license: cc-by-nc-4.0
dataset_info:
features:
- name: text
dtype: string
- name: url
dtype: string
- name: access_date
dtype: timestamp[s]
- name: index
dtype: int64
splits:
- name: train
num_bytes: 91743720104
num_examples: 34870177
download_size: 55584223083
dataset_size: 91743720104
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Dataset Information
Aroeira is a curated corpus designed for training large language models in the Portuguese language. Most existing research focuses on high-resource languages like English and Chinese, with considerable efforts made to develop multilingual corpora. However, there is a pressing need to develop large datasets for lower-resource languages. This works aims to make this gap smaller contributing for the development of state-of-art research in Portugues. The final corpus is a result from the combinated work of researchers in the Instituto de Ciência e Tecnologia Itaú (ICTi) and the details about how it was made are described in the paper (paper link or name).
Dataset developer: Instituto de Ciência e Tecnologia Itaú (ICTi)
Dataset Creation Architecture: The corpus creation is divided into two main objectives, (i) collect (Data Pipeline) and (ii) ensure content safety (Content Safety Pipeline). Our whole pipeline contains nine steps: data collection and sampling, text extraction, language identification, deduplication, and quality filters in Data Pipeline, and sexual content filter, toxic data filter, bias filter, and categorization in Content Safety Pipeline
Supported languages: Brazilian Portuguese (PT-BR) and Portuguese (PT-PT).
Dataset Release Date: October, 2024.
Status: The current corpus version, released October 2024, contains approximately 100GB, 35 millions documents and 15 billions tokens.
License
The corpus is published over the cc-by-nc-4.0 license. Other information and the complete license can be found at the Creative Commons
Intended Use
Intended Use Case: Aroeira is intended for research use in portuguese or multiple languages experiments setup. The use of the corpus for development of comercial products or any other comercicial use is not allowed. More details about the correct and permissible usage can be found in the license.
Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and the license by wich the corpus is published.
Data Structure
The corpus is saved as o jsonl (json line) file, where each line contains all the information for the respectively entry divide into 4 fields. Bellow we presente a example of a data point in the corpus.
{"text": "Olá pessoal,\nNo MSDN temos vários exemplos de código utilizando as tecnologias do Windows Azure no Windows Azure Samples.Lá existe a possibilidade de filtrar os vários exemplos por versão do Visual Studio, linguagem de programação, quem criou ocódigo (Microsoft ou comunidade) como também pelas tecnologias (Windows Azure, Service Bus, Storage, SQL Azure, etc).\nUma opção interessante é a possibilidade de upload para enviar seu código de exemplo, você já enviou o seu?\nRG",
"url": "http://blogs.msdn.com/b/rafaelgodinho/archive/2011/12/21/v-225-rios-exemplos-de-c-243-digo-para-windows-azure.aspx",
"access_date": "2014-07-12T07:40:01Z",
"index": 0}
Fields:
- Text: This field contains the plain text of the web page from where the data os crawled.
- URL: The url from where the the text was obtained
- Acess date: The acess date when Common Crawl crawled the data
- Index: The index of each line
Split: The data is only split into train set.
Data Freshness: The available data has a cutoff of December 2023.
How to use
Use with transformers
Benchmark with other Portuguese datasets
Aroeira is, in our knowlegde, the largest dataset available for Portuguese language. In the table bellow we present the comparision with other well know Portuguese datasets.
| Corpus | Size | #Documents | #Tokens |
|---|---|---|---|
| Aroeira | 100GB | 35.3 Mi | 15.1 Bi |
| brWac | 25GB | 3.53 Mi | 2.58 Bi |
| Carolina 1.2 Ada | 11GB | 2.11 Mi | 0.82 Bi |