| --- |
| license: apache-2.0 |
| language: |
| - en |
| datasets: |
| - BGLab/FlowBench |
| tags: |
| - OneScience |
| - computational-fluid-dynamics |
| - neural-operator |
| - lid-driven-cavity |
| --- |
| # FlowBench-FNO-LDC2D-Harmonics |
|
|
| This model reproduces the `LDC_NS_2D/128x128/harmonics` FNO baseline from [FlowBench](https://arxiv.org/abs/2409.18032). It maps a three-channel input to a four-channel flow field. |
|
|
| ## Download |
|
|
| ```bash |
| hf download OneScience-Group/FlowBench-FNO-LDC2D-Harmonics --local-dir ./FlowBench-FNO-LDC2D-Harmonics |
| cd FlowBench-FNO-LDC2D-Harmonics |
| hf download BGLab/FlowBench \ |
| LDC_NS_2D/128x128/harmonics_lid_driven_cavity_X.npz \ |
| LDC_NS_2D/128x128/harmonics_lid_driven_cavity_Y.npz \ |
| --repo-type dataset --local-dir ./data |
| ``` |
|
|
| The two original NPZ files are also available as a [provenance-preserving subset on ModelScope](https://www.modelscope.cn/datasets/OneScience/FlowBench-LDC2D-Harmonics). The original dataset is [CC BY-NC 4.0](https://huggingface.co/datasets/BGLab/FlowBench). No training data is included in this model repository. |
|
|
| ## Run |
|
|
| Install PyTorch, NumPy, PyYAML and Matplotlib in a compatible GPU/DCU environment, then run: |
|
|
| ```bash |
| python scripts/inference.py |
| python scripts/result.py |
| # To repeat the configured training run: |
| python scripts/train.py |
| ``` |
|
|
| The package includes `weight/best_model.pth`, trained on 800 examples and selected on 200 validation examples. Recorded validation MSE is 0.0012336337 and relative L2 is 0.0434478655. Weight SHA256: `13caed2d5d34db87cf728b1669601782642d2ab83a9c8b683fe05e2a80ab5905`. See `results/` for metrics and figures. |
|
|
| The model code is distributed under Apache 2.0. The original FlowBench data retains its separate CC BY-NC 4.0 terms. |
|
|
