Otto-Destiny/mufasa

The Model for Understanding the Frontiers of African Scientific Advancement. MUFASA is a scientific foundation model designed to reason across Africa's scientific and engineering landscape. It will learn from research papers, technical reports, engineering documents, textbooks, datasets, and other scientific literature relevant to Africa. # 🦁 MUFASA ### Models for Understanding the Frontiers of African Scientific Advancement **A small foundation model that knows what African science has already found — and can reason with you about what it hasn't.** Ask most language models what has been studied about groundwater in the Cross River Basin, or which local materials have been tested as partial cement replacements in Nigeria, and you get plausible prose assembled from a global average. MUFASA answers from a corpus of **10,480 African research papers** it was actually trained on, names the study, and tells you when it doesn't know. It runs offline, on a laptop, in about 800 MB. --- ## Table of contents 1. What makes MUFASA different 2. At a glance 3. The problem 4. System overview 5. Part I — Data engineering 6. Part II — Building the training set 7. Part III — Model engineering 8. Results · ADTC profiler 9. What it looks like in use 10. Running MUFASA 11. What we did not build, and why 12. Roadmap 13. Repository map 14. Reproducing this --- ## 1. What makes MUFASA different ### It is both closed-book and open-book Most domain models are one or the other. A retrieval-augmented assistant has no knowledge of its own — take away the database and it is an empty shell. A fine-tuned model has knowledge but cannot cite, ground, or be corrected by a document you hand it. MUFASA is trained on both behaviours, from the same corpus, in the same weights: ```mermaid graph LR subgraph CB["CLOSED BOOK — knowledge in the weights"] C1["Question only"] --> C2["Answer from learned research"] C2 --> C3["Provenance: LEARNED_STUDY Citation: (Nganje et al., 2015)"] end subgraph OB["OPEN BOOK — grounded in what you give it"] O1["Question + evidence"] --> O2["Answer from the evidence only"] O2 --> O3["Provenance: PROVIDED_EVIDENCE Citation + Evidence span"] end subgraph AB["ABSTENTION — the third behaviour"] A1["Question with no support"] --> A2["Says so plainly"] end style CB fill:#1d3a57,color:#fff style OB fil …