File size: 1,692 Bytes
e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 e2e5757 4d99a96 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 | ---
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.
|