@aws/ml-container-creator 0.2.0

package/templates/do/test

@@ -0,0 +1,1147 @@
+#!/bin/bash
+# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+set -e
+set -u
+set -o pipefail
+
+# Source configuration
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/config"
+
+<% if (deploymentTarget === 'managed-inference') { %>
+# ============================================================
+# SageMaker Managed Inference Testing
+# ============================================================
+
+# Parse arguments
+ENDPOINT_NAME="${1:-${ENDPOINT_NAME:-}}"
+
+if [ -z "${ENDPOINT_NAME}" ]; then
+    echo "🧪 Testing local container at localhost:8080"
+    echo "   Project: ${PROJECT_NAME}"
+    echo "   Framework: ${FRAMEWORK}"
+    echo "   Model server: ${MODEL_SERVER}"
+    TARGET_URL="http://localhost:8080"
+    TEST_MODE="local"
+else
+    echo "🧪 Testing SageMaker endpoint: ${ENDPOINT_NAME}"
+    echo "   Project: ${PROJECT_NAME}"
+    echo "   Framework: ${FRAMEWORK}"
+    echo "   Model server: ${MODEL_SERVER}"
+    echo "   Region: ${AWS_REGION}"
+    TEST_MODE="sagemaker"
+fi
+
+echo ""
+
+# Test 1: Health check (/ping)
+echo "🔍 Test 1: Health check"
+if [ "${TEST_MODE}" = "local" ]; then
+    echo "   Sending GET request to ${TARGET_URL}/ping"
+    
+    if ! PING_RESPONSE=$(curl -s -w "\n%{http_code}" -X GET "${TARGET_URL}/ping" 2>&1); then
+        echo "❌ Health check failed: Could not connect to local container"
+        echo "   Make sure the container is running: ./do/run"
+        exit 1
+    fi
+    
+    HTTP_CODE=$(echo "${PING_RESPONSE}" | tail -n1)
+    RESPONSE_BODY=$(echo "${PING_RESPONSE}" | sed '$d')
+    
+    if [ "${HTTP_CODE}" = "200" ]; then
+        echo "✅ Health check passed (HTTP ${HTTP_CODE})"
+    else
+        echo "❌ Health check failed (HTTP ${HTTP_CODE})"
+        echo "   Response: ${RESPONSE_BODY}"
+        exit 1
+    fi
+else
+    # For SageMaker endpoints, /ping is not directly accessible
+    # We'll verify the endpoint exists and is InService
+    echo "   Checking endpoint status..."
+    
+    if ! ENDPOINT_STATUS=$(aws sagemaker describe-endpoint \
+        --endpoint-name "${ENDPOINT_NAME}" \
+        --region "${AWS_REGION}" \
+        --query 'EndpointStatus' \
+        --output text 2>&1); then
+        echo "❌ Endpoint not found or not accessible"
+        echo "   Error: ${ENDPOINT_STATUS}"
+        exit 1
+    fi
+    
+    if [ "${ENDPOINT_STATUS}" = "InService" ]; then
+        echo "✅ Endpoint is InService"
+    else
+        echo "❌ Endpoint is not InService (Status: ${ENDPOINT_STATUS})"
+        exit 1
+    fi
+fi
+
+echo ""
+
+# Test 2: Inference request (/invocations)
+echo "🔍 Test 2: Inference request"
+
+# Create framework-specific test payload
+case "${FRAMEWORK}" in
+    sklearn|xgboost)
+        # Traditional ML: JSON with instances array
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    tensorflow)
+        # TensorFlow: JSON with instances array
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    transformers)
+        # Transformers: payload format depends on model server
+        case "${MODEL_SERVER}" in
+            vllm|sglang)
+                # OpenAI-compatible chat completions format
+                # For S3/JumpStart models, vLLM registers the model under the local path
+                VLLM_MODEL_NAME="${MODEL_NAME}"
+                if [[ "${MODEL_NAME}" == jumpstart://* ]] || [[ "${MODEL_NAME}" == jumpstart-hub://* ]] || [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                    VLLM_MODEL_NAME="/opt/ml/model"
+                fi
+                TEST_PAYLOAD='{"model": "'"${VLLM_MODEL_NAME}"'", "messages": [{"role": "user", "content": "What is machine learning?"}], "max_tokens": 50, "temperature": 0.7}'
+                echo "   Payload: OpenAI-compatible chat completion request"
+                echo "   Model: ${VLLM_MODEL_NAME}"
