Original PyTorch implementation of UPS proposed in the paper "UPS: Towards Building Foundation Models for PDE Solving via Cross-Modal Adaptation". UPS is developed for solving diverse spatiotemporal PDEs defined over various domains, dimensions, and resolutions. It unifies different PDEs into a consistent representation space and processes diverse collections of PDE data using a unified network architecture that combines LLMs with domain-specific neural operators.
pip install -r requirements.txt
Note that the attrdict package might not be compatible for python 3.10 or newer versions. If getting ImportError: cannot import name 'Mapping' from 'collections', change
from collections import Mapping
to
from collections.abc import Mapping
- Download PDEBench datasets to
./datasets - Generate the PDE metadata
python3 generate_text_embeddings.py
- Generate the data files for data loading
python3 generate_data.py
- Use an existing configuration file or add a new one to
./configs - Run training
python3 main.py --config configs/config_file_name.yaml
Model checkpoints will be released later.