search

Palashio / libra

Ergonomic machine learning for everyone.
http://libradocs.org/
MIT License
1.92k stars 109 forks source link

Question Anwering Query #370

Closed pranavnt closed 4 years ago

pranavnt commented 4 years ago

name: b) New feature - Question Answering Query about: Select this if you're adding a new feature to libra. A query that fits a pre-trained BERT QA model to the dataset (more explained later). labels: feature

This pull request closes #252

- What I did I developed a query that fine-tunes a question answering model on the user's dataset.

- How I did it I used the cdqa library which allowed me to use and fine tune a pre-trained BERT based QA model. Their repo can be found here: https://github.com/cdqa-suite/cdQA

- How to verify it Go through the instructions in this colab notebook to run the code: https://colab.research.google.com/gist/pranavnt/96b426f9a44f0c0ef6abf38bf81fbded/qa-testing.ipynb

This pull request adds a new feature to libra. @Palashio, could you please take a look at it?

anas-awadalla commented 4 years ago

@pranavnt This looks alright but I have a couple of concerns here that need to be addressed before merging. 1st cdqa is built on top of Pytorch which we decided to not be a part of our Libra for now. Also, you don't seem to be using our common framework of allowing the user to specify column names or extracting those from the command which would be a helpful feature to have in this setting. Also, there seems to be no modularity in the training loop which is inconsistent with other queries. Let's chat more on slack!

pranavnt commented 4 years ago

Sounds good! I'll close this PR and reopen once I fix those issues