osiria/roberta-large-italian

---

 
  
    Model: RoBERTa Large
    Lang: IT
  
 

---

Model description

This is a RoBERTa Large [1] model for the Italian language, obtained using XLM-RoBERTa-Large [2] (xlm-roberta-large) as a starting point and focusing it on the italian language by modifying the embedding layer (as in [3], computing document-level frequencies over the Wikipedia dataset)

The resulting model has 356M parameters, a vocabulary of 50.670 tokens, and a size of ~1.42 GB.

Quick usage

from transformers import RobertaTokenizerFast, RobertaForMaskedLM
from transformers import pipeline

tokenizer = RobertaTokenizerFast.from_pretrained("osiria/roberta-large-italian")
model = RobertaForMaskedLM.from_pretrained("osiria/roberta-large-italian")

pipe = pipeline("fill-mask", model=model, tokenizer=tokenizer)

pipe("Milano è una <mask> italiana")

[{'score': 0.9284337759017944,
  'token': 7786,
  'token_str': 'città',
  'sequence': 'Milano è una città italiana'},
 {'score': 0.03296631574630737,
  'token': 26960,
  'token_str': 'capitale',
  'sequence': 'Milano è una capitale italiana'},
 {'score': 0.015821034088730812,
  'token': 8043,
  'token_str': 'provincia',
  'sequence': 'Milano è una provincia italiana'},
 {'score': 0.007335659582167864,
  'token': 18841,
  'token_str': 'regione',
  'sequence': 'Milano è una regione italiana'},
 {'score': 0.006183209829032421,
  'token': 50152,
  'token_str': 'cittadina',
  'sequence': 'Milano è una cittadina italiana'}]

References

[1] https://arxiv.org/abs/1907.11692

[2] https://arxiv.org/abs/1911.02116

[3] https://arxiv.org/abs/2010.05609

License

The model is released under MIT license