Closed drussellmrichie closed 2 years ago
@drussellmrichie I have faced with the slimilar problem.
Reverting this commit solved my problem. The commit causes mismatch of the tensor size. https://github.com/cooelf/SemBERT/commit/f849452f864b5dd47f94e2911cffc15e9f6a5a2a
@a1phamath That seems to have worked! Thank you so much!
Hi, @cooelf. Thanks again for providing this really interesting work! I am trying to get the code working so I can ultimately adapt it for some other work, and I'm running into some trouble. Hope you can clarify.
I have cloned the repository to my server, set up a Py3.6 virtual environment, installed PyTorch (1.0.0), AllenNLP (0.8.1), and spacy==2.0.18 (also did
pip install --pre allennlp-models
), and downloaded the GLUE data. I then run the following:(You'll notice that I replaced 'bert-wwm-uncased' with 'bert-base-uncased' because I got an error when I tried to use 'bert-wwm-uncased'. I know #8 asks about this as well but even with Google Translate I was not able to figure out how to solve the issue with 'bert-wwm-uncased'.)
I get the following output.
That seems to be saying that there's some mismatch in some matrix shapes? Any idea what's causing that?