File size: 3,612 Bytes
5646b6d | 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 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 | ---
library_name: braindecode
license: mit
tags:
- EpiNT
- braindecode
- eeg
- ieeg
- seeg
- ecog
- epilepsy
- foundation-model
- model_hub_mixin
- pytorch_model_hub_mixin
---
# EpiNT — Epilepsy Neurophysiological Transformer
Pretrained weights for [`braindecode.models.EpiNT`](https://braindecode.org/stable/generated/braindecode.models.EpiNT.html),
a braindecode port of **EpiNT** (Zhang et al., 2025), a channel-independent
transformer pretrained on 2,700 hours of scalp EEG and intracranial EEG (ECoG,
sEEG) from 1,199 patients.
- Paper: [Cross-Modal Epileptic Signal Harmonization: Frequency Domain Mapping Quantization for Pre-training a Unified Neurophysiological Transformer](https://arxiv.org/abs/2506.17068) (arXiv:2506.17068)
- Original code & weights: [RunKZhang/EpiNT](https://github.com/RunKZhang/EpiNT/tree/781e1d61e72899d59b4955b4c075404eef532a40)
- braindecode docs: https://braindecode.org
## Provenance & license
These weights are the authors' pretrained backbone, `weights/representations.bin`
at commit `781e1d61` (sha256 `b4449974058fd6d314debeef0a09c05730a70e822f17244e88d1ee4e7d9a8b1d`),
converted into the braindecode `EpiNT` module through its `mapping` (all 102
backbone tensors; the pretraining mask token `embed.mask_encoding` is dropped).
The classification head is not part of the release and is randomly initialized.
The original release is under the MIT license, which this repository preserves:
> MIT License
>
> Copyright (c) 2024 Auton Lab, Carnegie Mellon University
>
> Permission is hereby granted, free of charge, to any person obtaining a copy
> of this software and associated documentation files (the "Software"), to deal
> in the Software without restriction, including without limitation the rights
> to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
> copies of the Software, and to permit persons to whom the Software is
> furnished to do so, subject to the following conditions:
>
> The above copyright notice and this permission notice shall be included in all
> copies or substantial portions of the Software.
>
> THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
> AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
> OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
> SOFTWARE.
## Model configuration
This checkpoint uses the paper's configuration (~19.2M parameters):
| | |
|---|---|
| `patch_size` | 256 |
| `embed_dim` | 512 |
| `ffn_dim` | 2048 |
| `n_layers` | 6 |
| `n_heads` | 8 |
| `n_times` | 3072 (12 patches) |
Each encoder `LayerNorm` spans tokens and features, so keep `n_times=3072`:
12 s at 256 Hz for scalp EEG, or 3 s at 1024 Hz for iEEG. Signals are expected
in microvolts. Every channel is encoded alone and the class tokens are averaged,
so `n_chans` is free; `n_outputs` resets the task head.
## Usage
```python
from braindecode.models import EpiNT
model = EpiNT.from_pretrained("braindecode/epint-pretrained", n_chans=17, n_outputs=2)
```
## Citation
```bibtex
@article{zhang2025epint,
title = {Cross-Modal Epileptic Signal Harmonization: Frequency Domain Mapping Quantization for Pre-training a Unified Neurophysiological Transformer},
author = {Zhang, Runkai and Yu, Hua and Gan, John Q. and Wang, Haixian},
journal = {arXiv preprint arXiv:2506.17068},
year = {2025}
}
```
|