+                ;;
+            *)
+                # HuggingFace-style format for LMI, DJL, TensorRT-LLM
+                TEST_PAYLOAD='{"inputs": "What is machine learning?", "parameters": {"max_new_tokens": 50, "temperature": 0.7}}'
+                echo "   Payload: HuggingFace-style text generation request"
+                ;;
+        esac
+        ;;
+    diffusors)
+        # Diffusors: OpenAI DALL-E compatible image generation request
+        TEST_PAYLOAD='{"prompt": "A white cat", "n": 1, "size": "512x512"}'
+        echo "   Payload: OpenAI DALL-E compatible image generation request"
+        ;;
+    *)
+        echo "❌ Unknown framework: ${FRAMEWORK}"
+        exit 3
+        ;;
+esac
+
+if [ "${TEST_MODE}" = "local" ]; then
+    if [ "${FRAMEWORK}" = "diffusors" ]; then
+        echo "   Sending POST request to ${TARGET_URL}/v1/images/generations"
+        
+        if ! INVOKE_RESPONSE=$(curl -s -m 120 -w "\n%{http_code}" -X POST "${TARGET_URL}/v1/images/generations" \
+            -H "Content-Type: application/json" \
+            -d "${TEST_PAYLOAD}" 2>&1); then
+            echo "❌ Inference request failed: Could not connect to local container"
+            exit 1
+        fi
+        
+        HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+        RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+        
+        if [ "${HTTP_CODE}" = "200" ]; then
+            # Validate response contains data array with b64_json
+            if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json
+resp = json.load(sys.stdin)
+assert 'data' in resp, 'Missing data array'
+assert len(resp['data']) > 0, 'Empty data array'
+assert 'b64_json' in resp['data'][0], 'Missing b64_json in data entry'
+" 2>/dev/null; then
+                echo "✅ Image generation successful (HTTP ${HTTP_CODE})"
+                echo "   Response contains valid data array with b64_json image"
+                
+                # Save generated image to file
+                OUTPUT_IMAGE="${SCRIPT_DIR}/../test_output.png"
+                if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json, base64
+resp = json.load(sys.stdin)
+img_data = base64.b64decode(resp['data'][0]['b64_json'])
+with open('${OUTPUT_IMAGE}', 'wb') as f:
+    f.write(img_data)
+" 2>/dev/null; then
+                    echo "   🖼️  Image saved to: ${OUTPUT_IMAGE}"
+                fi
+            else
+                echo "⚠️  Image generation returned HTTP ${HTTP_CODE} but response format unexpected"
+                echo "   Response preview: ${RESPONSE_BODY:0:200}"
+            fi
+        else
+            echo "❌ Image generation failed (HTTP ${HTTP_CODE})"
+            echo "   Response: ${RESPONSE_BODY}"
+            exit 1
+        fi
+    else
+        echo "   Sending POST request to ${TARGET_URL}/invocations"
+        
+        if ! INVOKE_RESPONSE=$(curl -s -w "\n%{http_code}" -X POST "${TARGET_URL}/invocations" \
+            -H "Content-Type: application/json" \
+            -d "${TEST_PAYLOAD}" 2>&1); then
+            echo "❌ Inference request failed: Could not connect to local container"
+            exit 1
+        fi
+        
+        HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+        RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+        
+        if [ "${HTTP_CODE}" = "200" ]; then
+            echo "✅ Inference request successful (HTTP ${HTTP_CODE})"
+            echo "   Response preview: ${RESPONSE_BODY:0:200}"
+            if [ ${#RESPONSE_BODY} -gt 200 ]; then
+                echo "   (truncated, full response is ${#RESPONSE_BODY} characters)"
+            fi
+        else
+            echo "❌ Inference request failed (HTTP ${HTTP_CODE})"
+            echo "   Response: ${RESPONSE_BODY}"
+            exit 1
+        fi
+    fi
+else
+    echo "   Invoking SageMaker endpoint..."
