HNO

Model and experiment

This package reproduces the Darcy flow operator learning at 85 x 85 resolution experiment from the paper. The hyperbolic neural operator uses a width-96, depth-4, four-head model. Darcy coefficients are downsampled from 421 x 421 to 85 x 85 before prediction.

The package includes model code, config/config.yaml, training, inference and visualization scripts, the recorded results/, and the selected weight/best_model.pth checkpoint.

Data

Download OneScience/HNO-Darcy with modelscope download --dataset OneScience/HNO-Darcy --local_dir ./data. Two Darcy MAT files; each has coeff and sol [1024,421,421], both MATLAB double (float64). First file provides 1,000 training cases; second supplies 200 evaluation cases.

Original dataset: https://drive.google.com/drive/folders/1UnbQh2WWc6knEHbLn-ZaXrKUZhp7pjt- Download the dataset before running inference where an external dataset is required. The release configuration uses relative paths, not SCNet personal-directory paths.

Installation and usage

hf download OneScience-Group/HNO --local-dir ./HNO
cd HNO
pip install -r requirements.txt
python scripts/inference.py
python scripts/result.py

For a fresh training run, use an allocated GPU/DCU compute node with a compatible PyTorch build, keep a separate copy of the supplied checkpoint, then run python scripts/train.py from the package root. The full budget is specified in config/config.yaml. Cluster-specific Slurm scripts, if present, need adaptation outside the original cluster.

Training and reported result

The run completed 500 epochs with AdamW and OneCycle scheduling. The best checkpoint came from epoch 488, selected on the same 200 samples later used for evaluation; these are not an additional independent test set.

Mean evaluation relative L2 was 0.00450221 versus 0.0045 in paper Table 2, about 0.049% higher. This is one run, whereas the paper reports three-run means. See results/ for the recorded metrics and visualizations. The supplied checkpoint comes from a completed run; another run may differ.

Source and license

• Paper: https://guobapei.github.io/Hyperbolic-Neural-Operator/assets/paper.pdf.
• Dataset source: https://drive.google.com/drive/folders/1UnbQh2WWc6knEHbLn-ZaXrKUZhp7pjt-.
• This model package uses the license indicated in the card metadata. Dataset redistribution terms, where not separately stated, must not be inferred from the code license.