monsoon-nlp/bangla-electra

This is a second attempt at a Bangla/Bengali language model trained with Google Research's ELECTRA . This notebook uses Bangla questions from Google's TyDi dataset:. The model can be used for Question Answering . The model is trained on a web crawl from https://oscar-corpus.com/ (deduped version, 5.8GB) and 1 July 2020 dump of bn.wikipedia.org (414MB) The model was trained on the news classification task:. (Random: 16.7%, mBERT: 72.3%, Bangla-Electra: 82.3%) On Soham Chatterjee's,