RuBERT (Russian, cased, 12‑layer, 768‑hidden, 12‑heads, 180M parameters) was trained on the Russian part of Wikipedia and news data. We used this training data to build a vocabulary of Russian subtokens and took a multilingual version of BERT‑base as an initialization for RuBERT[1].
[1]: Kuratov, Y., Arkhipov, M. (2019). Adaptation of Deep Bidirectional Multilingual Transformers for Russian Language. arXiv preprint
arXiv:1905.07213.
