InkubaLM 0.4B
InkubaLM is a compact text model trained from scratch for five African languages, with English and French also in its training data. It is a downloadable research release, rather than the model name for Vulavula's managed services.
Research release
Model and access
The model uses a Llama-style architecture and was trained on 2.4 billion tokens. Its card describes text continuation and warns that generated output can switch languages. Access requires accepting the repository conditions.InkubaLM official model card
Practical fit
Practical guidance. Choose InkubaLM when the task is to study or adapt a small model for African-language text. Review each language separately and include locally written examples. A short continuation prompt is different from a conversation with an instruction-tuned assistant. For a ready transcription workflow, compare the named Vulavula ASR models instead of assuming this text checkpoint can process audio.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | lelapa/InkubaLM-0.4B |
| Context window | 2,048 tokens Maximum training sequence length published in the model card. InkubaLM official model card |
| Parameter count | 0.422 billionInkubaLM official model card |
| Languages | isiZulu, isiXhosa, Yoruba, Hausa, Swahili, English and FrenchInkubaLM official model card |
| Weight license | CC BY-NC 4.0InkubaLM official model card |
Related models
- Lelapa-X-Glot
Lelapa AI · Lelapa-X-Glot
A proprietary African-language translation model used by Vulavula Translate.
Launch date not confirmedAPI · Provider product