| --- |
| title: Leotsha |
| emoji: 🎮 |
| colorFrom: blue |
| colorTo: purple |
| sdk: docker |
| pinned: false |
| tags: |
| - sepedi |
| - african-language |
| - sediba |
| - datafication |
| - edu |
| - game |
| - cdi |
| --- |
| |
| # Leotsha — 7-Mode CDI Datafication Surface |
|
|
| **Leotsha** (Leotsa la Setshaba) is the public-facing **7-mode datafication game surface** of the Sediba AI Sepedi platform — the interactive application layer where people play the 7 mode games that the CDI (Content/Domain Intelligence) layer produced. |
|
|
| Leotsha sits on top of the **Leotsa la Setshaba 7-model CDI/d-acidification layer** (the Sepedi-xlvi Content/Domain Intelligence layer held in the `leotsha_project` repo). It gives end users a Sepedi-native way to interact with and play the datafication modes that the CDI layer produced. |
|
|
| ## What Leotsha does |
|
|
| - **7 mode games** — the interactive datafication surface of the Sediba Sepedi platform. Each mode is a game built on CDI-layer content/domain intelligence produced by the 7-model CDI layer. |
| - **Sepedi-first** — the whole surface is Sepedi-native; it is built for Sepedi speakers and Sepedi contexts first. |
| - **An entry point for the broader Sediba AI Sepedi stack** — datasets, models (sediba-XLM-R, Zabantu Nso sentiment), and the CDI training pipeline feed into the 7 modes. |
|
|
| ## The 7 modes |
|
|
| The Space exposes the **7 modes** of the CDI datafication layer as interactive games. These 7 modes are the product of the 7-model CDI/d-acidification layer held in `leotsha_project`; Leotsha is the surface that makes them playable on the internet. |
|
|
| ## The Sediba stack |
|
|
| | Layer | Component | Where | |
| |---|---|---| |
| | Application / game layer | **Leotsha** (this Space) — 7-mode datafication surface | `huggingface.co/spaces/Sediba-AI/leotsha` | |
| | CDI / model layer | Leotsa la Setshaba (conversational Sepedi) + sediba-XLM-R (MLM) + 7-model CDI layer | held in `leotsha_project` repo + Sediba vaults | |
| | Data layer | Sepedi training corpora + Sepedi-xlvi datasets | `huggingface.co/datasets/Sediba-AI/sepedi-training-v1` | |
|
|
| ## Model family context |
|
|
| | Model | Role | Status | Where | |
| |---|---|---|--| |
| | **sediba-XLM-R** | Masked-language Sepedi NLU anchor | **Shipped** | `Sediba-AI/sediba-XLM-R` on HF | |
| | **Leotsa la Setshaba** | Conversational Sepedi (RLHF tier) | **Pending — publication holds until 1B+ base tier** | Sediba vaults; HF publication pending | |
| | **SedibaLM V7** | QLoRA experiment: Qwen2.5-1.5B + custom Sepedi vocab | **In training (Kaggle trail)** | Kaggle: `sedibaai/sedibalm-v7-sepedi-qlora` | |
|
|
| Leotsa la Setshaba and SedibaLM V7 are **not** published under this Space's model references. They are tracked separately in Sediba's vaults and training pipeline. |
|
|
| ## Status |
|
|
| This Space is the **7-mode CDI datafication surface** in public — the application-layer game surface that Leotsha (the 7-model CDI layer) owns. The underlying models are at different stages: sediba-XLM-R is shipped; Leotsa la Setshaba publication holds until the 1B+ base tier is reached; SedibaLM V7 is in training on the Kaggle trail. |
|
|
| Built by Sediba AI | Mankweng, Limpopo. |