+    
+    # Create temporary file for payload
+    TEMP_PAYLOAD=$(mktemp)
+    echo "${TEST_PAYLOAD}" > "${TEMP_PAYLOAD}"
+    
+    # Create temporary file for response
+    TEMP_RESPONSE=$(mktemp)
+    
+    # Invoke endpoint via inference component
+    IC_NAME="${INFERENCE_COMPONENT_NAME:-}"
+    INVOKE_ARGS=(
+        --endpoint-name "${ENDPOINT_NAME}"
+        --region "${AWS_REGION}"
+        --content-type "application/json"
+        --body "fileb://${TEMP_PAYLOAD}"
+    )
+    if [ -n "${IC_NAME}" ]; then
+        INVOKE_ARGS+=(--inference-component-name "${IC_NAME}")
+        echo "   Inference component: ${IC_NAME}"
+    fi
+
+    INVOKE_ERROR=$(mktemp)
+    if ! aws sagemaker-runtime invoke-endpoint \
+        "${INVOKE_ARGS[@]}" \
+        "${TEMP_RESPONSE}" 2>"${INVOKE_ERROR}"; then
+        echo "❌ Inference request failed"
+        echo "   Error: $(cat "${INVOKE_ERROR}")"
+        rm -f "${TEMP_PAYLOAD}" "${TEMP_RESPONSE}" "${INVOKE_ERROR}"
+        exit 1
+    fi
+    rm -f "${INVOKE_ERROR}"
+    
+    # Read response
+    RESPONSE_BODY=$(cat "${TEMP_RESPONSE}")
+    
+    # Clean up temp files
+    rm -f "${TEMP_PAYLOAD}" "${TEMP_RESPONSE}"
+    
+    if [ "${FRAMEWORK}" = "diffusors" ]; then
+        # Validate response contains data array with b64_json
+        if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json
+resp = json.load(sys.stdin)
+assert 'data' in resp, 'Missing data array'
+assert len(resp['data']) > 0, 'Empty data array'
+assert 'b64_json' in resp['data'][0], 'Missing b64_json in data entry'
+" 2>/dev/null; then
+            echo "✅ Image generation successful"
+            echo "   Response contains valid data array with b64_json image"
+            
+            # Save generated image to file
+            OUTPUT_IMAGE="${SCRIPT_DIR}/../test_output.png"
+            if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json, base64
+resp = json.load(sys.stdin)
+img_data = base64.b64decode(resp['data'][0]['b64_json'])
+with open('${OUTPUT_IMAGE}', 'wb') as f:
+    f.write(img_data)
+" 2>/dev/null; then
+                echo "   🖼️  Image saved to: ${OUTPUT_IMAGE}"
+            fi
+        else
+            echo "⚠️  Image generation returned but response format unexpected"
+            echo "   Response preview: ${RESPONSE_BODY:0:200}"
+        fi
+    else
+        echo "✅ Inference request successful"
+        echo "   Response preview: ${RESPONSE_BODY:0:200}"
+        if [ ${#RESPONSE_BODY} -gt 200 ]; then
+            echo "   (truncated, full response is ${#RESPONSE_BODY} characters)"
+        fi
+    fi
+fi
+
+echo ""
+echo "✅ All tests passed!"
+echo ""
+
+if [ "${TEST_MODE}" = "local" ]; then
+    echo "Next steps:"
+    echo "  • Push to ECR: ./do/push"
+    echo "  • Deploy to SageMaker: ./do/deploy"
+else
+    echo "Endpoint is ready for production use!"
