Building Question Answer Agent – Bani Bot (1)

Bani's Model Training

Description

Creating the question answer agent begins with the training of dataset on a pretrained model using the package, Bani: https://github.com/captanlevi/Bani. Training is ran using training script and loading of Bani bot is ran as another process.

Installation/ Set up:

• Ensure docker is available on your machine.
  
• Create 2 docker volume (this can be done via docker app or using command line).
  
  • 1 volume: Store faq store files (.pkl) after building of FAQ classes.
    
  • 1 volume: Store generated model files, after bot training has been completed.
    
• Placed FAQs files in the csv_folder.
  

Steps:

• Access to the project directory.
  
• Build docker image: docker build -t bani_training_script .
  
• Run docker image with created volume:
  

docker run -v $(pwd):/bani_training -it --mount source=model_vol_name,target=/model --mount source=faq_vol_name,target=/faq_store bani_training_script

To note:

FAQ file:

• Each FAQ file contains question and answer for 1 topic.
  
• The first column is the question, second column is the answer. Question and answer data begins on the first line (no header).
  
• File should be available in a CSV format.
  

Volume:

• Subsequent training of FAQ classes with the same file name will overwrite the FAQ files in the faq_store volume.
  
• If subsequent training does not support a topic that was trained previously, do remove the files in the volume, in case there may be issue with loading of bot for answering.