Howardqlz

I met the same problem. I fix it like this: `p = self.word_emb([kwargs['p'].cuda(self.args.gpu)])`

I just want to use your model as a pretrained model to finetune on my own dataset. Wish your reply if you have time to explain the training step about...

how about dcn+ performs on SQuAD2.0?Can f1 or em reaches 50%?And it's seem that the model runs an epoch needs 3hours，i don't know it's correct or not(the f1 and em...