+    echo "  • Endpoint name: ${ENDPOINT_NAME}"
+    echo "  • Region: ${AWS_REGION}"
+    echo ""
+    echo "📝 Register this deployment:"
+    echo "  ./do/register"
+fi
+
+<% } else if (deploymentTarget === 'async-inference') { %>
+# ============================================================
+# SageMaker Managed Inference - Async Testing
+# ============================================================
+
+# Parse arguments
+ENDPOINT_NAME="${1:-${ENDPOINT_NAME:-}}"
+
+if [ -z "${ENDPOINT_NAME}" ]; then
+    echo "🧪 Testing local container at localhost:8080"
+    echo "   Project: ${PROJECT_NAME}"
+    echo "   Framework: ${FRAMEWORK}"
+    echo "   Model server: ${MODEL_SERVER}"
+    TARGET_URL="http://localhost:8080"
+    TEST_MODE="local"
+else
+    echo "🧪 Testing SageMaker async endpoint: ${ENDPOINT_NAME}"
+    echo "   Project: ${PROJECT_NAME}"
+    echo "   Framework: ${FRAMEWORK}"
+    echo "   Model server: ${MODEL_SERVER}"
+    echo "   Region: ${AWS_REGION}"
+    echo "   S3 output: ${ASYNC_S3_OUTPUT_PATH}"
+    TEST_MODE="sagemaker"
+fi
+
+echo ""
+
+# Test 1: Health check (/ping)
+echo "🔍 Test 1: Health check"
+if [ "${TEST_MODE}" = "local" ]; then
+    echo "   Sending GET request to ${TARGET_URL}/ping"
+    
+    if ! PING_RESPONSE=$(curl -s -w "\n%{http_code}" -X GET "${TARGET_URL}/ping" 2>&1); then
+        echo "❌ Health check failed: Could not connect to local container"
+        echo "   Make sure the container is running: ./do/run"
+        exit 1
+    fi
+    
+    HTTP_CODE=$(echo "${PING_RESPONSE}" | tail -n1)
+    RESPONSE_BODY=$(echo "${PING_RESPONSE}" | sed '$d')
+    
+    if [ "${HTTP_CODE}" = "200" ]; then
+        echo "✅ Health check passed (HTTP ${HTTP_CODE})"
+    else
+        echo "❌ Health check failed (HTTP ${HTTP_CODE})"
+        echo "   Response: ${RESPONSE_BODY}"
+        exit 1
+    fi
+else
+    # For SageMaker endpoints, verify the endpoint exists and is InService
+    echo "   Checking endpoint status..."
+    
+    if ! ENDPOINT_STATUS=$(aws sagemaker describe-endpoint \
+        --endpoint-name "${ENDPOINT_NAME}" \
+        --region "${AWS_REGION}" \
+        --query 'EndpointStatus' \
+        --output text 2>&1); then
+        echo "❌ Endpoint not found or not accessible"
+        echo "   Error: ${ENDPOINT_STATUS}"
+        exit 1
+    fi
+    
+    if [ "${ENDPOINT_STATUS}" = "InService" ]; then
+        echo "✅ Endpoint is InService"
+    else
+        echo "❌ Endpoint is not InService (Status: ${ENDPOINT_STATUS})"
+        exit 1
+    fi
+fi
+
+echo ""
+
+# Test 2: Inference request
+echo "🔍 Test 2: Inference request"
+
+# Create framework-specific test payload
+case "${FRAMEWORK}" in
+    sklearn|xgboost)
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    tensorflow)
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    transformers)
+        case "${MODEL_SERVER}" in
+            vllm|sglang)
+                VLLM_MODEL_NAME="${MODEL_NAME}"
+                if [[ "${MODEL_NAME}" == jumpstart://* ]] || [[ "${MODEL_NAME}" == jumpstart-hub://* ]] || [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                    VLLM_MODEL_NAME="/opt/ml/model"
+                fi
+                TEST_PAYLOAD='{"model": "'"${VLLM_MODEL_NAME}"'", "messages": [{"role": "user", "content": "What is machine learning?"}], "max_tokens": 50, "temperature": 0.7}'
+                echo "   Payload: OpenAI-compatible chat completion request"
+                echo "   Model: ${VLLM_MODEL_NAME}"
+                ;;
+            *)
+                TEST_PAYLOAD='{"inputs": "What is machine learning?", "parameters": {"max_new_tokens": 50, "temperature": 0.7}}'
