convert_tf_checkpoint_to_pytorch.py error

Question

convert_tf_checkpoint_to_pytorch.py error

mrmutator opened this issue 5 years ago · comments

Hi,

When running the conversion script convert_tf_checkpoint_to_pytorch.py I get the following error when trying to convert the BERT-Base, uncased model:

Traceback (most recent call last):
  File "convert_tf_checkpoint_to_pytorch.py", line 107, in <module>
    convert()
  File "convert_tf_checkpoint_to_pytorch.py", line 100, in convert
    pointer.data = torch.from_numpy(array)
TypeError: expected np.ndarray (got numpy.ndarray)

Any ideas? Thanks

OmGraja · Answer 1 · Wed Feb 27 2019 00:29:56 GMT+0800 (China Standard Time)

Please provide more infos about the Bert model you are using.
It worked for me in the case of BERT-Base, Uncased: 12-layer, 768-hidden, 12-heads, 110M parameters.

Roger Wechsler · Answer 2 · Wed Feb 27 2019 16:59:53 GMT+0800 (China Standard Time)

I'm also using the Bert model BERT-Base, Uncased: 12-layer, 768-hidden, 12-heads, 110M parameters.
But I get the error above on Windows with Pytorch 1.01 and Tensorflow 1.13.0rc2

OmGraja · Answer 3 · Thu Feb 28 2019 02:21:08 GMT+0800 (China Standard Time)

Windows does not support some of ner-bert packages and libraries. I recommend that you implement it on linux OS or MacOS. A lot of bugs will occur if you implement it on windows.

Anton Emelyanov · Answer 4 · Mon Mar 04 2019 19:25:54 GMT+0800 (China Standard Time)

All experiments have done on linux :).