transformers/model_cards/rdenadai/BR_BERTo
Rodolfo De Nadai d115872b38
Create README.md (#6020)
* Create README.md

* Update README.md

* Update model_cards/rdenadai/BR_BERTo/README.md

Co-authored-by: Julien Chaumond <chaumond@gmail.com>

* Update model_cards/rdenadai/BR_BERTo/README.md

Co-authored-by: Julien Chaumond <chaumond@gmail.com>
2020-07-24 14:11:14 -04:00
..
README.md Create README.md (#6020) 2020-07-24 14:11:14 -04:00

language tags
pt
portuguese
brazil
pt_BR

BR_BERTo

Portuguese (Brazil) model for text inference.

Params

Trained on a corpus of 5_258_624 sentences, with 132_807_374 non unique tokens (992_418 unique tokens).

But since my machine doesn`t support bigger model, at the end it has a vocab size of 54_000 tokens. The rest of the parameters are the default used in the HuggingFace tutorial.

How to train a new language model from scratch using Transformers and Tokenizers

Results

fill_mask("gostei muito dessa <mask>")

#[{'sequence': '<s>gostei muito dessa experiência</s>',
#  'score': 0.0719294399023056,
#  'token': 2322,
#  'token_str': 'Ġexperiência'},
# {'sequence': '<s>gostei muito dessa diferença</s>',
#  'score': 0.05286405608057976,
#  'token': 3472,
#  'token_str': 'Ġdiferença'},
# {'sequence': '<s>gostei muito dessa atenção</s>',
#  'score': 0.027575725689530373,
#  'token': 2557,
#  'token_str': 'Ġatenção'},
# {'sequence': '<s>gostei muito dessa história</s>',
#  'score': 0.026764703914523125,
#  'token': 1329,
#  'token_str': 'Ġhistória'},
# {'sequence': '<s>gostei muito dessa razão</s>',
#  'score': 0.0250675268471241,
#  'token': 3323,
#  'token_str': 'Ġrazão'},
# {'sequence': '<s>gostei muito dessa resposta</s>',
#  'score': 0.024784332141280174,
#  'token': 2403,
#  'token_str': 'Ġresposta'},
# {'sequence': '<s>gostei muito dessa dose</s>',
#  'score': 0.01720510423183441,
#  'token': 1042,
#  'token_str': 'Ġdose'}]