+                echo "   Payload: HuggingFace-style text generation request"
+                ;;
+        esac
+        ;;
+    diffusors)
+        TEST_PAYLOAD='{"prompt": "A white cat", "n": 1, "size": "512x512"}'
+        echo "   Payload: OpenAI DALL-E compatible image generation request"
+        ;;
+    *)
+        echo "❌ Unknown framework: ${FRAMEWORK}"
+        exit 3
+        ;;
+esac
+
+if [ "${TEST_MODE}" = "local" ]; then
+    if [ "${FRAMEWORK}" = "diffusors" ]; then
+        echo "   Sending POST request to ${TARGET_URL}/v1/images/generations"
+        
+        if ! INVOKE_RESPONSE=$(curl -s -m 120 -w "\n%{http_code}" -X POST "${TARGET_URL}/v1/images/generations" \
+            -H "Content-Type: application/json" \
+            -d "${TEST_PAYLOAD}" 2>&1); then
+            echo "❌ Inference request failed: Could not connect to local container"
+            exit 1
+        fi
+        
+        HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+        RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+        
+        if [ "${HTTP_CODE}" = "200" ]; then
+            if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json
+resp = json.load(sys.stdin)
+assert 'data' in resp, 'Missing data array'
+assert len(resp['data']) > 0, 'Empty data array'
+assert 'b64_json' in resp['data'][0], 'Missing b64_json in data entry'
+" 2>/dev/null; then
+                echo "✅ Image generation successful (HTTP ${HTTP_CODE})"
+                echo "   Response contains valid data array with b64_json image"
+                
+                # Save generated image to file
+                OUTPUT_IMAGE="${SCRIPT_DIR}/../test_output.png"
+                if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json, base64
+resp = json.load(sys.stdin)
+img_data = base64.b64decode(resp['data'][0]['b64_json'])
+with open('${OUTPUT_IMAGE}', 'wb') as f:
+    f.write(img_data)
+" 2>/dev/null; then
+                    echo "   🖼️  Image saved to: ${OUTPUT_IMAGE}"
+                fi
+            else
+                echo "⚠️  Image generation returned HTTP ${HTTP_CODE} but response format unexpected"
+                echo "   Response preview: ${RESPONSE_BODY:0:200}"
+            fi
+        else
+            echo "❌ Image generation failed (HTTP ${HTTP_CODE})"
+            echo "   Response: ${RESPONSE_BODY}"
+            exit 1
+        fi
+    else
+        echo "   Sending POST request to ${TARGET_URL}/invocations"
+        
+        if ! INVOKE_RESPONSE=$(curl -s -w "\n%{http_code}" -X POST "${TARGET_URL}/invocations" \
+            -H "Content-Type: application/json" \
+            -d "${TEST_PAYLOAD}" 2>&1); then
+            echo "❌ Inference request failed: Could not connect to local container"
+            exit 1
+        fi
+        
+        HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+        RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+        
+        if [ "${HTTP_CODE}" = "200" ]; then
+            echo "✅ Inference request successful (HTTP ${HTTP_CODE})"
+            echo "   Response preview: ${RESPONSE_BODY:0:200}"
+            if [ ${#RESPONSE_BODY} -gt 200 ]; then
+                echo "   (truncated, full response is ${#RESPONSE_BODY} characters)"
+            fi
+        else
+            echo "❌ Inference request failed (HTTP ${HTTP_CODE})"
+            echo "   Response: ${RESPONSE_BODY}"
+            exit 1
+        fi
+    fi
+else
+    # SageMaker async invocation: upload payload to S3, invoke async, poll for result
+    echo "   Uploading test payload to S3..."
