Discord-GPT2-bot

A simple Discord chatbot powered by OpenAI's GPT-2 model, designed to run on local TensorFlow setups.

Install

1. Clone the repository

git clone https://github.com/FlyingFathead/Discord-GPT2-bot
cd Discord-GPT2-bot/

2. Install dependencies

pip install -r requirements.txt

3. Get the OpenAI GPT-2 model and tune it according to your requirements. Place the model in the corresponding subdirectory (e.g., models/124M/). Ensure that the directory contains all necessary TensorFlow model files, including vocab.bpe, encoder.json, and hparams.json.

4. Set up the bot (see the code; fix the variables to suit your use scenario):

   • Save your Discord API bot token in token.txt (one line).
   • Define the bot's authorized users in authorized_users.txt (one username per line; syntax i.e. UserName#0 or UserName#xxxx).

5. Optional & additional configuration

   • Feel free to configure various replacement lists for elements like unwanted URLs, words, strings, allowed channels etc. Refer to the source code for a complete list of replaceable items.

6. Run with python Discord-GPT2-bot.py and enjoy! (?)

Usage

Talk to the bot and it should respond.

Commands:

• !goodnight - the bot shuts down (user must be in the authorized users list, authorized_users.txt by default)
• !maxrolls x - max number of replies to a single line (x = number of replies)
• !odds x - odds (probability) of the bot answering; 0 for 0%, 1 for 100%, use fractions like 0.5 = 50%
• !temp x - set the model's temperature (between 0 and 2). Optimal for an optimized model usually hangs at around 1.
• !alert - send out a test alert (for testing, debug purposes etc)

Notes

This is mostly for humorous & testing purposes, although the code works perfectly fine when set up correctly, don't expect miracles out of the bot. Actual performance depends on fine-tuning and model size, etc. -- Feel free to expand on its functionalities and adapt it to your needs.

From FlyingFathead, in collaboration with ChaosWhisperer.