Idea: replace deep learning models in MuZero with LLM models.
create conda environment with python 3.11 (any versions between 3.9 and 3.11 should work, but only 3.11 is tested).
conda create -n llmzero python=3.11
conda activate llmzero
The code is tested to work with pytorch 2.5.1 with cuda. Check https://pytorch.org/get-started/locally/ to install for your machine.
Install other dependencies after installing pytorch with cuda.
pip install -r requirements.txt
This implementation of LLMZero supports environments created using PyRDDLGym, install it with
pip install -q git+https://github.com/tasbolat1/pyRDDLGym.git --force-reinstall
pip install numpy==1.24.2 --force-reinstall
Create .env file in the root folder and add
OPENAI_API_KEY=<your api key>