# dv-wave

This is a second attempt at a Dhivehi language model trained with
Google Research's [ELECTRA](https://github.com/google-research/electra).

Tokenization and training CoLab: https://colab.research.google.com/drive/1ZJ3tU9MwyWj6UtQ-8G7QJKTn-hG1uQ9v?usp=sharing

V1: similar performance to mBERT after 3 epochs

V2: fixed tokenizers do_lower_case=False and strip_accents=False to preserve vowel signs of Dhivehi

## Corpus

Trained on @Sofwath's 307MB corpus of Dhivehi news: https://github.com/Sofwath/DhivehiDatasets

[OSCAR](https://oscar-corpus.com/) was considered; as of this writing their web crawl has 126MB
of Dhivehi text (79MB deduped).

## Vocabulary

Included as vocab.txt in the upload - vocab_size is 29874