AllenAI's scibert_scivocab_uncased fine-tuned on SQuAD 2.0 evaluated with F1 = 86.85
To load the model:
from transformers import BertTokenizerFast
from transformers import BertForQuestionAnswering
tokenizer = BertTokenizerFast.from_pretrained("LoudlySoft/scibert_scivocab_uncased_squad")
model = BertForQuestionAnswering.from_pretrained("LoudlySoft/scibert_scivocab_uncased_squad")