MORPH: PDE Foundation Models with Arbitrary Data Modality

Pretraining sets

Finetuning sets

---

User Guide

The guide for using MORPH as a standalone surrogate and a foundation model is available in ./docs.

Clone the repository

To clone the repository, click on top-right 'code' and select 'clone with HTTPS' and copy the code path and paste in the terminal.

git clone https://github.com/lanl/MORPH.git

Go to the directory

cd MORPH

Check the directory structure

directory_structure.md

Install the requirements

• Install dependencies via environment.yml

conda env create -f environment.yml

• Activate the environment

conda activate pytorch_py38_env

• Install pytorch

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118                    

• Check pytorch installation

python -c "import torch; print('PyTorch version:', torch.__version__); print('CUDA available:', torch.cuda.is_available())"

Output: PyTorch version: 2.4.1+cu118 CUDA available: True

Run the scripts

1. Pretraining script

• Check arguments:

python scripts/pretrain_MORPH.py -h 

2. Finetuning script

• Check arguments:

python scripts/finetune_MORPH.py -h

3. Inference script

• Check arguments:

python scripts/infer_MORPH.py -h

If you use MORPH in your research, please cite:

@article{rautela2025morph,
  title={MORPH: PDE Foundation Models with Arbitrary Data Modality},
  author={Rautela, Mahindra Singh and Most, Alexander and Mansingh, Siddharth and Love, Bradley C and Biswas, Ayan and Oyen, Diane and Lawrence, Earl},
  journal={arXiv preprint arXiv:2509.21670},
  year={2025}
}

Note: EIDR number O#4999 - MORPH: Shape-agnostic PDE Foundational Models. This program is Open-Source under the BSD-3 License.