language:
- pt
tags:
- albertina-pt*
- albertina-100m-portuguese-ptpt
- albertina-100m-portuguese-ptbr
- albertina-900m-portuguese-ptpt
- albertina-900m-portuguese-ptbr
- albertina-1b5-portuguese-ptpt
- albertina-1b5-portuguese-ptbr
- fill-mask
- bert
- deberta
- portuguese
- encoder
- foundation model
license: mit
datasets:
- PORTULAN/glue-ptpt
widget:
- text: >-
A culinária portuguesa é rica em sabores e [MASK], tornando-se um dos
maiores tesouros do país.
This is the model card for Albertina 1.5B PTBR 256. You may be interested in some of the other models in the Albertina (encoders) and Gervásio (decoders) families.
Albertina 1.5B PTBR 256
Albertina 1.5B PTBR 256 is a foundation, large language model for the American variant of Portuguese.
It is an encoder of the BERT family, based on the neural architecture Transformer and developed over the DeBERTa model, with most competitive performance for this language. It has different versions that were trained for different variants of Portuguese (PT), namely the European variant, spoken in Portugal (PTPT) and the American variant, spoken in Brazil (PTBR), and it is openly distributed free of charge under an open license.
| Albertina's Family of Models |
|---|
| Albertina 1.5B PTPT |
| Albertina 1.5B PTBR |
| Albertina 1.5B PTPT 256 |
| Albertina 1.5B PTBR 256 |
| Albertina 900M PTPT |
| Albertina 900M PTBR |
| Albertina 100M PTPT |
| Albertina 100M PTBR |
Albertina 1.5B PTBR 256 is a version for the American variant of Portuguese, and to the best of our knowledge, this is an encoder specifically for this language and variant that, at the time of its initial distribution, with its 1.5 billion parameters and performance scores sets a new state of the art for it, and is made publicly available and distributed for reuse.
It is an encoder of the BERT family, based on the neural architecture Transformer and developed over the DeBERTa model, with most competitive performance for this language. It is distributed free of charge and under a most permissible license.
Albertina 1.5B PTBR 256 is developed by a joint team from the University of Lisbon and the University of Porto, Portugal. For a fully detailed description, check the respective publication:
@misc{albertina-pt-fostering,
title={Fostering the Ecosystem of Open Neural Encoders
for Portuguese with Albertina PT-* family},
author={Rodrigo Santos and João Rodrigues
and Luís Gomes and João Silva
and António Branco and Henrique Lopes Cardoso
and Tomás Freitas Osório and Bernardo Leite},
year={2024},
eprint={2403.01897},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
Please use the above cannonical reference when using or citing this model.
Model Description
This model card is for Albertina 1.5B PTBR 256, with 1.5 billion parameters, 48 layers and a hidden size of 1536.
Albertina 1.5B PTBR 256 is distributed under an MIT license.
DeBERTa is distributed under an MIT license.
Training Data
Albertina 1.5B PTBR 256 was trained over a 36 billion token data set that resulted from gathering some openly available corpora of American Portuguese from the following sources:
- CulturaX: the CulturaX is a multilingual corpus, freely available for research and AI development, created by combining and extensively cleaning two other large datasets, mC4 and OSCAR. It is the result of a selection performed over the Common Crawl data set, crawled from the Web, that retains only pages whose metadata indicates permission to be crawled, that performs deduplication, and that removes some boilerplate, among other filters. Given that it does not discriminate between the Portuguese variants, we performed extra filtering by retaining only documents whose meta-data indicate the Internet country code top-level domain of Portugal.
Preprocessing
We filtered the PTBR corpora using the BLOOM pre-processing pipeline. We skipped the default filtering of stopwords since it would disrupt the syntactic structure, and also the filtering for language identification given the corpus was pre-selected as Portuguese.
Training
As codebase, we resorted to the DeBERTa V2 xxlarge, for English.
To train Albertina 1.5B PTBR 256, the data set was tokenized with the original DeBERTa tokenizer with a 128-token sequence truncation and dynamic padding for 250k steps and a 256-token sequence-truncation for 80k steps. These steps correspond to the equivalent setup of 48 hours on a2-megagpu-16gb Google Cloud A2 node for the 128-token input sequences and 24 hours of computation for the 256-token input sequences. We opted for a learning rate of 1e-5 with linear decay and 10k warm-up steps.
