start.sh

#!/bin/bash

# Set environment variables for optimization
export OMP_NUM_THREADS=4  # Optimize OpenMP threading
export MKL_NUM_THREADS=4  # Optimize Intel MKL threading
export CUDA_VISIBLE_DEVICES=0  # Use the first GPU if available

# Start Ollama in the background with optimized settings
ollama serve &  # Use GPU 0 if available

# Pull the model if not already present
if ! ollama list | grep -q "llama3.2:1b"; then
    ollama pull llama3.2:1b
fi

# Wait for Ollama to start up (use a more robust check)
max_attempts=30
attempt=0
while ! curl -s http://localhost:11434/api/tags >/dev/null; do
    sleep 1
    attempt=$((attempt + 1))
    if [ $attempt -eq $max_attempts ]; then
        echo "Ollama failed to start within 30 seconds. Exiting."
        exit 1
    fi
done

echo "Ollama is ready."

# Print the API URL
echo "API is running on: http://0.0.0.0:7860"

# Start the FastAPI server with optimized settings
uvicorn app:app --host 0.0.0.0 --port 7860 --workers 4 --limit-concurrency 20