ml-jku/gyroswin_cbc_id_ood

---
license: mit
---

This repository contains data used in the paper [GyroSwin: 5D Surrogates for Gyrokinetic Plasma Turbulence Simulations](https://arxiv.org/abs/2510.07314).

We provide snapshots for the in-distribution and out-of-distribution test cases presented in the paper and to be used with the pre-trained GyroSwin model series.

The data was generated using GKW considering the cyclone base case (CBC) mainly focusing on ion temperature gradient based turbulence. The provided simulations can be found at

📁 preprocessed
├── iteration_8.h5
├── iteration_115.h5
├── iteration_131.h5
├── iteration_148.h5
├── iteration_235.h5
├── iteration_262.h5
├── ood_iteration_0.h5
├── ood_iteration_1.h5
├── ood_iteration_2.h5
├── ood_iteration_3.h5
└── ood_iteration_4.h5

where files starting with `ood_` are out-of-distribution with respect to the dataset the GyroSwin models were trained on and the remaining ones are considered in-distribution.

The data contains a single snapshot that was used for autoregressive prediction in the paper. Due to the prohibitive storage demands, we only provide the first snapshot, i.e. they can only be used for autoregressive prediction, but cannot be compared to ground-truth.

Furthermore, normalization statistics are provided at `normalization_stats.pkl` as they are needed for denormalizing the predictions to the original space before transforming back to fourier space.

## Pre-trained model checkpoints

The pre-trained model checkpoints can be found at these links:

- [Small](https://huggingface.co/ml-jku/gyroswin_small)
- [Medium](https://huggingface.co/ml-jku/gyroswin_medium)
- [Large](https://huggingface.co/ml-jku/gyroswin_large)

## Example Usage

A script to run inference using any of our pre-trained models can be found in our [Github repository](https://github.com/ml-jku/neural-gyrokinetics) at `gyroswin/eval/inference_from_hf.py`.

Furthermore, we provide different ways of visualizing 5D Gyrokinetics data in our github repository to visualize the predictions of our GyroSwin models.

## Citation

If you found our work useful, please consider citing it:

@inproceedings{paischergyroswin,
title={GyroSwin: 5D Surrogates for Gyrokinetic Plasma Turbulence Simulations},
author={Paischer, Fabian and Galletti, Gianluca and Hornsby, William and Setinek, Paul and Zanisi, Lorenzo and Carey, Naomi and Pamela, Stanislas and Brandstetter, Johannes},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems}
}