AI Potluck
Back to Gap Map Model components / Language-specific datasets

African UltraChat

Masakhane
open / Overall score: 1.7

African UltraChat is a multi-turn chat dataset of about 55,000 conversations in 11 African languages, among them Amharic, Hausa, Igbo, Kinyarwanda, Swahili, Yoruba and isiZulu, with a small test split. Each record holds a user prompt and the exchange of user and assistant messages, tagged by language. Masakhane publishes it.

The card carries only a license and a language list, so how the conversations were produced is not documented.

Openness

4 medium confidence
4.0
license
mit(card metadata only)
access
public(Hugging Face repository is not gated)
dataset_card
partial(metadata only (license, languages)

It downloads without a gate under the MIT license given in the card metadata. The card says nothing about where the conversations came from, so any terms inherited from a source dataset are unknown.

Adoption

1 high confidence
1.0

Hugging Face downloads of the single repository. Downloads count file fetches and cannot show how many chat models were tuned on it.

Capability

2 medium confidence
2.0

African UltraChat offers chat-format conversations in 11 African languages, but its card documents neither sources nor method, so it is unclear whether the text was translated, generated or written. It sits well below Inkuba-Instruct, which names every dataset it was built from and has a named model tuned on it.

Verified 2026-09-24