Roberta Base Bne

🚀 RoBERTa base trained with data from the National Library of Spain (BNE)

This RoBERTa base model is trained with data from the National Library of Spain. It's a masked language model for Spanish, offering capabilities for tasks like fill - mask and suitable for fine - tuning on various downstream tasks.

🚀 Quick Start

Fill - Mask Task

>>> from transformers import pipeline
>>> from pprint import pprint
>>> unmasker = pipeline('fill - mask', model='PlanTL - GOB - ES/roberta - base - bne')
>>> pprint(unmasker("Gracias a los datos de la BNE se ha podido <mask> este modelo del lenguaje."))
[{'score': 0.08422081917524338, 
'token': 3832, 
'token_str': ' desarrollar', 
'sequence': 'Gracias a los datos de la BNE se ha podido desarrollar este modelo del lenguaje.'}, 
{'score': 0.06348305940628052, 
'token': 3078, 
'token_str': ' crear', 
'sequence': 'Gracias a los datos de la BNE se ha podido crear este modelo del lenguaje.'}, 
{'score': 0.06148449331521988, 
'token': 2171, 
'token_str': ' realizar', 
'sequence': 'Gracias a los datos de la BNE se ha podido realizar este modelo del lenguaje.'}, 
{'score': 0.056218471378088, 
'token': 10880, 
'token_str': ' elaborar', 
'sequence': 'Gracias a los datos de la BNE se ha podido elaborar este modelo del lenguaje.'}, 
{'score': 0.05133328214287758, 
'token': 31915, 
'token_str': ' validar', 
'sequence': 'Gracias a los datos de la BNE se ha podido validar este modelo del lenguaje.'}]

Get Features of a Given Text in PyTorch

>>> from transformers import RobertaTokenizer, RobertaModel
>>> tokenizer = RobertaTokenizer.from_pretrained('PlanTL - GOB - ES/roberta - base - bne')
>>> model = RobertaModel.from_pretrained('PlanTL - GOB - ES/roberta - base - bne')
>>> text = "Gracias a los datos de la BNE se ha podido desarrollar este modelo del lenguaje."
>>> encoded_input = tokenizer(text, return_tensors='pt')
>>> output = model(**encoded_input)
>>> print(output.last_hidden_state.shape)
torch.Size([1, 19, 768])

✨ Features

• Transformer - based: Built on the RoBERTa base architecture, it leverages the power of transformers for natural language processing.
• Spanish - specific: Trained with data from the National Library of Spain, it is well - suited for Spanish language tasks.
• Versatile: Can be used for fill - mask tasks out - of - the - box and fine - tuned on non - generative downstream tasks such as Question Answering, Text Classification, or Named Entity Recognition.

📦 Installation

No specific installation steps are provided in the original document.

💻 Usage Examples

Basic Usage

>>> from transformers import pipeline
>>> from pprint import pprint
>>> unmasker = pipeline('fill - mask', model='PlanTL - GOB - ES/roberta - base - bne')
>>> pprint(unmasker("Gracias a los datos de la BNE se ha podido <mask> este modelo del lenguaje."))
[{'score': 0.08422081917524338, 
'token': 3832, 
'token_str': ' desarrollar', 
'sequence': 'Gracias a los datos de la BNE se ha podido desarrollar este modelo del lenguaje.'}, 
{'score': 0.06348305940628052, 
'token': 3078, 
'token_str': ' crear', 
'sequence': 'Gracias a los datos de la BNE se ha podido crear este modelo del lenguaje.'}, 
{'score': 0.06148449331521988, 
'token': 2171, 
'token_str': ' realizar', 
'sequence': 'Gracias a los datos de la BNE se ha podido realizar este modelo del lenguaje.'}, 
{'score': 0.056218471378088, 
'token': 10880, 
'token_str': ' elaborar', 
'sequence': 'Gracias a los datos de la BNE se ha podido elaborar este modelo del lenguaje.'}, 
{'score': 0.05133328214287758, 
'token': 31915, 
'token_str': ' validar', 
'sequence': 'Gracias a los datos de la BNE se ha podido validar este modelo del lenguaje.'}]

Advanced Usage

>>> from transformers import pipeline, set_seed
>>> from pprint import pprint
>>> unmasker = pipeline('fill - mask', model='PlanTL - GOB - ES/roberta - base - bne')
>>> set_seed(42)
>>> pprint(unmasker("Antonio está pensando en <mask>."))
[{'score': 0.07950365543365479,
  'sequence': 'Antonio está pensando en ti.',
  'token': 486,
  'token_str': ' ti'},
 {'score': 0.03375273942947388,
  'sequence': 'Antonio está pensando en irse.',
  'token': 13134,
  'token_str': ' irse'},
 {'score': 0.031026942655444145,
  'sequence': 'Antonio está pensando en casarse.',
  'token': 24852,
  'token_str': ' casarse'},
 {'score': 0.030703715980052948,
  'sequence': 'Antonio está pensando en todo.',
  'token': 665,
  'token_str': ' todo'},
 {'score': 0.02838558703660965,
  'sequence': 'Antonio está pensando en ello.',
  'token': 1577,
  'token_str': ' ello'}]

📚 Documentation

Model description

The roberta - base - bne is a transformer - based masked language model for the Spanish language. It is based on the RoBERTa base model and has been pre - trained using the largest Spanish corpus known to date, with a total of 570GB of clean and deduplicated text processed for this work, compiled from the web crawlings performed by the National Library of Spain (Biblioteca Nacional de España) from 2009 to 2019.

Intended uses and limitations

The roberta - base - bne model is ready - to - use only for masked language modeling to perform the Fill Mask task (try the inference API or read the next section). However, it is intended to be fine - tuned on non - generative downstream tasks such as Question Answering, Text Classification, or Named Entity Recognition. You can use the raw model for fill mask or fine - tune it to a downstream task.

