Similarity learning can be handy in a large amount of domains and tasks. It can be used in CV, NLP, recommendation systems, etc.
You may need to look at problems you are already used to from a different angle in order to solve them with similarity learning. At first, it may be unusual and to mitigate "mind-shift" we present our series of tutorials.
This tutorial covers one of NLP problems - Q&A. Here we train a model capable of answering to questions from F.A.Q. pages of popular cloud providers.
The whole tutorial is available here
We use poetry to manage our dependencies.
All packages required are listed in pyproject.toml.
If you have never used poetry earlier, we've collected a bunch of commands to make it seamless in install_dependencies.sh.
You can install all the dependencies with only one command:
./install_dependencies.shMake this file executable with chmod +x install_dependencies.sh before usage.
In this tutorial we use our own dataset.
You can download it with
./download_data.shMake this file executable with chmod +x download_data.sh before usage.
It will fetch cloud_faq_dataset.jsonl file from our Google Storage and make a train-val split.