lrex93497/bert_qa_pt_3

This is a model for bert QA to answer question for SQuAD 2.0. It trained on SQuAD 2.0 train dataset, epoch 4 (start from 1). It itself has no ability to answer unawserable question.
For details please see https://github.com/lrex93497/fine-tune-BERT-base-uncased-QA-SQuAD2.0/

In our system, we achieved Exact Match (EM): 53.0784, f1: 59.1615
This QA alone only have F1: 32.7591, EM:24.9474
If including answerable question only, this QA alone have F1: 62.0359, EM: 46.3900