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.
08.11.2021: upload model with MLM and NSP heads
: Kuratov, Y., Arkhipov, M. (2019). Adaptation of Deep Bidirectional Multilingual Transformers for Russian Language. arXiv preprint arXiv:1905.07213.
- Downloads last month