Ogoun09gerbad/sunutech-wolof-chatbot

# 🇸🇳 Wolof Everyday Chatbot — SunuTech **Team SunuTech — Applied Generative and Agentic AI, AIMS Senegal** A conversational chatbot in Wolof covering everyday topics (culture, education, agriculture, health, transport) and Wolof orthography correction, fine-tuned from Qwen/Qwen3-0.6B with LoRA. --- ## 📁 Project Structure ``` sunutech-wolof-chatbot/ ├── src/ │ └── context_state_machine.py # IDF-weighted retrieval + safety filter ├── hf_space/ │ ├── app.py # Gradio Space (gr.Blocks) │ ├── context_state_machine.py # Copy for the Space │ ├── requirements.txt │ └── data/splits/ # Train + validation only (no eval leakage) ├── data/ │ ├── README.md # Data sources documentation │ ├── chat_synth.json # ⚠️ large file — see note below │ ├── wolof_synth.jsonl # ⚠️ large file — see note below │ └── splits/ # Train / validation / eval splits ├── configs/ │ └── wolof_training_config.yml # LoRA training configuration ├── evaluation_by_category.py # Per-category evaluation script ├── test_assistant_only_masking.py # Label masking verification ├── test_space_context_integration.py # Space integration test (no GPU needed) ├── DATA_SOURCING_GUIDE.md # How to fetch aya & soynade on Kaggle ├── MODEL_CARD_DRAFT.md # Draft model card ├── TEAM_GUIDE.md # Team coordination guide ├── INDIVIDUAL_NOTES.md # Per-member notes └── STATUS.md # What is done / what remains ``` --- ## ⚠️ Large Data Files The large data files (`chat_synth.json`, `wolof_synth.jsonl`, `data/splits/chat_synth_train.json`, etc.) are excluded from this repository via `.gitignore`. **To get the data:** - `synth` data was generated as course material — contact the team. - `aya` and `soynade` sources must be downloaded from Kaggle — see `DATA_SOURCING_GUIDE.md`. --- ## 🚀 Quick Start …