---
tags:
  - pytorch
  - safetensors
  - windformer
  - esawaai
---

# WindFormer

*Developed for the ESAWAAI Project and ESAWAAI Legacy Dataset HF Bucket*

A PyTorch implementation that loads a pre-trained `WindFormerDist` architecture and prepares it in evaluation mode for inference.

## Requirements

Ensure the necessary dependencies are installed:

```bash
pip install torch safetensors
```

*Note: CUDA is utilized automatically when available; otherwise, execution defaults to the CPU.*

## File Structure

| **File** | **Description** |
| :--- | :--- |
| `main.py` | Script to load the model weights and set it to evaluation mode |
| `windformer.py` | Model architecture definition for `WindFormerDist` |
| `model.safetensors` | Pre-trained model weights |

## Execution

To run the model inference pipeline:

```bash
python main.py
```

## Model Weights & Configuration

Weights are stored using [Safetensors](https://github.com/huggingface/safetensors).

> **Important:** The instantiation parameters (e.g., `WindFormerDist(fusion=False)`) must strictly match the configuration used during training. Any discrepancy will cause `load_state_dict` to fail due to missing or unexpected keys.

## Cite

Benchaabane, A., Toft, L. D. D. S., Ristea, N.-C., Dimitriadou, K., Husson, R., Hasager, C. B., Anghel, A., Longépé, N., Mouche, A., Grouazel, A., & Datcu, M. (2026). Explainable SAR measurements for Wind Assessment with Artificial Intelligence (ESAWAAI).

## License

This dataset and its associated documentation are released under the terms of the [Creative Commons Attribution 4.0 International (CC-BY 4.0)](https://creativecommons.org/licenses/by/4.0/) license.