fastAI

A minimal GPT-2 based chatbot with a simple plugin system. Each response is printed character by character using textengine.py and saved to saved_responses.txt. The bot now includes rudimentary natural language understanding, optional voice input/output, conversation memory and a feedback system.

Usage

1. Clone this repository
2. Run python chatbot/chat.py
3. Type a prompt and wait for the generated reply

Web and API

Run python chatbot/api.py to start a simple JSON API. chatbot/web_app.py offers a local web chat interface using templates/index.html.

You can also deploy the static interface found in docs/index.html on GitHub Pages. It sends requests to the API endpoint, so make sure api.py is running on a publicly accessible server.

Plugins

Plugins live under chatbot/plugins/. When chat.py starts it automatically loads any Python file in that directory (recursively) that defines a Plugin class. If the class provides a generator(prompt) method, its output will be printed alongside the main GPT-2 response. Example plugins include a GPT-3 wrapper and a translation plugin.

To create a plugin:

# chatbot/plugins/my_plugin.py
class Plugin:
    def generator(self, prompt):
        return "Custom response"

Simply place the file in the plugins folder and run chat.py again. The plugin will be discovered automatically. See chatbot/plugins/offical/gpt3.py for a more detailed example that calls the OpenAI API.

Multi-language

The bot uses a simple translation plugin to demonstrate multi-language support. Enable it by placing a plugin like chatbot/plugins/utils/translation.py in the plugins folder.