ARIS-Gold-1.5B

🌾 ARIS-Gold-1.5B

Offline AI Agricultural Advisor for Nigerian Farmers

ARIS-Gold is a fine-tuned version of Qwen2.5-1.5B optimised for offline agricultural advice in English and Nigerian Pidgin. It is quantized to GGUF Q4_K_M for efficient CPU inference on 8GB laptops.

---

🏆 ADTC 2026 Submission

This model was built for the Africa Deep Tech Challenge 2026 (Laptop LLM Track) – designed to run on budget hardware with no internet connection.

---

🌍 What It Does

• Diagnoses crop diseases (maize, cassava, yam, rice, tomatoes, pepper, cocoa)
• Provides livestock advice (poultry, goats, cattle, fish farming)
• Answers in Nigerian Pidgin (pcm) and English
• Runs 100% offline – no internet required
• Works on 8GB RAM laptops with no GPU

---

📥 How to Use

With llama.cpp

# Download the model
wget https://huggingface.co/Vicgrace/ARIS-Gold-1.5B/resolve/main/qwen2.5-1.5b-instruct.Q4_K_M.gguf

# Run inference (temperature 0.0 for safety)
llama-cli -m qwen2.5-1.5b-instruct.Q4_K_M.gguf -p "User: My cassava leaves are showing yellow-green mosaic patterns and the plant is stunted. What disease is this and how can I manage it?\nAssistant:" -n 256 --temp 0.0

# Interactive chat mode
llama-cli -m qwen2.5-1.5b-instruct.Q4_K_M.gguf -cnv -t 4 --chat-template qwen

---

🔧 Fine-Tuning Details

• Base model: Qwen/Qwen2.5-1.5B-Instruct
• Method: QLoRA (8 epochs)
• Training data: 497 agricultural Q&A pairs + 50 ARC-Easy questions
• Languages: English + Nigerian Pidgin
• Framework: Unsloth + llama.cpp

---

📊 Evaluation

---

📁 Files

• qwen2.5-1.5b-instruct.Q4_K_M.gguf – Quantized model file (941 MB)

---

📄 License

Apache 2.0 – open-source and free to use.

---

🤝 Credits

• Built by Victor Nwaruwe for the Africa Deep Tech Challenge 2026
• Fine-tuned with Unsloth
• Inference with llama.cpp

---

ARIS — AI for the hardware Africa actually has. 🌾