Based off of: https://github.com/DoktorHolmes/Maxwell and https://github.com/polakowo/gpt2bot
- Fixed issues with long messages causing the original Maxwell bot to freeze (clears history if an error occurs and prints traceback in Discord)
- Fixed issues with translation causing no response from bot due to misread language
- Removed statistics reporting which had errors that prevented the original program from running
Please ignore instructions from original readme, I updated packages to the latest working versions:
- Clone this repository
- Install Python 3.7.9 if not installed
- Install required libraries (I reccomend using a virtual environment or an IDE like PyCharm). You can install all requirements with
pip install -r requirements.txt:
requests~=2.24.0
torch==1.4.0+cu92
tqdm~=4.48.2
transformers~=2.3.0
python-telegram-bot~=12.8
numpy~=1.19.1
discord~=1.0.1
textblob~=0.15.3
googletrans~=4.0.0-rc1
matplotlib~=3.3.1
python-decouple==3.5
- Copy
.env.confto.env(cp .env.conf .env) - In `.env``, replace "YOUR_TOKEN_GOES_HERE" with your discord bot's API token and save the file
- Open the folder "gpt2bot"
- Run discord_bot.py (
python discord_bot.py). The model will download automatically. - @ the bot or DM to get a response!
After step 5, run the dockerfile and run in a container.. I think should work... maybe.
To reduce confusion, see the MAPLE_README. Install info contained therein is outdated.
To reduce confusion, see the TELEGRAM_README. Install info contained therein is outdated.
- Update readme
- Fix deps
- pull token code out into a dotenv file
- EOS token is being checked during generation -> gpt2bot is now fast enough to be run on CPU.
- Add support for maximum mutual information (MMI) -> more quality, but slower.
You can wait for a full DialoGPT release and then replace the decoder.