Error when trying to create a record_numericalizer in the Deduplication notebook

[azfarulis]

Hi vintasoftware,

This is my first time writing up a bug report.

I was trying out the Deduplication notebook on Google Colab (GPU + High-ram shape), and when I get to the stage for creating record_numericalizer it throws an AttributeError saying that:

'Vocab' object has no attribute 'load_vectors'

Here is a screenshot attached:

I was following the notebook as structured without any deviation.

Please let me know if I can add any more information to help.

[fjsj]

Thanks for the report, I'll check this in the next couple of days.

[fjsj]

@azfarulis there has been a change in torchtext that broke it's API. I'll be fixing the requirements today. Thanks for reporting the issue. A quick solution right now is forcing the right versions in your Colab by adding below the "Boilerplate" section a new cell including this:

!pip install "torch>=1.7.1,<1.9"
!pip install "torchtext>=0.8,<0.10"
!pip install "torchvision>=0.8.2<0.10"
!pip install "pytorch_lightning>=1.1.6,<1.3"

[azfarulis]

@fjsj Thanks for the heads up. Will try it out and report back!

[fjsj]

@azfarulis
This issue is now fixed on main branch and released as version 0.0.6

Please factory reset your runtimes at Colab and try again. Closing this issue for now. Thanks again for reporting the issue! Please let us know if you find other problems.