An offline agricultural advisory assistant fine tuned for English and Yoruba, built for the Africa Deep Tech Challenge 2026 Laptop LLM track by Team Support Vector.

Model details

Base model: Llama-3.2-3B-Instruct
Fine tuning: LoRA (rank 16, alpha 32), 1 epoch, via Unsloth on a Kaggle T4 GPU
Quantization: GGUF Q4_0, 1.78 GiB, runs through llama.cpp
Domain: agriculture (crop, livestock, soil, and market advisory)
Languages: English and Yoruba
Training data: 1nnocent/adtc-agri-yoruba-training-mix, 43,640 English/Yoruba agricultural extension examples

Intended use

Offline agricultural advisory for Yoruba speaking farmers and extension workers, running entirely on CPU with no GPU and no internet connection, on hardware with as little as 8 GB RAM.

Usage

llama-cli -m llama-3.2-3b-agri-yoruba.Q4_0.gguf -cnv -p "Kí ni erùlé àti báwo ni a ṣe lè lò ó fún oko?"

License

Subject to the Llama 3.2 Community License, consistent with the base model.

1nnocent/llama-3.2-3b-agri-yoruba

llama-3.2-3b-agri-yoruba

Model details

Intended use

Usage

License

1nnocent/
llama-3.2-3b-agri-yoruba