Performance
We resorted to extraGLUE, a PTBR version of the GLUE and SUPERGLUE benchmark. We automatically translated the tasks from GLUE and SUPERGLUE using DeepL Translate, which specifically provides translation from English to PTPT or PTBR as possible options.
| Model | RTE (Accuracy) | WNLI (Accuracy) | MRPC (F1) | STS-B (Pearson) | COPA (Accuracy) | CB (F1) | MultiRC (F1) | BoolQ (Accuracy) |
|---|---|---|---|---|---|---|---|---|
| Albertina 1.5B PTBR | 0.8676 | 0.4742 | 0.8622 | 0.9007 | 0.7767 | 0.6372 | 0.7667 | 0.8654 |
| Albertina 1.5B PTBR 256 | 0.8123 | 0.4225 | 0.8638 | 0.8968 | 0.8533 | 0.6884 | 0.6799 | 0.8509 |
| Albertina 900M PTBR | 0.7545 | 0.4601 | 0.9071 | 0.8910 | 0.7767 | 0.5799 | 0.6731 | 0.8385 |
| BERTimbau (335M) | 0.6446 | 0.5634 | 0.8873 | 0.8842 | 0.6933 | 0.5438 | 0.6787 | 0.7783 |
| Albertina 100M PTBR | 0.6582 | 0.5634 | 0.8149 | 0.8489 | n.a. | 0.4771 | 0.6469 | 0.7537 |
| DeBERTa 1.5B (English) | 0.7112 | 0.5634 | 0.8545 | 0.0123 | 0.5700 | 0.4307 | 0.3639 | 0.6217 |
| DeBERTa 100M (English) | 0.5716 | 0.5587 | 0.8060 | 0.8266 | n.a. | 0.4739 | 0.6391 | 0.6838 |
How to use
You can use this model directly with a pipeline for masked language modeling:
>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='PORTULAN/albertina-1b5-portuguese-ptbr-encoder-256')
>>> unmasker("A culinária portuguesa é rica em sabores e [MASK], tornando-se um dos maiores tesouros do país.")
[{'score': 0.8332648277282715, 'token': 14690, 'token_str': ' costumes', 'sequence': 'A culinária portuguesa é rica em sabores e costumes, tornando-se um dos maiores tesouros do país.'},
{'score': 0.07860890030860901, 'token': 29829, 'token_str': ' cores', 'sequence': 'A culinária portuguesa é rica em sabores e cores, tornando-se um dos maiores tesouros do país.'},
{'score': 0.03278181701898575, 'token': 35277, 'token_str': ' arte', 'sequence': 'A culinária portuguesa é rica em sabores e arte, tornando-se um dos maiores tesouros do país.'},
{'score': 0.009515956044197083, 'token': 9240, 'token_str': ' cor', 'sequence': 'A culinária portuguesa é rica em sabores e cor, tornando-se um dos maiores tesouros do país.'},
{'score': 0.009381960146129131, 'token': 33455, 'token_str': ' nuances', 'sequence': 'A culinária portuguesa é rica em sabores e nuances, tornando-se um dos maiores tesouros do país.'}]
The model can be used by fine-tuning it for a specific task:
>>> from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer
>>> from datasets import load_dataset
>>> model = AutoModelForSequenceClassification.from_pretrained("PORTULAN/albertina-1b5-portuguese-ptbr-encoder-256", num_labels=2)
>>> tokenizer = AutoTokenizer.from_pretrained("PORTULAN/albertina-1b5-portuguese-ptbr-encoder-256")
>>> dataset = load_dataset("PORTULAN/glue-ptbr", "rte")
>>> def tokenize_function(examples):
... return tokenizer(examples["sentence1"], examples["sentence2"], padding="max_length", truncation=True)
>>> tokenized_datasets = dataset.map(tokenize_function, batched=True)
>>> training_args = TrainingArguments(output_dir="albertina-ptbr-rte", evaluation_strategy="epoch")
>>> trainer = Trainer(
... model=model,
... args=training_args,
... train_dataset=tokenized_datasets["train"],
... eval_dataset=tokenized_datasets["validation"],
... )
>>> trainer.train()
Citation
When using or citing this model, kindly cite the following publication:
@misc{albertina-pt-fostering,
title={Fostering the Ecosystem of Open Neural Encoders
for Portuguese with Albertina PT-* family},
author={Rodrigo Santos and João Rodrigues
and Luís Gomes and João Silva
and António Branco and Henrique Lopes Cardoso
and Tomás Freitas Osório and Bernardo Leite},
year={2024},
eprint={2403.01897},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
Acknowledgments
The research reported here was partially supported by: PORTULAN CLARIN—Research Infrastructure for the Science and Technology of Language, funded by Lisboa 2020, Alentejo 2020 and FCT—Fundação para a Ciência e Tecnologia under the grant PINFRA/22117/2016; research project ALBERTINA - Foundation Encoder Model for Portuguese and AI, funded by FCT—Fundação para a Ciência e Tecnologia under the grant CPCA-IAC/AV/478394/2022; innovation project ACCELERAT.AI - Multilingual Intelligent Contact Centers, funded by IAPMEI, I.P. - Agência para a Competitividade e Inovação under the grant C625734525-00462629, of Plano de Recuperação e Resiliência, call RE-C05-i01.01 – Agendas/Alianças Mobilizadoras para a Reindustrialização; and LIACC - Laboratory for AI and Computer Science, funded by FCT—Fundação para a Ciência e Tecnologia under the grant FCT/UID/CEC/0027/2020.