Finetuned on - google-bert-base/bert-multilingual-based with a custom tweet dataset accuracy - 82 trained with 30 epochs