diya-chat

Korean chatbot with various language models

Stars
3
Committers
4

KoreanChatbot

Korean chatbot with various language models. This chatbot runs on the local Django website, which will be incorporated to the official DIYA(Do It Yourself AI) webpage

1. Overview

  • This project is written by both tensorflow and pytorch frameworks.
  • The models are trained by Korean QA-pair chatbot data.
  • We tried to train various language model using this Korean chatbot data.

LMs used in this project

  1. BERT(by fine-tuning pretrained model: KorBERT)
  2. Transformer (based on Attention is all you need)
  3. Sequence to sequence with attention mechanism

Data

2. Install and Run

  • How to set virtualenv

    # Install
    $ sudo apt-get install python3-virtualenv   # for linux
    $ pip install virtualenv                    # for mac / windows
    
    # Run
    $ virtualenv --python=python3 diyachat-env
    $ source diyachat-env/bin/activate
    
  • How to install and set packages

    • Install requirements

      $ pip install -r requirements.txt
      $ sudo pip install [package_name] --upgrade   # update packages
      
    • Save current packages into requirements.txt

      $ pip freeze > requirements.txt
      
  • Run docker redis server (for chatting)

    • Installation

      1. Install docker
      $ brew install Docker
      
      1. docker pull redis
    • Run

        sudo docker run -p 6379:6379 -d redis
      
  • Run django server

      $ python manage.py runserver
    
  • Setup sqlite DB

      $ python manage.py makemigrations
      $ python manage.py migrate
    
    • If you want to reset the whole DB, you need to delete db.sqlite3 file and follow the process above again
    • For more detail check Djang tutorial: PART2