RubbaBot is an experimental project in attempt to train the 1558M model gpt-2 Neural Network on over 132,000 messages by only myself on Discord.
- Scrape your messages via DiscordChatExporter in the CSV setting.
- Run
com.uddernetworks.rubbabot.MessageFilter
to filter out the messages with the ID constant in the file to create anoutput.txt
file. Alternatively, put messages to train on in anoutput.txt
- Set up GPT-2 1558M via:
git clone https://github.com/RubbaBoy/gpt-2.git
./gpt-2/setup.sh 1558M output.txt
./gpt-2/train.sh