479 |
- `text`: a `string` feature.
|
5480 |
|
5481 |
-
###
|
5482 |
|
5483 |
|
5484 |
<details>
|
@@ -5657,9 +5657,9 @@ The data fields are the same among all configs.
|
|
5657 |
|
5658 |
</details>
|
5659 |
|
5660 |
-
##
|
5661 |
|
5662 |
-
###
|
5663 |
|
5664 |
OSCAR was constructed new pipeline derived from the [fastText's one](https://github.com/facebookresearch/fastText), called [_goclassy_](https://github.com/pjox/goclassy). Goclassy reuses the [fastText linear classifier](https://fasttext.cc) and the pre-trained fastText model for language recognition, but it completely rewrites and parallelises their pipeline in an asynchronous manner.
|
5665 |
|
@@ -5667,7 +5667,7 @@ The order of operations is more or less the same as in the fastText pre-processi
|
|
5667 |
|
5668 |
Filtering and cleaning processes at line level are done before feeding each line to the classifier. Lines shorter than 100 UTF-8 characters and lines containing invalid UTF-8 characters are discarted and are not classified. After all files are proccesed the deduplicated versions are constructed and everything is then splitted in shards and compressed.
|
5669 |
|
5670 |
-
###
|
5671 |
|
5672 |
[Common Crawl](https://commoncrawl.org/) is a non-profit foundation which produces and maintains an open repository of web crawled data that is both accessible and analysable. Common Crawl's complete web archive consists of petabytes of data collected over 8 years of web crawling. The repository contains raw web page HTML data (WARC files), metdata extracts (WAT files) and plain text extracts (WET files). The organisation's crawlers has always respected [nofollow](http://microformats.org/wiki/rel-nofollow) and [robots.txt](https://www.robotstxt.org/) policies.
|
5673 |
|
@@ -5675,35 +5675,35 @@ Each monthly Common Crawl snapshot is in itself a massive multilingual corpus, w
|
|
5675 |
|
5676 |
To construct OSCAR the WET files of Common Crawl were used. These contain the extracted plain texts from the websites mostly converted to UTF-8, as well as headers containing the metatada of each crawled document. Each WET file comes compressed in gzip format and is stored on Amazon Web Services. In the case of OSCAR, the **November 2018** snapshot was used. It surpasses 20TB of uncompressed data and contains more than 50 thousand plain text files where each file consists of the plain text from multiple websites along its metadata header.
|
5677 |
|
5678 |
-
###
|
5679 |
|
5680 |
The dataset does not contain any additional annotations.
|
5681 |
|
5682 |
-
###
|
5683 |
|
5684 |
Being constructed from Common Crawl, Personal and sensitive information might be present. This **must** be considered before training deep learning models with OSCAR, specially in the case of text-generation models.
|
5685 |
|
5686 |
-
##
|
5687 |
|
5688 |
-
###
|
5689 |
|
5690 |
OSCAR is intended to bring more data to a wide variety of lanuages, the aim of the corpus is to make large amounts of data available to lower resource languages in order to facilitate the pre-training of state-of-the-art language modeling architectures.
|
5691 |
|
5692 |
-
###
|
5693 |
|
5694 |
OSCAR is not properly filtered yet and this can be reflected on the models trained with it. Care is advised specially concerning biases of the resulting models.
|
5695 |
|
5696 |
-
###
|
5697 |
|
5698 |
The [fastText linear classifier](https://fasttext.cc) is limed both in performance and the variety of languages it can recognize, so the quality of some OSCAR sub-corpora might be lower than expected, specially for the lowest-resource langiuages. Some audits have already been done by [third parties](https://arxiv.org/abs/2010.14571).
|
5699 |
|
5700 |
-
##
|
5701 |
|
5702 |
-
###
|
5703 |
|
5704 |
The corpus was put together by [Pedro J. Ortiz](https://pjortiz.eu/), [Benoît Sagot](http://pauillac.inria.fr/~sagot/), and [Laurent Romary](https://cv.archives-ouvertes.fr/laurentromary), during work done at [Inria](https://www.inria.fr/en), particularly at the [ALMAnaCH team](https://team.inria.fr/almanach/).
|
5705 |
|
5706 |
-
###
|
5707 |
|
5708 |
These data are released under this licensing scheme
|
5709 |
We do not own any of the text from which these data has been extracted.
