search

victordibia / neuralqa

NeuralQA: A Usable Library for Question Answering on Large Datasets with BERT
https://victordibia.github.io/neuralqa/
MIT License
233 stars 32 forks source link

Pytorch in requirements? #53

Open dconathan opened 3 years ago

dconathan commented 3 years ago

Both torch and torchvision are in requirements.txt and setup.py but they don't seem to be used anywhere. Any particular reason you have both torch and tensorflow in requirements?

victordibia commented 3 years ago

Hi @dconathan

I recall this being a requirement to load/convert HF models that do not have TF weights.
I am happy to get your thoughts on ways to completely sidestep pytorch/torchvision (as it makes the install really heavy ). I'll investigate a bit more .. let me know if you have any additional ideas.

dconathan commented 3 years ago

Ah that makes sense. Are you open to pytorch-only backend and eliminating tensorflow as a dependency?

victordibia commented 3 years ago

At this point, that seems like a good way to simply things. I'll take a pass this weekend, articulate out an issue and post an update. Given that the library is not more focused on usability/visual interface, it probably does not matter much if its pt or tf underneath.