miku-discord/setup-dual-gpu.sh

#!/bin/bash
# Setup script for dual GPU configuration (NVIDIA + AMD RX 6800)

# Note: Not using 'set -e' to allow arithmetic operations

echo "================================================"
echo "Miku Discord Bot - Dual GPU Setup"
echo "================================================"
echo ""

# Colors for output
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
NC='\033[0m' # No Color

# Check if ROCm is installed on host (optional - Docker container has it)
echo -e "${YELLOW}Checking ROCm installation (optional)...${NC}"
if command -v rocm-smi &> /dev/null; then
    echo -e "${GREEN}✓ ROCm is installed on host${NC}"
    rocm-smi --version
    
    # Check AMD GPU via rocm-smi
    echo ""
    echo -e "${YELLOW}Checking AMD GPU...${NC}"
    if rocm-smi -i 0 &> /dev/null; then
        echo -e "${GREEN}✓ AMD GPU detected via rocm-smi${NC}"
        rocm-smi -i 0 --showproductname
    else
        echo -e "${YELLOW}! AMD GPU check via rocm-smi failed${NC}"
    fi
else
    echo -e "${YELLOW}! ROCm not found on host (this is OK - Docker container includes ROCm)${NC}"
    
    # Fallback: Check AMD GPU via lspci
    echo ""
    echo -e "${YELLOW}Checking AMD GPU via lspci...${NC}"
    if lspci | grep -i "VGA.*AMD\|VGA.*Radeon" &> /dev/null; then
        echo -e "${GREEN}✓ AMD GPU detected${NC}"
        lspci | grep -i "VGA.*AMD\|VGA.*Radeon"
    else
        echo -e "${YELLOW}! AMD GPU not detected via lspci${NC}"
        echo "Note: AMD RX 6800 should be visible in lspci output"
    fi
fi

# Check NVIDIA GPU
echo ""
echo -e "${YELLOW}Checking NVIDIA GPU...${NC}"
if command -v nvidia-smi &> /dev/null; then
    echo -e "${GREEN}✓ NVIDIA GPU detected${NC}"
    nvidia-smi --query-gpu=name --format=csv,noheader
else
    echo -e "${RED}✗ NVIDIA GPU not found${NC}"
    echo "Warning: Primary GPU (NVIDIA) not detected"
fi

# Check Docker
echo ""
echo -e "${YELLOW}Checking Docker...${NC}"
if command -v docker &> /dev/null; then
    echo -e "${GREEN}✓ Docker is installed${NC}"
    docker --version
else
    echo -e "${RED}✗ Docker not found${NC}"
    exit 1
fi

# Check device permissions
echo ""
echo -e "${YELLOW}Checking device permissions...${NC}"
if [ -c /dev/kfd ] && [ -r /dev/kfd ]; then
    echo -e "${GREEN}✓ /dev/kfd accessible${NC}"
else
    echo -e "${RED}✗ /dev/kfd not accessible${NC}"
    echo "You may need to add your user to the 'render' group:"
    echo "  sudo usermod -aG render \$USER"
fi

if [ -d /dev/dri ]; then
    echo -e "${GREEN}✓ /dev/dri exists${NC}"
else
    echo -e "${RED}✗ /dev/dri not found${NC}"
fi

# Check if models exist
echo ""
echo -e "${YELLOW}Checking model files...${NC}"
MODEL_DIR="./models"
if [ -d "$MODEL_DIR" ]; then
    echo -e "${GREEN}✓ Models directory exists${NC}"
    
    # Check for specific models
    models_found=0
    if [ -f "$MODEL_DIR/Llama-3.1-8B-Instruct-UD-Q4_K_XL.gguf" ]; then
        echo -e "${GREEN}  ✓ Llama 3.1 8B model found${NC}"
        ((models_found++))
    fi
    
    if [ -f "$MODEL_DIR/DarkIdol-Llama-3.1-8B-Instruct-1.3-Uncensored_Q4_K_M.gguf" ]; then
        echo -e "${GREEN}  ✓ DarkIdol model found${NC}"
        ((models_found++))
    fi
    
    if [ -f "$MODEL_DIR/moondream2-text-model-f16_ct-vicuna.gguf" ]; then
        echo -e "${GREEN}  ✓ Moondream2 model found${NC}"
        ((models_found++))
    fi
    
    if [ $models_found -eq 0 ]; then
        echo -e "${YELLOW}  ! No models found in $MODEL_DIR${NC}"
        echo "  Please download GGUF models before starting"
    fi
else
    echo -e "${YELLOW}! Models directory not found${NC}"
    echo "Creating models directory..."
    mkdir -p "$MODEL_DIR"
fi

echo ""
echo "================================================"
echo "Setup Commands"
echo "================================================"
echo ""

echo "1. Build the AMD ROCm container:"
echo -e "   ${GREEN}docker compose build llama-swap-amd${NC}"
echo ""

echo "2. Start both GPU services:"
echo -e "   ${GREEN}docker compose up -d llama-swap llama-swap-amd${NC}"
echo ""

echo "3. Check status:"
echo -e "   ${GREEN}docker compose ps${NC}"
echo -e "   ${GREEN}docker compose logs -f llama-swap-amd${NC}"
echo ""

echo "4. Test endpoints:"
echo -e "   ${GREEN}curl http://localhost:8090/health  # NVIDIA GPU${NC}"
echo -e "   ${GREEN}curl http://localhost:8091/health  # AMD GPU${NC}"
echo ""

echo "5. Monitor GPU usage:"
echo -e "   ${GREEN}watch -n 1 nvidia-smi              # NVIDIA${NC}"
echo -e "   ${GREEN}watch -n 1 rocm-smi                # AMD${NC}"
echo ""

echo "================================================"
echo "Optional: Enable AMD GPU preference"
echo "================================================"
echo ""
echo "To prefer AMD GPU for load balancing, add to docker-compose.yml:"
echo "  environment:"
echo "    - PREFER_AMD_GPU=true"
echo ""

echo -e "${GREEN}Setup check complete!${NC}"
echo "See DUAL_GPU_SETUP.md for detailed documentation"