ictBioRtc/hausa_health_assistant

# Fine-tune Hausa Health Assistant AI Model This workshop demonstrates how to fine-tune a multilingual AI model to create a health assistant that responds in Hausa language, providing culturally appropriate medical guidance for Nigerian communities. ## Prerequisites ### 1. GPU Requirements - Recommended: GPU with at least 16GB VRAM - Minimum: GPU with 8GB VRAM (will use smaller batch sizes) - CPU training is possible but very slow ### 2. HuggingFace Setup (Optional - for saving models) 1. Create an account at huggingface.co 2. Go to Settings → Access Tokens 3. Create a new token with write access (needed for uploading) 4. Copy and save your token - you'll need it later ## Quick Start 1. Open terminal: ```bash File > New Launcher > Terminal ``` 2. Clone the repository: ```bash git clone github.com ``` 3. Navigate to project directory: ```bash cd hausa_health_assistant ``` 4. Install dependencies: ```bash pip install -r requirements.txt ``` 5. Run the application: ```bash python app.py --share ``` 6. Copy the public URL provided (e.g., abc123.gradio.live) 7. Open in a new browser tab ## Using the Application ### Step 1: Initialize the Base Model 1. Go to "🤖 Model Management" tab 2. Click "🚀 Load Base Model" and wait for completion 3. Once loaded, click "⚙️ Prepare for Training" ### Step 2: Test the Untrained Model (Optional) 1. Go to "💬 Chat" tab 2. Try sample queries or type your own in Hausa: - "Ina jin ciwon kai da zazzabi tun kwana biyu. Me ya kamata in yi?" - "Dana yana da gudawa sosai. Ina bukatan taimako." 3. Observe how the base model responds (before training) ### Step 3: Train the Model 1. Return to "🤖 Model Management" tab 2. Adjust training parameters: - **Epochs**: Start with 1 (full training run) - **Batch Size**: Use 2 (adjust based on GPU memory) - **Gradient Accumulation**: Keep default 4 - **Learning Rate**: Keep default 2e-4 3. Click "🎯 Start Training" 4. Monitor the status display …