attilanagy234 / neural-punctuator Goto Github PK

In line 16 of src/neural_punctuator/models/BertPunctuator.py add return_dict=False to from_pretrained(), otherwise a dict is returned, from which your code takes the name of the layer, rather than the tensor.

(This will fix the bug with a traceback of dropout expecting a tensor rather than a string)

• create random seed
• save with model for reproducibility

Execuse me, when I train the model in GPU,I meet this error.Could you help me ?
Traceback (most recent call last):
File "main.py", line 6, in
pipe.train()
File "/workspace/neural-punctuator-main/src/neural_punctuator/wrappers/BertPunctuatorWrapper.py", line 17, in train
self._trainer.train()
File "/workspace/neural-punctuator-main/src/neural_punctuator/trainers/BertPunctuatorTrainer.py", line 106, in train
mask = ((targets == 0) & (np.random.rand(*targets.shape) < .1)) | (targets > 0)
TypeError: and() received an invalid combination of arguments - got (numpy.ndarray), but expected one of:

• (Tensor other)
  didn't match because some of the arguments have invalid types: (numpy.ndarray)
• (Number other)
  didn't match because some of the arguments have invalid types: (numpy.ndarray)

How did you decide id2target for converting back from prediction to original.
id2target = {-1: 0,
9: 1, # .
60: 2, # ?
15: 3, # ,
-2: -1, # will be masked
}

I have class 1 as , 2 as ? 3 as . and 4 as ! 0 for all others.
How should I use this?

• Need to show that transformers actually bring improvement.
• Train a simpler model (LSTM/1d conv, etc.)

Hi, I wanted to know if you have implemented to check the model output at test time. Because we had aligned labels according to tokens so model output would be according to that. How can we convert them back to original which was before aligning ?

Best Regards

Can you please share the pre-trained model?

What can we do to convert it into tflite or use the pt file on the mobile device?

https://arxiv.org/pdf/1911.02855.pdf

Hello, can you share the data set of the experiment, the displayed network link can no longer be downloaded.Thanks!

Hello,

I'm kind of new with BERT etc, sorry in advance for the question, but how and where in the code sould I add my own texts, in order to have in the output my texts with the punctuation ?

Thanks,
Camille

Hi, I am getting validation accuracy greater than training and validation loss lesser than train. Can you tell why? Maybe due to class imbalance? I am using different dataset