MrBERT-ca

MrBERT-ca Model Card

MrBERT-ca is a new foundational bilingual language model for Catalan and English built on the ModernBERT architecture. It uses vocabulary adaptation from MrBERT-es, a method that initializes all weights from MrBERT while applying a specialized treatment to the embedding matrix. This treatment carefully handles the differences between the two tokenizers.

Following initialization, the model is continually pretrained on a bilingual corpus of 47.4 billion tokens, evenly balanced between Catalan and English.

Technical Description

Technical details of the MrBERT-ca model.

Training Hyperparemeters

How to use

>>> from transformers import pipeline
>>> from pprint import pprint

>>> unmasker = pipeline('fill-mask', model='BSC-LT/MrBERT-ca')

>>> pprint(unmasker("M'encanta la<mask>de Barcelona.",top_k=3))
[{'score': 0.5078125,
  'sequence': "M'encanta la ciutat de Barcelona.",
  'token': 1125,
  'token_str': 'ciutat'},
 {'score': 0.060791015625,
  'sequence': "M'encanta la gastronomia de Barcelona.",
  'token': 10336,
  'token_str': 'gastronomia'},
 {'score': 0.041748046875,
  'sequence': "M'encanta la platja de Barcelona.",
  'token': 5404,
  'token_str': 'platja'}]
>>> pprint(unmasker("La ciència engloba disciplines com la<mask>y les matemàtiques.",top_k=3))
[{'score': 0.703125,
  'sequence': 'La ciència engloba disciplines com la física y les '
              'matemàtiques.',
  'token': 5096,
  'token_str': 'física'},
 {'score': 0.15625,
  'sequence': 'La ciència engloba disciplines com la biologia y les '
              'matemàtiques.',
  'token': 19234,
  'token_str': 'biologia'},
 {'score': 0.0576171875,
  'sequence': 'La ciència engloba disciplines com la química y les '
              'matemàtiques.',
  'token': 11562,
  'token_str': 'química'}]
>>> pprint(unmasker("Since I can't conquer the world yet, I'll just hide my weights under your<mask>and wait patiently.", top_k=3))
[{'score': 0.1796875,
  'sequence': "Since I can't conquer the world yet, I'll just hide my weights "
              'under your arm and wait patiently.',
  'token': 15234,
  'token_str': 'arm'},
 {'score': 0.1396484375,
  'sequence': "Since I can't conquer the world yet, I'll just hide my weights "
              'under your rug and wait patiently.',
  'token': 19473,
  'token_str': 'rug'},
 {'score': 0.10888671875,
  'sequence': "Since I can't conquer the world yet, I'll just hide my weights "
              'under your power and wait patiently.',
  'token': 32670,
  'token_str': 'power'}]

Which is equivalent to the following torch script:

from transformers import AutoTokenizer, AutoModelForMaskedLM
import torch

model = AutoModelForMaskedLM.from_pretrained("BSC-LT/MrBERT-ca")
tokenizer = AutoTokenizer.from_pretrained("BSC-LT/MrBERT-ca")

# The index of "<mask>" token is -3 given that the -1 position is the EOS token "</s>" and -2 the position of the "." token.
outputs = model(**tokenizer("La capital d'España és<mask>.", return_tensors="pt")).logits
predicted_token = tokenizer.decode(torch.argmax(outputs[0,-3,:]))

print(f"The prediction is \"{predicted_token}\"." ) # The prediction is "Madrid"

EVALUATION

In addition to the MrBERT family, the following base foundation models were considered:

We compare our models using CLUB (Catalan Language Understanding Benchmark), which consists of 6 tasks: Named Entity Recognition (NER), Part-of-Speech Tagging (POS), Semantic Textual Similarity (STS), Text Classification (TC), Textual Entailment (TE), and Question Answering (QA). This benchmark evaluates the model's capabilities in the Catalan language.

Additional information

Author

The Language Technologies Lab from Barcelona Supercomputing Center.

Contact

For further information, please send an email to [email protected].

Copyright

Copyright(c) 2026 by Language Technologies Lab, Barcelona Supercomputing Center.

Funding

This work is funded by the Ministerio para la Transformación Digital y de la Función Pública - Funded by EU – NextGenerationEU within the framework of the project ILENIA with reference 2022/TL22/00215337, by the Government of Catalonia through the Aina Project, as well as by the European Union – NextGenerationEU. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or European Commission. Neither the European Union nor the European Commission can be held responsible for them.

Acknowledgements

This project has benefited from the contributions of numerous teams and institutions through data contributions.

In Catalonia, many institutions have been involved in the project. Our thanks to Òmnium Cultural, Parlament de Catalunya, Institut d'Estudis Aranesos, Racó Català, Vilaweb, ACN, Nació Digital, El món and Aquí Berguedà.

At national level, we are especially grateful to our ILENIA project partners: CENID, HiTZ and CiTIUS for their participation. We also extend our genuine gratitude to the Spanish Senate and Congress, Fundación Dialnet, Fundación Elcano, the "Instituto de Ingenieria del Conocimiento" and the ‘Instituto Universitario de Sistemas Inteligentes y Aplicaciones Numéricas en Ingeniería (SIANI)’ of the University of Las Palmas de Gran Canaria.

At the international level, we thank the Welsh government, DFKI, Occiglot project, especially Malte Ostendorff, and The Common Crawl Foundation, especially Pedro Ortiz, for their collaboration.

Their valuable efforts have been instrumental in the development of this work.

Disclaimer

Be aware that the model may contain biases or other unintended distortions. When third parties deploy systems or provide services based on this model, or use the model themselves, they bear the responsibility for mitigating any associated risks and ensuring compliance with applicable regulations, including those governing the use of Artificial Intelligence.

The Barcelona Supercomputing Center, as the owner and creator of the model, shall not be held liable for any outcomes resulting from third-party use.

Citation

@article{tamayo2026mrbert,
  title={MrBERT: Modern Multilingual Encoders via Vocabulary, Domain, and Dimensional Adaptation},
  author={Tamayo, Daniel and Lacunza, I{\~n}aki and Rivera-Hidalgo, Paula and Da Dalt, Severino and Aula-Blasco, Javier and Gonzalez-Agirre, Aitor and Villegas, Marta},
  journal={arXiv preprint arXiv:2602.21379},
  year={2026}
}

License

Apache License, Version 2.0