monsoon-nlp/tamillion

This is the second version of a Tamil language model trained with Google Research's ELECTRA . The model outperformed mBERT on news classification and movie reviews . It could be finetuned to train a QA model for Hindi and Bengali . The Tamil-language question-answering question-ansistability question-a-question-analyst dataset is not complete, but this model could be used to train QA models . It has been trained on IndicCorp Tamil (11GB) and 1 October 2020 dump of https://ta.org/corpora/tamil-nlp.org (482MB) The model was trained with a small model with,