Skip to content

End-to-End text sumarization, QAs generation using flask.

License

Notifications You must be signed in to change notification settings

Anku5hk/Help-Me-Read

Repository files navigation

Help-Me-Read

A web application created with Flask + BootStrap + HuggingFace🤗 to generate summary and question-answer from given input text. It uses T5 (Text-To-Text Transfer Transformer) for summaries and 'Question Generation using transformers' for question answer generation. For deployment gunicorn(python wsgi server) is used.

Some details:

This application is created with flask(a python microframework), for NLP models HuggingFace is used and for styling and other purposses HTML+CSS+Javascript is used. The goal was to help user read their text, it can be a blog text, some long passage etc. This application takes advantage of multitask model such as T5 to generate abstractive summary, generate questions from the given text and verify thier answers using a NLP technique called Semantic textual similarity (MRPC in short). To get started, user needs to input some text they want to read, then can summarize the given text or also can generate questions based on the texts summary. User can later attend the questions generated to verify their knowledge about the text and can also get results of they did from the model. Models are quantized to save space and increase performance.

Requirements

python 3.9 or above
pytorch 1.8.1 or above
transformers 4.4.2 or above

Installation

  • Install conda/miniconda.
  • Inside Anaconda prompt create a new env $ conda create --name helpmeread_env
  • Activate the env $ conda activate helpmeread_env
  • Upgrade pip $ pip install -U pip
  • Install dependencies $ pip install -r requirements.txt --no-cache-dir
  • Finally Install punkt $ python -m nltk.downloader punkt
  • DONE!!.

Run

  • From Anaconda prompt cd to the directory and hit gunicorn --bind 0.0.0.0:5000 wsgi:app
  • Alternative. if gives error run using python wsgi.py

Note: When running this for the first time models will be downloaded(~400mb).

Docker Installation

  • Build image using docker build --tag helpmeread:1.3 .
  • Run the image as a container docker run --publish 5000:5000 --detach --name hmr helpmeread:1.3
  • Visit the application in browser at localhost:5000

Update

  • Support for newer pytorch, transformers versions.

Screens

1 2

About

End-to-End text sumarization, QAs generation using flask.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published