Medical GraphRAG Assistant

#!/usr/bin/env bash # # Deploy NVIDIA NIM Vision Container # # Deploys NVIDIA NIM Vision service for multi-modal image embeddings. # Requires: # - NVIDIA GPU with drivers installed # - Docker with GPU runtime configured # - NVIDIA API key set in environment # # Usage: # ./scripts/aws/deploy-nim-vision.sh [--remote PUBLIC_IP] [--ssh-key SSH_KEY_PATH] [--force-recreate] # # Options: # --remote Deploy to remote instance via SSH # --ssh-key Path to SSH private key (required with --remote) # --force-recreate Stop and remove existing container before deploying # # Environment Variables: # NVIDIA_API_KEY NVIDIA NGC API key (required) # NGC_API_KEY Alternative name for NVIDIA_API_KEY # # Exit Codes: # 0 Success # 1 Missing requirements or deployment failed # # Example: # ./scripts/aws/deploy-nim-vision.sh # ./scripts/aws/deploy-nim-vision.sh --remote 34.xxx.xxx.xxx --ssh-key ~/.ssh/my-key.pem set -euo pipefail # Colors for output RED='\033[0;31m' GREEN='\033[0;32m' YELLOW='\033[1;33m' BLUE='\033[0;34m' NC='\033[0m' # No Color # Configuration CONTAINER_NAME="nim-vision" NIM_VISION_IMAGE="nvcr.io/nim/nvidia/nv-clip-vit:latest" VISION_PORT=8002 INTERNAL_PORT=8000 SHARED_MEMORY="8g" HEALTH_ENDPOINT_PATH="/health" HEALTH_CHECK_TIMEOUT=300 # 5 minutes HEALTH_CHECK_INTERVAL=15 # 15 seconds between checks # Parse command-line arguments REMOTE_HOST="" SSH_KEY_PATH="" FORCE_RECREATE=false while [[ $# -gt 0 ]]; do case $1 in --remote) REMOTE_HOST="$2" shift 2 ;; --ssh-key) SSH_KEY_PATH="$2" shift 2 ;; --force-recreate) FORCE_RECREATE=true shift ;; --help) head -n 28 "$0" | tail -n +3 exit 0 ;; *) echo -e "${RED}Unknown option: $1${NC}" echo "Use --help for usage information" exit 1 ;; esac done # Validation if [[ -n "$REMOTE_HOST" && -z "$SSH_KEY_PATH" ]]; then echo -e "${RED}Error: --ssh-key required when using --remote${NC}" exit 1 fi if [[ -n "$SSH_KEY_PATH" && ! -f "$SSH_KEY_PATH" ]]; then echo -e "${RED}Error: SSH key not found: $SSH_KEY_PATH${NC}" exit 1 fi # Function to execute commands (local or remote) execute_cmd() { if [[ -n "$REMOTE_HOST" ]]; then ssh -i "$SSH_KEY_PATH" -o StrictHostKeyChecking=no "ubuntu@$REMOTE_HOST" "$@" else bash -c "$@" fi } # Function to check if API key is set check_api_key() { local check_cmd=' if [ -z "${NVIDIA_API_KEY:-}" ] && [ -z "${NGC_API_KEY:-}" ]; then echo "MISSING" else echo "OK" fi ' local result result=$(execute_cmd "$check_cmd") if [[ "$result" == "MISSING" ]]; then echo -e "${RED}✗ NVIDIA API key not found${NC}" echo " Set NVIDIA_API_KEY or NGC_API_KEY environment variable" echo " Get your API key at: https://org.ngc.nvidia.com/setup/api-key" return 1 fi echo -e "${GREEN}✓ NVIDIA API key found${NC}" return 0 } # Function to check GPU availability check_gpu() { echo -e "${BLUE}→ Verifying GPU availability...${NC}" local gpu_check_cmd=' if ! command -v nvidia-smi &> /dev/null; then echo "MISSING" elif ! nvidia-smi &> /dev/null; then echo "ERROR" else echo "OK" fi ' local result result=$(execute_cmd "$gpu_check_cmd") if [[ "$result" == "MISSING" ]]; then echo -e "${RED}✗ nvidia-smi not found${NC}" echo " Install NVIDIA drivers first: ./scripts/aws/install-gpu-drivers.sh" return 1 elif [[ "$result" == "ERROR" ]]; then echo -e "${RED}✗ GPU not accessible${NC}" echo " Reboot may be required after driver installation" return 1 fi echo -e "${GREEN}✓ GPU accessible${NC}" return 0 } # Function to handle existing container handle_existing_container() { echo -e "${BLUE}→ Checking for existing container...${NC}" local container_exists container_exists=$(execute_cmd "docker ps -a --filter name=^/${CONTAINER_NAME}$ --format '{{.Names}}' 2>/dev/null || true") if [[ -n "$container_exists" ]]; then if [[ "$FORCE_RECREATE" == true ]]; then echo -e "${YELLOW}! Removing existing container: $CONTAINER_NAME${NC}" execute_cmd "docker stop $CONTAINER_NAME 2>/dev/null || true" execute_cmd "docker rm $CONTAINER_NAME 2>/dev/null || true" echo -e "${GREEN}✓ Existing container removed${NC}" else echo -e "${YELLOW}! Container $CONTAINER_NAME already exists${NC}" echo " Use --force-recreate to remove and redeploy" # Check if it's running local is_running is_running=$(execute_cmd "docker ps --filter name=^/${CONTAINER_NAME}$ --format '{{.Names}}' 2>/dev/null || true") if [[ -n "$is_running" ]]; then echo -e "${GREEN}✓ Container is already running${NC}" return 0 else echo " Starting stopped container..." execute_cmd "docker start $CONTAINER_NAME" echo -e "${GREEN}✓ Container started${NC}" return 0 fi fi fi } # Function to pull NIM Vision image pull_image() { echo -e "${BLUE}→ Pulling NVIDIA NIM Vision image...