search

facebookresearch / DrQA

Reading Wikipedia to Answer Open-Domain Questions
Other
4.48k stars 898 forks source link

Interactive.py: Why I got KeyError: 'unexpected key "module.embedding.weight" in state_dict' #142

Closed jasperzhong closed 6 years ago

jasperzhong commented 6 years ago

I trained my model using pre-trained word embedding Train command: python3 train.py --embedding-file word2vec_baike.txt --batch-size 128 --data-dir data/train/preprocessed --train-file train-processed-corenlp.txt --dev-file dev-processed-corenlp.txt --dev-json dev.json --embed-dir data --model-name model3 --parallel True --weight-decay 0.0001 --momentum 0.9

After training, I try to use the trained model with interactive.py but I got such error:

Traceback (most recent call last): File "interactive.py", line 55, in normalize=False) File "/home/oyyj/Programs/QA/drqa/reader/predictor.py", line 61, in init normalize=normalize) File "/home/oyyj/Programs/QA/drqa/reader/model.py", line 436, in load return DocReader(args, word_dict, feature_dict, state_dict, normalize) File "/home/oyyj/Programs/QA/drqa/reader/model.py", line 59, in init self.network.load_state_dict(state_dict) File "/usr/local/lib/python3.5/dist-packages/torch/nn/modules/module.py", line 490, in load_state_dict .format(name)) KeyError: 'unexpected key "module.embedding.weight" in state_dict'

I am very confused and don't know why. Can anyone help me?

Env: python3.5 pytorch0.3

ajfisch commented 6 years ago

Oh, hm, I think this is a bug with saving models trained with DataParallel. I'll put in a fix...