liat-ml-roberta 1.1.3

Multi-Launguage RoBERTa trained by RIKEN-AIP LIAT.

liat_ml_roberta

Multi-Language RoBERTa trained by RIKEN-AIP LIAT.

How to install

Can use pip to install.

pip install liat_ml_roberta

How to use

The loaded models and configurations can be used in the same way as transformers.roberta.

from liat_ml_roberta import RoBERTaTokenizer


def main():
    tokenizer = RoBERTaTokenizer.from_pretrained(version="en_20190121_m10000_v24000_base")
    print(tokenizer.tokenize("This is a pen."))

    config = RoBERTaConfig.from_pretrained("roberta_base_en_20190121_m10000_v24000_u125000")
    model = RoBERTaModel.from_pretrained("roberta_base_en_20190121_m10000_v24000_u125000", config=config)


if __name__ == "__main__":
    main()

Models

name	lang	size	bpe merges	vocab size	updates	wikipedia version
roberta_base_ja_20190121_m10000_v24000_u125000	ja	roberta-base	10000	24000	125000	20190121
roberta_base_ja_20190121_m10000_v24000_u500000	ja	roberta-base	10000	24000	500000	20190121
roberta_base_en_20190121_m10000_v24000_u125000	en	roberta-base	10000	24000	125000	20190121
roberta_base_en_20190121_m10000_v24000_u500000	en	roberta-base	10000	24000	500000	20190121