${NC}" echo " Image: $NIM_VISION_IMAGE" echo " This may take several minutes..." if execute_cmd "docker pull $NIM_VISION_IMAGE 2>&1" | grep -q "Downloaded\|Image is up to date"; then echo -e "${GREEN}✓ Image pulled${NC}" return 0 else echo -e "${RED}✗ Failed to pull image${NC}" echo " Check NGC API key and network connectivity" return 1 fi } # Function to start NIM Vision container start_container() { echo -e "${BLUE}→ Starting NVIDIA NIM Vision container...${NC}" local run_cmd=" docker run -d \\ --name $CONTAINER_NAME \\ --gpus all \\ --restart unless-stopped \\ -p $VISION_PORT:$INTERNAL_PORT \\ -e NGC_API_KEY=\${NVIDIA_API_KEY:-\${NGC_API_KEY}} \\ -e NIM_MODEL_PROFILE=auto \\ --shm-size=$SHARED_MEMORY \\ $NIM_VISION_IMAGE " if execute_cmd "$run_cmd" &> /dev/null; then echo -e "${GREEN}✓ Container started${NC}" echo " Container name: $CONTAINER_NAME" echo " Port mapping: $VISION_PORT:$INTERNAL_PORT" return 0 else echo -e "${RED}✗ Failed to start container${NC}" execute_cmd "docker logs $CONTAINER_NAME --tail 50 2>&1 || true" return 1 fi } # Function to wait for service health wait_for_health() { echo -e "${BLUE}→ Waiting for NIM Vision to initialize...${NC}" echo " This may take 3-5 minutes (model download on first run)" local elapsed=0 local endpoint if [[ -n "$REMOTE_HOST" ]]; then endpoint="http://$REMOTE_HOST:$VISION_PORT$HEALTH_ENDPOINT_PATH" else endpoint="http://localhost:$VISION_PORT$HEALTH_ENDPOINT_PATH" fi while [[ $elapsed -lt $HEALTH_CHECK_TIMEOUT ]]; do # Check if container is still running local is_running is_running=$(execute_cmd "docker ps --filter name=^/${CONTAINER_NAME}$ --format '{{.Names}}' 2>/dev/null || true") if [[ -z "$is_running" ]]; then echo -e "${RED}✗ Container stopped unexpectedly${NC}" echo " Checking logs..." execute_cmd "docker logs $CONTAINER_NAME --tail 50" return 1 fi # Try health check if execute_cmd "command -v curl &> /dev/null" &> /dev/null; then if execute_cmd "curl -sf $endpoint" &> /dev/null; then echo -e "${GREEN}✓ NIM Vision service is healthy${NC}" return 0 fi fi # Show progress echo " Still initializing... (${elapsed}s/${HEALTH_CHECK_TIMEOUT}s)" sleep $HEALTH_CHECK_INTERVAL elapsed=$((elapsed + HEALTH_CHECK_INTERVAL)) done echo -e "${YELLOW}! Health check timeout${NC}" echo " Service may still be initializing - check logs:" echo " docker logs $CONTAINER_NAME" return 0 # Don't fail deployment, just warn } # Function to show deployment info show_deployment_info() { local host if [[ -n "$REMOTE_HOST" ]]; then host="$REMOTE_HOST" else host="localhost" fi cat << EOF ${GREEN}========================================== NVIDIA NIM Vision Deployed ==========================================${NC} Model: CLIP Vision Transformer Endpoint: http://$host:$VISION_PORT Health: http://$host:$VISION_PORT/health Container Details: Name: $CONTAINER_NAME Port: $VISION_PORT (external) → $INTERNAL_PORT (internal) GPU: Enabled Shared Memory: $SHARED_MEMORY Test with curl: curl -X POST http://$host:$VISION_PORT/v1/embeddings \\ -H "Content-Type: application/json" \\ -d '{ "input": "base64_encoded_image_here", "model": "nv-clip-vit" }' Check logs: docker logs $CONTAINER_NAME Monitor GPU usage: nvidia-smi --loop=1 EOF } # Main deployment flow main() { echo "" echo -e "${BLUE}╔══════════════════════════════════════════════════════════════╗${NC}" echo -e "${BLUE}║ NVIDIA NIM Vision Deployment ║${NC}" echo -e "${BLUE}╚══════════════════════════════════════════════════════════════╝${NC}" echo "" if [[ -n "$REMOTE_HOST" ]]; then echo -e "${BLUE}Remote deployment to: $REMOTE_HOST${NC}" echo "" fi # Run deployment steps check_api_key || exit 1 check_gpu || exit 1 handle_existing_container || exit 1 pull_image || exit 1 start_container || exit 1 wait_for_health || echo -e "${YELLOW}(Continuing despite health check timeout)${NC}" show_deployment_info echo -e "${GREEN}✅ NIM Vision deployment complete!${NC}" echo "" } # Execute main function main