start.sh

#!/bin/bash

# 1. Set the OLLAMA_HOST to listen on all interfaces (CRITICAL NETWORK FIX for Docker)
export OLLAMA_HOST=0.0.0.0

# 2. Start Ollama in the background
echo "Starting Ollama server in background (listening on $OLLAMA_HOST:11434)..."
ollama serve &

# 3. Wait for the Ollama server process to initialize
echo "Waiting 45 seconds for Ollama server to fully initialize..."
sleep 45

# 4. Pull and CONFIRM MODEL READINESS (Pre-load the model)
MODEL="moondream:1.8b"
echo "Ensuring model $MODEL is pulled and ready..."

# Pull and force a synchronous run to load the model into VRAM
ollama pull "$MODEL"
ollama run "$MODEL" "Describe a house" --verbose --system "You are a helpful assistant." --keep < /dev/null

# 5. Start the Gradio web application
echo "Starting Gradio application on port 7860..."
exec python app.py