DipaHealth/DipaMed-1
DipaMed-1 is an 8 billion parameter causal language model developed by DipaHealth, based on Meta's Llama-3.1-8B architecture with an 8192 token context length. It is specifically specialized and fine-tuned on Nigerian clinical guidelines and biomedical text. This model provides locally-appropriate clinical guidance reflecting Nigerian disease priorities, essential medicines, and national treatment protocols, outperforming its base model on Nigeria-specific medical topics.
Loading preview...
DipaMed-1: Nigerian Clinical Guideline Specialist
DipaMed-1 is an 8 billion parameter language model developed by Destiny Ebhodaghe Ibhate (DipaHealth), adapting Meta's Llama-3.1-8B. Its core distinction lies in its specialization on Nigerian clinical guidelines from the Federal Ministry of Health (FMOH) and Nigeria Centre for Disease Control (NCDC). This grounding allows it to provide clinical guidance tailored to Nigerian disease priorities, the Nigerian Essential Medicines List, and national treatment protocols, which general medical models trained on Western data cannot.
Key Capabilities & Performance
The model was built through continued pretraining on 156 million words of Nigerian biomedical text and instruction tuning using Q&A generated from and verified against real Nigerian guidelines. On the NigeriaMedQA benchmark, DipaMed-1 significantly outperforms its base model on critical Nigeria-specific clinical topics:
- Mental health: 95.8% (+8.3% improvement)
- Maternal emergencies: 90.6% (+6.2% improvement)
- Drug availability: 72.8% (+5.8% improvement)
- Hypertension: 68.5% (+3.7% improvement)
While performing comparably to its base model overall, DipaMed-1 delivers targeted gains where Nigerian specialization is crucial.
Intended Use & Limitations
DipaMed-1 is designed for clinical decision support for trained health workers, not as an autonomous diagnostic system. Users should be aware that it may not reliably memorize numeric dose tables and should be used in conjunction with a retrieval-augmented generation (RAG) pipeline for production to improve factual reliability and citation. As an 8B parameter model, its strength is domain specialization rather than matching frontier models on general medicine.