Dice Loss for Data-imbalanced NLP Tasks

Benchmark Model Rank Results
chinese-named-entity-recognition-on-msraBERT-MRC+DSC#1F1: 96.72
chinese-named-entity-recognition-on-ontonotesBERT-MRC+DSC#1F1: 84.47
named-entity-recognition-ner-on-conll-2003BERT-MRC+DSC#23F1: 93.33
named-entity-recognition-ner-on-ontonotes-v5BERT-MRC+DSC#1F1: 92.07
question-answering-on-squad11-devXLNet+DSC#3EM: 89.79F1: 95.77
question-answering-on-squad20-devXLNet+DSC#2F1: 89.51EM: 87.65