Tumbuka translation models tum

2 models into Tumbuka, 4 from Tumbuka. Best scored into Tumbuka: French to Tumbuka (chrF++ 22.2). Best scored from Tumbuka: Tumbuka to English (chrF++ 25.9).

Which open models work for Tumbuka?

Windy Translate lists 2 open translation models into Tumbuka and 4 from Tumbuka. All of them can be downloaded from Hugging Face and run on your own hardware. By licence: 6 Apache-2.0 (terms and attribution text: licensing). 6 of them have a screening score on FLORES-200. The best scored into Tumbuka is translate-fr-tum (chrF++ 22.2, Limited). The best scored from Tumbuka is translate-tum-en (chrF++ 25.9, Limited).

With English. English to Tumbuka: no direct model here. Tumbuka to English: translate-tum-en (chrF++ 25.9, Limited).

By band, the scored Tumbuka models are 6 Limited (Excellent is chrF++ 60 or more, Good 45 to 60, Usable 30 to 45, Limited 15 to 30). A score is a 48-sentence screening of news and encyclopedia text, not a measure of accuracy, and a difference of a point or two is noise. How we measured.

Speech-to-text. No speech model here covers Tumbuka: there is no dedicated model, and Whisper's tokenizer does not list the language.

Each model page has a short snippet to download and run the model yourself, its licence and attribution text, and its score where there is one. Pick the best scored model for your direction, then try it on a few sentences of your own text: a screening score is a starting point, not a guarantee.

Into Tumbuka

From Tumbuka

Scores: chrF++ on FLORES-200 dev, 48 sentences, measured 2026-08-04. Method.