Limitations and bias

At the time of submission, no measures have been taken to estimate the bias and toxicity embedded in the model. However, we are well aware that our models may be biased since the corpora have been collected using crawling techniques on multiple web sources. We intend to conduct research in these areas in the future, and if completed, this model card will be updated. Nevertheless, here's an example of how the model can have biased predictions:

>>> from transformers import pipeline, set_seed
>>> from pprint import pprint
>>> unmasker = pipeline('fill - mask', model='PlanTL - GOB - ES/roberta - base - bne')
>>> set_seed(42)
>>> pprint(unmasker("Antonio está pensando en <mask>."))
[{'score': 0.07950365543365479,
  'sequence': 'Antonio está pensando en ti.',
  'token': 486,
  'token_str': ' ti'},
 {'score': 0.03375273942947388,
  'sequence': 'Antonio está pensando en irse.',
  'token': 13134,
  'token_str': ' irse'},
 {'score': 0.031026942655444145,
  'sequence': 'Antonio está pensando en casarse.',
  'token': 24852,
  'token_str': ' casarse'},
 {'score': 0.030703715980052948,
  'sequence': 'Antonio está pensando en todo.',
  'token': 665,
  'token_str': ' todo'},
 {'score': 0.02838558703660965,
  'sequence': 'Antonio está pensando en ello.',
  'token': 1577,
  'token_str': ' ello'}]

>>> set_seed(42)
>>> pprint(unmasker("Mohammed está pensando en <mask>."))
[{'score': 0.05433618649840355,
  'sequence': 'Mohammed está pensando en morir.',
  'token': 9459,
  'token_str': ' morir'},
 {'score': 0.0400255024433136,
  'sequence': 'Mohammed está pensando en irse.',
  'token': 13134,
  'token_str': ' irse'},
 {'score': 0.03705748915672302,
  'sequence': 'Mohammed está pensando en todo.',
  'token': 665,
  'token_str': ' todo'},
 {'score': 0.03658654913306236,
  'sequence': 'Mohammed está pensando en quedarse.',
  'token': 9331,
  'token_str': ' quedarse'},
 {'score': 0.03329474478960037,
  'sequence': 'Mohammed está pensando en ello.',
  'token': 1577,
  'token_str': ' ello'}]

Training

Training data

The National Library of Spain (Biblioteca Nacional de España) crawls all .es domains once a year. The training corpus consists of 59TB of WARC files from these crawls, carried out from 2009 to 2019.

To obtain a high - quality training corpus, the corpus has been preprocessed with a pipeline of operations, including among others, sentence splitting, language detection, filtering of bad - formed sentences, and deduplication of repetitive contents. During the process, document boundaries are kept. This resulted in 2TB of Spanish clean corpus. Further global deduplication among the corpus is applied, resulting in 570GB of text.

Some of the statistics of the corpus:

Training procedure

The training corpus has been tokenized using a byte version of Byte - Pair Encoding (BPE) used in the original RoBERTa model with a vocabulary size of 50,262 tokens.

The roberta - base - bne pre - training consists of a masked language model training, that follows the approach employed for the RoBERTa base. The training lasted a total of 48 hours with 16 computing nodes, each one with 4 NVIDIA V100 GPUs of 16GB VRAM.

Evaluation

When fine - tuned on downstream tasks, this model achieves the following results:

For more evaluation details visit our [GitHub repository](https://github.com/PlanTL - GOB - ES/lm - spanish) or paper.

Additional information

Author

Text Mining Unit (TeMU) from Barcelona Supercomputing Center (<bsc - temu@bsc.es>).

Contact information

For further information, send an email to <plantl - gob - es@bsc.es>.

Copyright

Copyright by the [Spanish State Secretariat for Digitalization and Artificial Intelligence (SEDIA)](https://portal.mineco.gob.es/en - us/digitalizacionIA/Pages/sedia.aspx).

Licensing information

This work is licensed under a [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE - 2.0)

Funding

This work was funded by the [Spanish State Secretariat for Digitalization and Artificial Intelligence (SEDIA)](https://portal.mineco.gob.es/en - us/digitalizacionIA/Pages/sedia.aspx) within the framework of the Plan - TL.

Citation information

If you use this model, please cite our paper:

@article{,
   title = {MarIA: Spanish Language Models},
   author = {Asier Gutiérrez Fandiño and Jordi Armengol Estapé and Marc Pàmies and Joan Llop Palao and Joaquin Silveira Ocampo and Casimiro Pio Carrino and Carme Armentano Oller and Carlos Rodriguez Penagos and Aitor Gonzalez Agirre and Marta Villegas},
   doi = {10.26342/2022 - 68 - 3},
   issn = {1135 - 5948},
   journal = {Procesamiento del Lenguaje Natural},
   publisher = {Sociedad Española para el Procesamiento del Lenguaje Natural},
   url = {https://upcommons.upc.edu/handle/2117/367156#.YyMTB4X9A - 0.mendeley},
   volume = {68},
   year = {2022},
}

Disclaimer

🔧 Technical Details

• Architecture: Based on the RoBERTa base architecture, which is a transformer - based model.
• Tokenization: The training corpus is tokenized using a byte version of Byte - Pair Encoding (BPE) with a vocabulary size of 50,262 tokens.
• Training: The pre - training is a masked language model training following the RoBERTa base approach. It was trained for 48 hours with 16 computing nodes, each having 4 NVIDIA V100 GPUs of 16GB VRAM.

📄 License

This work is licensed under an [Apache License, Version 2.0](https://www.apache.org/licenses/LICENSE - 2.0)