BERT et ses Variantes

XLM-RoBERTa

Multilingual version of RoBERTa pre-trained on 100 languages using massive Common Crawl dataset. Outperforms XLM and mBERT thanks to improved pre-training and better handling of low-resource languages.

← Back