vultr-deepseek-r1.sh

#!/bin/bash
#########################################################################################
# Ollama + Chatbox + DeepSeek Installation & Configuration Script
#
# This script:
# - Deploys Ollama and ensures it's configured for remote access
# - Moves Ollama storage to /data (if needed) to optimize space
# - Installs NVIDIA drivers, CUDA, and configures GPU usage
# - Verifies CUDA and VRAM availability
# - Configures UFW to allow external access
# - Downloads and verifies the DeepSeek-r1:671b model
# - Implements a wait loop for DeepSeek to fully load
# - Starts watching logs at the end for debugging
#
# Usage:
# 1. Save and run this script on your server
# 2. Once completed, connect Chatbox to http://<your-server-ip>:11434
#
# DISCLAIMER:
# This script is provided "as is" without any warranties or guarantees.
# Use at your own risk. Always verify before running scripts on production systems.
#########################################################################################

# Stop on errors
set -e

echo "🚀 Starting Ollama + DeepSeek installation on H100..."

# Ensure /data and /data/ollama directories exist before proceeding
if [ ! -d "/data" ]; then
    echo "📂 Creating /data directory..."
    sudo mkdir -p /data
fi

if [ ! -d "/data/ollama" ]; then
    echo "📂 Creating /data/ollama directory..."
    sudo mkdir -p /data/ollama
fi

# Set correct ownership and permissions
echo "🔧 Setting correct permissions for Ollama storage..."
sudo chown -R root:root /data/ollama
sudo chmod -R 755 /data/ollama

# Install NVIDIA Drivers and CUDA
echo "⚙️ Checking for NVIDIA GPU..."
if ! command -v nvidia-smi &> /dev/null; then
    echo "📥 Installing NVIDIA Drivers & CUDA..."
    sudo apt update
    sudo apt install -y nvidia-driver-535 nvidia-utils-535 cuda-toolkit-12-2
else
    echo "✅ NVIDIA Drivers already installed."
fi

# Verify CUDA installation
if ! command -v nvcc &> /dev/null; then
    echo "❌ CUDA not installed properly. Please check manually."
    exit 1
fi
echo "✅ CUDA installed successfully: $(nvcc --version | grep release)"

# Check VRAM availability
echo "🖥 Checking GPU VRAM availability..."
nvidia-smi --query-gpu=memory.free,memory.total --format=csv

# Install Ollama if not already installed
if ! command -v ollama &> /dev/null; then
    echo "📥 Installing Ollama..."
    curl -fsSL https://ollama.com/install.sh | sh
else
    echo "✅ Ollama is already installed."
fi

# Verify Ollama binary
if [ ! -f "/usr/local/bin/ollama" ]; then
    echo "❌ Ollama binary not found. Exiting."
    exit 1
fi

# Move existing Ollama data if needed
OLLAMA_DIR="/usr/share/ollama/.ollama"
NEW_OLLAMA_DIR="/data/ollama"
if [ -d "$OLLAMA_DIR" ]; then
    echo "📦 Moving existing Ollama data to $NEW_OLLAMA_DIR..."
    sudo mv "$OLLAMA_DIR" "$NEW_OLLAMA_DIR"
fi

# Create a symlink for Ollama storage
echo "🔗 Setting up Ollama storage..."
sudo ln -sf "$NEW_OLLAMA_DIR" "$OLLAMA_DIR"

# Set environment variables for GPU execution
echo "⚙️ Configuring GPU usage for Ollama..."
export OLLAMA_CUDA_VISIBLE_DEVICES=0
echo 'export OLLAMA_CUDA_VISIBLE_DEVICES=0' >> ~/.bashrc

# Configure systemd for Ollama
echo "⚙️ Configuring Ollama for remote access..."
sudo tee /etc/systemd/system/ollama.service > /dev/null <<EOF
[Unit]
Description=Ollama AI Model Service
After=network.target

[Service]
ExecStart=/usr/local/bin/ollama serve
Restart=always
User=root
Environment="OLLAMA_HOST=0.0.0.0:11434"
Environment="OLLAMA_ORIGINS=*"
Environment="CUDA_VISIBLE_DEVICES=0"

[Install]
WantedBy=multi-user.target
EOF

# Reload systemd and enable Ollama service
echo "🚀 Starting and enabling Ollama..."
sudo systemctl daemon-reload
sudo systemctl enable ollama
sudo systemctl restart ollama

# Confirm Ollama is running
sleep 3
if ! pgrep -x "ollama" > /dev/null; then
    echo "❌ Ollama failed to start. Check logs with: journalctl -u ollama --no-pager --lines=50"
    exit 1
fi

# Configure UFW to allow external access
echo "🔓 Configuring firewall to allow remote connections..."
sudo ufw allow 11434/tcp
sudo ufw reload
sudo ufw status

# Pull DeepSeek model if not already present
echo "📥 Checking if DeepSeek-r1:671b is available..."
if ! ollama list | grep -q "deepseek-r1:671b"; then
    echo "📥 Downloading deepseek-r1:671b model..."
    ollama pull deepseek-r1:671b
fi

# Confirm model is available
if ! ollama list | grep -q "deepseek-r1:671b"; then
    echo "❌ Failed to download deepseek-r1:671b. Please check storage space and internet connection."
    exit 1
fi

# Wait for the model to be fully loaded
echo "⏳ Waiting for DeepSeek model to be fully loaded..."
while ! ollama run deepseek-r1:671b --verbose 2>&1 | grep -q "model loaded"; do
    echo "⏳ Model still loading..."
    sleep 10
done
echo "✅ DeepSeek model is fully loaded and ready!"

# Verify Ollama is using GPU
echo "🖥 Verifying Ollama is using GPU..."
ollama run deepseek-r1:671b --verbose | grep "CUDA"

# Log success messages
echo "✅ Ollama is installed, running in the background, and ready for Chatbox!"
echo "✅ DeepSeek-r1:671b model is downloaded and ready to use!"
echo "➡️ You can now connect Chatbox by pointing it to: http://<your-server-ip>:11434"

# Start watching logs for real-time debugging
echo "📜 Watching Ollama logs in real-time (Press Ctrl+C to exit)..."
journalctl -u ollama -f --no-pager