|
@@ -5718,7 +5718,7 @@ The corpus was put together by [Pedro J. Ortiz](https://pjortiz.eu/), [Benoît S
|
|
5718 |
|
5719 |
We will comply to legitimate requests by removing the affected sources from the next release of the corpus.
|
5720 |
|
5721 |
-
###
|
5722 |
|
5723 |
```
|
5724 |
@inproceedings{ortiz-suarez-etal-2020-monolingual,
|
|
|
214 |
- [Citation Information](#citation-information)
|
215 |
- [Contributions](#contributions)
|
216 |
|
217 |
+
## Dataset Description
|
218 |
|
219 |
- **Homepage:** [https://oscar-corpus.com](https://oscar-corpus.com)
|
220 |
- **Repository:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
221 |
- **Paper:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
222 |
- **Point of Contact:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
223 |
|
224 |
+
### Dataset Summary
|
225 |
|
226 |
OSCAR or **O**pen **S**uper-large **C**rawled [**A**LMAnaCH](https://team.inria.fr/almanach/) co**R**pus is a huge multilingual corpus obtained by language classification and filtering of the [Common Crawl](https://commoncrawl.org/) corpus using the [goclassy](https://github.com/pjox/goclassy) architecture. Data is distributed by language in both original and deduplicated form.
|
227 |
|
228 |
+
### Supported Tasks
|
229 |
|
230 |
OSCAR is mainly inteded to pretrain language models and word represantations.
|
231 |
|
232 |
+
### Languages
|
233 |
|
234 |
All the data is distributed by language, both the original and the deduplicated versions of the data are available. 166 different languages are available. The table in subsection [Data Splits Sample Size](#data-splits-sample-size) provides the language code for each subcorpus as well as the number of words (space separated tokens), lines and sizes for both the original and the deduplicated versions of OSCAR.
|
235 |
|
236 |
+
## Dataset Structure
|
237 |
|
238 |
We show detailed information for all the configurations of the dataset.
|
239 |
|
240 |
+
### Data Instances
|
241 |
|
242 |
|
243 |
<details>
|
|
|
5471 |
|
5472 |
</details>
|
5473 |
|
5474 |
+
### Data Fields
|
5475 |
|
5476 |
The data fields are the same among all configs.
|
5477 |
|
5478 |
- `id`: a `int64` feature.
|
5479 |
- `text`: a `string` feature.
|
5480 |
|
5481 |
+
### Data Splits Sample Size
|
5482 |
|
5483 |
|
5484 |
<details>
|
|
|
5657 |
|
5658 |
</details>
|
5659 |
|
5660 |
+
## Dataset Creation
|
5661 |
|
5662 |
+
### Curation Rationale
|
5663 |
|
5664 |
OSCAR was constructed new pipeline derived from the [fastText's one](https://github.com/facebookresearch/fastText), called [_goclassy_](https://github.com/pjox/goclassy). Goclassy reuses the [fastText linear classifier](https://fasttext.cc) and the pre-trained fastText model for language recognition, but it completely rewrites and parallelises their pipeline in an asynchronous manner.
|
5665 |
|
|
|
5667 |
|
5668 |
Filtering and cleaning processes at line level are done before feeding each line to the classifier. Lines shorter than 100 UTF-8 characters and lines containing invalid UTF-8 characters are discarted and are not classified. After all files are proccesed the deduplicated versions are constructed and everything is then splitted in shards and compressed.
|
5669 |
|
5670 |
+
### Source Data
|
5671 |
|
5672 |
[Common Crawl](https://commoncrawl.org/) is a non-profit foundation which produces and maintains an open repository of web crawled data that is both accessible and analysable. Common Crawl's complete web archive consists of petabytes of data collected over 8 years of web crawling. The repository contains raw web page HTML data (WARC files), metdata extracts (WAT files) and plain text extracts (WET files). The organisation's crawlers has always respected [nofollow](http://microformats.org/wiki/rel-nofollow) and [robots.txt](https://www.robotstxt.org/) policies.
|
5673 |
|
|
|
5675 |
|
5676 |
To construct OSCAR the WET files of Common Crawl were used. These contain the extracted plain texts from the websites mostly converted to UTF-8, as well as headers containing the metatada of each crawled document. Each WET file comes compressed in gzip format and is stored on Amazon Web Services. In the case of OSCAR, the **November 2018** snapshot was used. It surpasses 20TB of uncompressed data and contains more than 50 thousand plain text files where each file consists of the plain text from multiple websites along its metadata header.
