cygon24/east-africa-domain-expert

model_card = """--- language: - en - sw license: apache-2.0 base_model: Qwen/Qwen2.5-0.5B-Instruct tags: - fine-tuned - east-africa - aquaculture - agriculture - hospitality - lora - domain-expert

East Africa Domain Expert

Fine-tuned version of Qwen2.5-0.5B-Instruct on East African aquaculture, agriculture, and hospitality domain knowledge.

Part of the Local AI Lab series — building AI systems from scratch, one version at a time.

What this model knows

• Aquaculture — tilapia farming, cold chain logistics, fish diseases, Zanzibar seaweed, market pricing, permits, pond management
• Agriculture — cash crops, soil health, irrigation, maize varieties, climate adaptation, cooperatives, organic farming in Tanzania
• Hospitality — safari lodge operations, Zanzibar hotel management, food sourcing, Booking.com optimization, staff training, tourism regulations

How to use

from transformers import pipeline

pipe = pipeline(
    "text-generation",
    model="cygon24/east-africa-domain-expert",
    max_new_tokens=300,
)

prompt = \"\"\"<|im_start|>system
You are an expert AI assistant specializing in East African aquaculture, 
agriculture, and hospitality industries.<|im_end|>
<|im_start|>user
What fish species are most profitable for aquaculture in Tanzania?<|im_end|>
<|im_start|>assistant
\"\"\"

result = pipe(prompt, do_sample=True, temperature=0.3)
print(result[0]["generated_text"])

Training details

• Base model: Qwen/Qwen2.5-0.5B-Instruct
• Method: LoRA (r=16, alpha=32) via PEFT + TRL SFTTrainer
• Dataset: 30 domain Q&A pairs (10 per domain), East African context
• Hardware: Kaggle GPU (Tesla P100 16GB)
• Epochs: 3
• Framework: transformers, peft, trl

Intended use

Designed for East African business contexts — particularly useful for:

• AI assistants for fishing, farming, and hospitality businesses
• Knowledge base for FishHappy, AgroCare, and similar platforms
• Educational tools for farmers and hospitality workers in Tanzania

Limitations

Trained on a small dataset (30 examples). Best used as a starting point or domain-aware component within a larger RAG or agent system, not as a standalone knowledge base.

Author

Built by Godfrey Muganyizi (Cygon) in Arusha, Tanzania 🇹🇿

• GitHub: github.com/cygon23
• HF: huggingface.co/cygon24 """

Push the model card to HF Hub

from huggingface_hub import HfApi

api = HfApi() api.upload_file( path_or_fileobj=model_card.encode("utf-8"), path_in_repo="README.md", repo_id=HF_REPO, token=hf_token, commit_message="Add proper model card", ) print(f"Model card updated at: https://huggingface.co/{HF_REPO}")