| --- |
| library_name: revit-pde |
| tags: |
| - pde |
| - equivariant |
| - vision-transformer |
| - neural-operator |
| --- |
| |
| # ReViT pretrained models |
|
|
| Pretrained checkpoints for **ReViT: Rotational-equivariant Vision Transformers |
| for Neural PDE Solvers**. See the code at https://github.com/tum-pbs/revit. |
|
|
| | Model | Channels | Grid | Patch | Architecture | |
| |-------|----------|------|-------|--------------| |
| | `TCF` | 4 | 96³ | 3 | `ReViT3D` | |
| | `MHD_64` | 7 | 64³ | 2 | `ReViT3D` | |
|
|
| ## Usage |
|
|
| ```python |
| from revit import load_pretrained |
| model = load_pretrained("TCF") # or "MHD_64" |
| ``` |
|
|
| Each subfolder holds `config.json` (the model config), `best_model.pth` |
| (state dict), and the original `config.yaml` for provenance. |
|
|
