Bi-Directional Attention Flow for Machine Comprehension

This is a 🔥 PyTorch 🔥 implementation of the seminal BiDAF Paper from Allen AI(https://arxiv.org/pdf/1611.01603.pdf)

Architecture

My modifications include 🚀 :

• QA on a token level instead of a character level ( I use Spacy 2.1.9 to create the dataset from SQUAD )
• Swapping out GloVe for Elmo for improved performance

Dataset

I have uploaded a sample dataset from the SQUAD Dataset. Download the complete dataset at SQUAD QA Dataset

Demo

I have also implemented a Streamlit App to interact with the model!

Instructions

To Install Requirements

pip install -r requirements.txt

To Train a New Model

python train.py

To Run the Demo Streamlit App

streamlit run app.py

Improvements

✌️ Pull requests are welcome for any improvements/features ✌️