braindecode/ZUNA

1---2license: apache-2.03library_name: braindecode4tags:5- eeg6- foundation-model7- braindecode8pipeline_tag: feature-extraction9---10 11# ZUNA (braindecode re-host)12 13Faithful re-host of the **ZUNA** EEG foundation-model encoder weights for use with14[braindecode](https://github.com/braindecode/braindecode).15 16- **Original model:** [`Zyphra/ZUNA`](https://huggingface.co/Zyphra/ZUNA)17- **Original code:** https://github.com/Zyphra/zuna18- **Paper:** Warner, C., Mago, J., Huml, J.R., Osman, M. and Millidge, B. (2026).19  *ZUNA: Flexible EEG Superresolution with Position-Aware Diffusion Autoencoders.*20  arXiv:2602.1847821- **Original authors (Zyphra):** Chris Warner, Jonas Mago, Jon Huml, et al.22- **License:** Apache-2.0 (inherited from the upstream release)23 24## Why this re-host25 26The braindecode `ZUNA` port loads these weights through27`ZUNA.from_pretrained(...)`. Re-hosting under the `braindecode` org gives a28stable, permanent location that the library can point to by default, so the29integration does not depend on the upstream repository staying unchanged. The30weights file is **bit-identical** to the upstream checkpoint (same SHA-256);31only the filename is normalised to the standard `model.safetensors`.32 33## What is (and is not) pretrained34 35These are the pretrained **encoder** weights (a position-aware diffusion36autoencoder trained for EEG superresolution). The braindecode wrapper adds a37**classification head that is randomly initialised** and must be fine-tuned on38your downstream task — loading these weights alone does not give a trained39classifier.40 41## Usage42 43```python44from braindecode.models import ZUNA45 46# Defaults to this repo (braindecode/ZUNA); n_chans / n_outputs are montage-47# and task-dependent and must be supplied.48model = ZUNA.from_pretrained(n_chans=19, n_outputs=4)49 50# Inputs are 5 s EEG windows sampled at 256 Hz (n_times = 1280).51```52