Japanese RoBERTa base size trained by wikipedia dump 20220905 using fairseq. Tokenizer is japanese_roberta_tokenizer.