reloadbrain / paper-qa-bert-api

Api to send a pdf of a paper and use question answerer of bert from deeppavlov

Paper Question Answerer with Bert - API

About

Upload the pdf of any article in english and ask her any questions about its content "where the dataset came from?", "what is the neural network architecture used?" then and the model will answer the question.

Some filters are applied to the article to remove the garbage and leave only the necessary information and the model used is Bert from the DeepPavlov library, which is a transformer that uses the seq2seq neural network architecture to model language, the model was trained in english with the Squad dataset to answer questions given a text.

Webapp

How to use the project? - Without docker

To configure the API for development use:

• Have Java 8 installed

Run

• Git Clone Repo
• Run pip install -r requirements.txt
• Run python -m deeppavlov install squad_bert
• Run python3 paper-qa-api.py

How to use the project? - With docker

To configure the API for development use:

• Have Docker Desktop installed
• In the docker settings, check the 'Expose daemon ...' option and mark your disks as shared drives
• Increase the docker ram limit to at least 4gb

Run

to start the API: Run docker-compose up