Sediba-AI/leotsha_project

Leotsha (Leotsa la Setshaba) — the 7-model CDI/d-acidification layer feeding Morutabana. Sepedi-first African-language AI, built in Mankweng, Limpopo, South Africa. # Leotsa la Setshaba — Leotsha Project The **Leotsha** (Leotsa la Setshaba) repository is the public-facing CDI/d-acidification layer for Sediba's Sepedi model family. ## What Leotsha is **Leotsa la Setshaba** ("Language of the Nation") is the canonical Sepedi-anchor name in Sediba's vaults. This repository tracks the Sepedi-xlvi (Content/Domain Intelligence, CDI) layer: the model card, datasets, eval logs, and pipeline notes that underpin Sediba's Sepedi-first AI. ## The model family | Component | Role | Where | |---|---|---| | **sediba-XLM-R** | Sepedi masked-language NLU anchor | `huggingface.co` | | **Leotsa la Setshaba** | Conversational Sepedi (post-training/RLHF tier) | Sediba vaults; HF publication pending 1B+ base | | **SedibaLM V7** | Sepedi QLoRA experiment (Qwen2.5-1.5B-Instruct + custom Sepedi vocab extension) | Kaggle trail + local `v7_launch_bundle/` | The CDI layer is the research/training backbone; **Morutabana** (on Hugging Face Spaces) is the **EduIntel application layer** that surfaces it to teachers and learners. ## Public vs. in-vault The public-facing repo holds model cards, dataset pointers, and high-level pipeline notes. Evaluation logs, governance specs (Esethu / NOODL / Itsebe / Yarena), and day-to-training details stay in Sediba's vaults and are shared with verified partners on request. ## Contact Sediba AI NPC | Mankweng, Limpopo GitHub: github.com