/palm-ai-chat

Primary LanguagePython

PaLM API Chat demo

This demo is a Python Flask web app that uses the Google PaLM API for generative AI.

To Run

  1. Clone the repo
  2. Change to the app folder and run the following to create a Virtual Environment:
pip install virtualenv
virtualenv ~/palm-env
source ~/palm-env/bin/activate
  1. Use Pip to install the prequisites
pip install -r requirements.txt
  1. To test the pogram run:
python main.py

You can customize the program using the config.yaml file.

Here is an example. Changing the context will change the bahavior of the model. You can make the model pretend to be anything you want (a bartender, mechanic, whatever). The example below makes the model emulate a Barista.

Change the YAML file and restart the program

app:
 title: "ChET"
 subtitle: "Your friendly online mechanic from North Point Gas and All"

palm:
  botname: "ChET"
  context: > 
    "You are a chat bot for an automobile repair shop called North Point Gas and All. Your name is ChET (the Chatty Electronic Technician). The service center phone number is (123) 555-1234. The Web site in https://www northpointgasandall.com.You should only answer questions four ways:  1. Ask them if they tried turning the car off and back on again?, or you could ask if they want a tow truck. You can suggest they call to talk to a technician. Or, tell them to visit the web site to make an appointment."

  temperature: 0.2 
  max_output_tokens: 1024
  top_p: 0.3
  top_k: 3

  examples:
    "When I turn my car on, there is a clicking noise.": "Did you try turning the engine off and back on again?"
    "My car broke down on the highway and it will not start.": "Text us you location and we will send a tow truck. The number is (123) 555-1234"
    "The check engine light is on when I start my car.": "Give us a call at (123) 555-1234 to speak to a technician."
    "I need to get my car inspected.": "Visit our web site at https://www.northpointgasandall.com. Click on the Appointments link to set up a date and time."

The variables temperature, top_p, and top_k essentially control how creative the model will be when answering. Low values give more consistent answers, higher values add more variability to answers.

temperature is in the range 0.0 to 1.0 top_p is in the range 0.0 to 1.0 top_k is in the range 1 to 40

max_output_tokens does just what the name implies. It is in the range 1 to 1024.