---
license: cc-by-nc-sa-4.0
language:
- bg
- cs
- da
- el
- es
- et
- fi
- hr
- hu
- it
- lt
- lv
- mt
- nl
- pl
- pt
- ro
- sk
- sl
- sv
---

# SpidR VP-20

SpidR VP-20 is a SpidR model pretrained  pretrained on a subset of 6k hours and 20 languages of VoxPopuli
(all EU languages except English, French, and German)
for the [DiscoPhon benchmark](https://benchmarks.cognitive-ml.fr/discophon).
It was pretrained using the [`spidr`](https://github.com/facebookresearch/spidr) library.

You can load it with:

```python
from spidr.models import SpidR
from torch.hub import load_state_dict_from_url

state_dict = load_state_dict_from_url("https://huggingface.co/coml/spidr-vp20/resolve/main/final.pt")
model = SpidR().eval()
model.load_state_dict(state_dict)
```

## Files:

- `config.json`: Model configuration.
- `final.pt`: Model checkpoint.
- `full_checkpoint.pt`: Full checkpoint, with model, optimizer, etc.

## Citing

Please cite the DiscoPhon paper

```bibtex
@misc{poli2026discophon,
  title={{DiscoPhon}: Benchmarking the Unsupervised Discovery of Phoneme Inventories With Discrete Speech Units},
  author={Maxime Poli and Manel Khentout and Angelo Ortiz Tandazo and Ewan Dunbar and Emmanuel Chemla and Emmanuel Dupoux},
  year={2026},
  eprint={2603.18612},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2603.18612},
}
```

along with [SpidR](https://openreview.net/forum?id=E7XAFBpfZs).