dev-chauhan / pqg-pytorch Goto Github PK

Hello there,
Do you have the pretrained models available somewhere?

May I know where can I find the score.py file in the repo.
I am anot able to find it here.

It seems like that after I trained the model with our own dataset using the code from master branch, I can not evaluate the model on the test set unless switching to ori- branch. However, I don't want to waste 2days training the model. What can I do to evaluate the testing data in the master branch.

How can i test the model with different paraphrase sentence after training?

Could you tell me the default epoch number to reproduce your results in the paper?

The delimiter in quora_prepro.py should be '\t' not ',' for the .tsv file

The nn.CrossEntropyLoss expects logits, not log of the probabilities.
This criterion combines nn.LogSoftmax() and nn.NLLLoss() in one single class.
https://pytorch.org/docs/stable/nn.html#torch.nn.CrossEntropyLoss

Use just nn.NLLLoss https://pytorch.org/docs/stable/nn.html#torch.nn.NLLLoss when working with log probabilties. Fix:

loss = nn.NLLLoss(ignore_index=data.PAD_token)