The computational chemistry and computational materials science community have both developed a great number of simulation tools. Still these tools typically require either rather cryptic input files or at least a fundamental programming experience in a language like Python to control them. Furthermore, many questions are only answered in the documentation, like:
- Which physical units does the code use?
- Which inputs match to which variables in the equations in the paper?
- ...
We address this challenge by developing a Large Language Model (LLM) extension which provides LLM agents to couple the LLM to scientific simulation codes and calculate physical properties from a natural language interface.
While our package is not yet available on the Python Package Index, you can install it directly using:
pip install git+https://github.com/jan-janssen/LangSim.git
The pip package includes optional dependencies for the mace model and the jupyter integration.
As the conda package is not yet available on Anaconda.org still you can clone the repository and install the dependencies directly from conda using the environment.yml file.
Prerequisites:
git clone https://github.com/jan-janssen/LangSim
cd LangSim
conda env create -f environment.yml --name LangSimWe build a docker container on every commit to the main branch.
You can pull the container from the Docker Hub using:
docker run -p 8866:8866 ltalirz/langsimThe package currently provides two interfaces, one for python / jupyter users to query the large language model directly from a python environment and a second web based interface.