The total size of the corpus after document-level deduplication and filtering was 14.82B tokens, split equally between the four languages. The model was trained using a WordPiece tokenizer with a vocabulary size of 96,105 for one million steps with a batch size of 256, and otherwise with default settings.
This research was supported with Cloud TPUs from Google's TPU Research Cloud (TRC).
This project was funded by the Language Technology Programme for Icelandic 2019-2023. The programme, which is managed and coordinated by
Almannarómur, is funded by the Icelandic Ministry of Education, Science and Culture.
