IRC-GPT2-Chatbot

An extremely simple Python-based Internet Relay Chat (IRC) bot client for local GPT-2 TensorFlow models.

Setup

1. Clone this repository to your local machine:

   git clone https://github.com/FlyingFathead/IRC-GPT2-Chatbot/
   cd IRC-GPT2-Chatbot/
   

2. Use pip to install the required packages:

   pip install -r requirements.txt
   

3. Get your local GPT-2 TensorFlow model files and place them underneath the main program folder (i.e. ./models/<yourmodelname>)

   To get the model files, you can use i.e. the original OpenAI repo at: https://github.com/openai/gpt-2/

4. Edit the config.txt

   Set your model name to point out to a directory that holds them, i.e. put them inside a directory tree that corresponds to ./models/<yourmodelname> in the config.txt, i.e. if your model name is 124M, make sure its files are in .models/124M/

   You might want to configure the input_prefix, the output_prefix and starting_context to suit your needs.

   Set up your bot's IRC server and other variables from config.txt: SERVER, PORT, NICKNAME, REALNAME, USERNAME, CHANNEL

5. Run the bot:

   python ircbot.py
   

---

By default, the bot will connect to the IRC server, join the designated channel and answer to all incoming messages on it.

The bot is equipped with a conversational context memory, and also has a default context memory trimmer that works in a rolling window-type of context memory management, and it is set up to filter non-UTF-8 encodings (since encodings can be anything on IRC). All in all a very simple and neat way to add a local GPT-2 model to your IRC channel.

More options might be added in later; stay tuned for updates.

---

Enjoy!

Brought to you by FlyingFathead with the kind whispering helps of ChaosWhisperer.