|
5677 |
|
5678 |
+
### Annotations
|
5679 |
|
5680 |
The dataset does not contain any additional annotations.
|
5681 |
|
5682 |
+
### Personal and Sensitive Information
|
5683 |
|
5684 |
Being constructed from Common Crawl, Personal and sensitive information might be present. This **must** be considered before training deep learning models with OSCAR, specially in the case of text-generation models.
|
5685 |
|
5686 |
+
## Considerations for Using the Data
|
5687 |
|
5688 |
+
### Social Impact of Dataset
|
5689 |
|
5690 |
OSCAR is intended to bring more data to a wide variety of lanuages, the aim of the corpus is to make large amounts of data available to lower resource languages in order to facilitate the pre-training of state-of-the-art language modeling architectures.
|
5691 |
|
5692 |
+
### Discussion of Biases
|
5693 |
|
5694 |
OSCAR is not properly filtered yet and this can be reflected on the models trained with it. Care is advised specially concerning biases of the resulting models.
|
5695 |
|
5696 |
+
### Other Known Limitations
|
5697 |
|
5698 |
The [fastText linear classifier](https://fasttext.cc) is limed both in performance and the variety of languages it can recognize, so the quality of some OSCAR sub-corpora might be lower than expected, specially for the lowest-resource langiuages. Some audits have already been done by [third parties](https://arxiv.org/abs/2010.14571).
|
5699 |
|
5700 |
+
## Additional Information
|
5701 |
|
5702 |
+
### Dataset Curators
|
5703 |
|
5704 |
The corpus was put together by [Pedro J. Ortiz](https://pjortiz.eu/), [Benoît Sagot](http://pauillac.inria.fr/~sagot/), and [Laurent Romary](https://cv.archives-ouvertes.fr/laurentromary), during work done at [Inria](https://www.inria.fr/en), particularly at the [ALMAnaCH team](https://team.inria.fr/almanach/).
|
5705 |
|
5706 |
+
### Licensing Information
|
5707 |
|
5708 |
These data are released under this licensing scheme
|
5709 |
We do not own any of the text from which these data has been extracted.
|
|
|
5718 |
|
5719 |
We will comply to legitimate requests by removing the affected sources from the next release of the corpus.
|
5720 |
|
5721 |
+
### Citation Information
|
5722 |
|
5723 |
```
|
5724 |
@inproceedings{ortiz-suarez-etal-2020-monolingual,
|
oscar.py
CHANGED
@@ -20,11 +20,13 @@ from __future__ import absolute_import, division, print_function
|
|
20 |
|
21 |
import collections
|
22 |
import gzip
|
23 |
-
import logging
|
24 |
|
25 |
import datasets
|
26 |
|
27 |
|
|
|
|
|
|
|
28 |
_DESCRIPTION = """\
|
29 |
The Open Super-large Crawled ALMAnaCH coRpus is a huge multilingual corpus \
|
30 |
obtained by language classification and filtering of the Common Crawl corpus \
|
@@ -353,7 +355,7 @@ class Oscar(datasets.GeneratorBasedBuilder):
|
|
353 |
id_ = 0
|
354 |
current_lines = []
|
355 |
for filepath in filepaths:
|
356 |
-
|
357 |
with gzip.open(filepath, "rt") as f:
|
358 |
for line in f:
|
359 |
if len(line.strip()) > 0:
|
|
|
20 |
|
21 |
import collections
|
22 |
import gzip
|
|
|
23 |
|
24 |
import datasets
|
25 |
|
26 |
|
27 |
+
logger = datasets.logging.get_logger(__name__)
|
28 |
+
|
29 |
+
|
30 |
_DESCRIPTION = """\
|
31 |
The Open Super-large Crawled ALMAnaCH coRpus is a huge multilingual corpus \
|
32 |
obtained by language classification and filtering of the Common Crawl corpus \
|
|
|
355 |
id_ = 0
|
356 |
current_lines = []
|
357 |
for filepath in filepaths:
|
358 |
+
logger.info("generating examples from = %s", filepath)
|
359 |
with gzip.open(filepath, "rt") as f:
|
360 |
for line in f:
|
361 |
if len(line.strip()) > 0:
|