Example

This example shows how to use kilroy to train a language model on some social media environment.

It uses GPT-2 from HuggingFace as the language model and targets Discord as the environment.

Prerequisites

First, you need to get the necessary files. You can get them here.

Then, you need to create a Discord bot and obtain its token. You can do it here.

You also need to find two channels in some Discord server, one for user messages and one for bot messages. It can be the same channel, if you want it that way.

Paste the bot token and channel IDs into appropriate entries in the .env file.

You also need to install Docker and Docker Compose. You can get them here.

Running

To run the example, simply run the following command:

docker-compose up

This will start all the services. The web app will be available at http://localhost:14000.

Persistence

This stack uses Docker volumes to persist data. The services save their state on exit and load it back on start.