Spaces:

WindyTranslate
/

README

Running

App Files Files Community

README / README.md

sneakyfree

Org card: who we are, what is here, scoring and licensing

7f98161 verified 24 days ago

preview code

Raw History Blame Contribute Delete

1.94 kB

---
title: README
emoji: 🌬️
colorFrom: blue
colorTo: indigo
sdk: static
pinned: false
---
# Windy Translate

**Windy Translate is the developer catalogue of [Windstorm Labs](https://windstormlabs.com)**, part of the [Windstorm Institute](https://windstorminstitute.org). This organisation is the canonical home of our translation models on Hugging Face. Scores, licence and attribution for every model are at **[windytranslate.com](https://windytranslate.com)**.

## What is here

1,660 machine-translation models, one language direction (or language group) per repository:

- 1,607 derived from [OPUS-MT](https://huggingface.co/Helsinki-NLP) (Helsinki-NLP, University of Helsinki), in transformers / safetensors format.
- 53 derived from the [HPLT project](https://huggingface.co/HPLT), in CTranslate2 INT8 format.

Each card says what was changed relative to the upstream model and how that was verified.

## How we score

788 models have a screening score: chrF++ and BLEU on 48 sentences of FLORES-200 dev, measured 2026-08-04. FLORES is professionally translated and is not the data these models were trained on. A 48-sentence screen is not a publication result; treat it as a first filter. Models without a score say so on their card. 53 of those are language-verified with GlotLID, which confirms the output language but says nothing about quality.

## Licensing

Every model keeps its upstream licence: 1,495 are Apache-2.0 and 165 are CC-BY-4.0. Each repository carries the full licence text (`LICENSE`) and an attribution notice (`NOTICE.md`). We did not train the base models; the attribution is there because the licences require it and because the credit is due.

## Related organisations

- [WindstormLabs](https://huggingface.co/WindstormLabs): upstream mirrors, speech models and video-model mirrors.
- [WindyWord](https://huggingface.co/WindyWord): the model copies used by the [Windy Word](https://windyword.ai) apps.

	---
	title: README
	emoji: 🌬️
	colorFrom: blue
	colorTo: indigo
	sdk: static
	pinned: false
	---

	# Windy Translate

	Windy Translate is the developer catalogue of [Windstorm Labs](https://windstormlabs.com), part of the [Windstorm Institute](https://windstorminstitute.org). This organisation is the canonical home of our translation models on Hugging Face. Scores, licence and attribution for every model are at [windytranslate.com](https://windytranslate.com).

	## What is here

	1,660 machine-translation models, one language direction (or language group) per repository:

	- 1,607 derived from [OPUS-MT](https://huggingface.co/Helsinki-NLP) (Helsinki-NLP, University of Helsinki), in transformers / safetensors format.
	- 53 derived from the [HPLT project](https://huggingface.co/HPLT), in CTranslate2 INT8 format.

	Each card says what was changed relative to the upstream model and how that was verified.

	## How we score

	788 models have a screening score: chrF++ and BLEU on 48 sentences of FLORES-200 dev, measured 2026-08-04. FLORES is professionally translated and is not the data these models were trained on. A 48-sentence screen is not a publication result; treat it as a first filter. Models without a score say so on their card. 53 of those are language-verified with GlotLID, which confirms the output language but says nothing about quality.

	## Licensing

	Every model keeps its upstream licence: 1,495 are Apache-2.0 and 165 are CC-BY-4.0. Each repository carries the full licence text (`LICENSE`) and an attribution notice (`NOTICE.md`). We did not train the base models; the attribution is there because the licences require it and because the credit is due.

	## Related organisations

	- [WindstormLabs](https://huggingface.co/WindstormLabs): upstream mirrors, speech models and video-model mirrors.
	- [WindyWord](https://huggingface.co/WindyWord): the model copies used by the [Windy Word](https://windyword.ai) apps.