+    
+    # Create temporary file for payload
+    TEMP_PAYLOAD=$(mktemp)
+    echo "${TEST_PAYLOAD}" > "${TEMP_PAYLOAD}"
+    
+    # Upload payload to S3 input location
+    ASYNC_INPUT_KEY="${PROJECT_NAME}/input/test-payload-$(date +%s).json"
+    ASYNC_S3_BUCKET=$(echo "${ASYNC_S3_OUTPUT_PATH}" | sed 's|s3://||' | cut -d'/' -f1)
+    S3_INPUT_LOCATION="s3://${ASYNC_S3_BUCKET}/${ASYNC_INPUT_KEY}"
+    
+    if ! aws s3 cp "${TEMP_PAYLOAD}" "${S3_INPUT_LOCATION}" --region "${AWS_REGION}" &> /dev/null; then
+        echo "❌ Failed to upload test payload to S3"
+        echo "   Location: ${S3_INPUT_LOCATION}"
+        echo "   Check that your IAM credentials have s3:PutObject permission"
+        rm -f "${TEMP_PAYLOAD}"
+        exit 1
+    fi
+    rm -f "${TEMP_PAYLOAD}"
+    echo "✅ Test payload uploaded to: ${S3_INPUT_LOCATION}"
+    
+    # Invoke endpoint asynchronously (no inference components for async)
+    echo "   Invoking async endpoint..."
+    
+    INVOKE_ARGS=(
+        --endpoint-name "${ENDPOINT_NAME}"
+        --input-location "${S3_INPUT_LOCATION}"
+        --region "${AWS_REGION}"
+        --content-type "application/json"
+    )
+    
+    if ! INVOKE_RESULT=$(aws sagemaker-runtime invoke-endpoint-async \
+        "${INVOKE_ARGS[@]}" 2>&1); then
+        echo "❌ Async invocation failed"
+        echo "   Error: ${INVOKE_RESULT}"
+        exit 1
+    fi
+    
+    # Extract output location from response
+    OUTPUT_LOCATION=$(echo "${INVOKE_RESULT}" | python3 -c "import sys,json; print(json.load(sys.stdin).get('OutputLocation',''))" 2>/dev/null || echo "")
+    
+    if [ -z "${OUTPUT_LOCATION}" ]; then
+        echo "⚠️  Async invocation accepted but no output location returned"
+        echo "   Check the S3 output path for results: ${ASYNC_S3_OUTPUT_PATH}"
+        echo ""
+        echo "✅ Async invocation submitted successfully"
+    else
+        echo "✅ Async invocation accepted"
+        echo "   Output location: ${OUTPUT_LOCATION}"
+        
+        # Poll S3 output location for result
+        POLL_TIMEOUT=300
+        POLL_INTERVAL=10
+        ELAPSED=0
+        
+        echo "⏳ Polling for async result (timeout: ${POLL_TIMEOUT}s)..."
+        
+        while [ ${ELAPSED} -lt ${POLL_TIMEOUT} ]; do
+            if aws s3 ls "${OUTPUT_LOCATION}" --region "${AWS_REGION}" &> /dev/null; then
+                echo "✅ Async inference result available"
+                
+                # Download and display result
+                TEMP_RESULT=$(mktemp)
+                if aws s3 cp "${OUTPUT_LOCATION}" "${TEMP_RESULT}" --region "${AWS_REGION}" &> /dev/null; then
+                    RESPONSE_BODY=$(cat "${TEMP_RESULT}")
+                    rm -f "${TEMP_RESULT}"
+                    
+                    echo "   Response preview: ${RESPONSE_BODY:0:200}"
+                    if [ ${#RESPONSE_BODY} -gt 200 ]; then
+                        echo "   (truncated, full response is ${#RESPONSE_BODY} characters)"
+                    fi
+                    
+                    # For diffusors, extract and save the generated image
+                    if [ "${FRAMEWORK}" = "diffusors" ]; then
+                        OUTPUT_IMAGE="${SCRIPT_DIR}/../test_output.png"
+                        if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json, base64
+resp = json.load(sys.stdin)
+if 'data' in resp and len(resp['data']) > 0 and 'b64_json' in resp['data'][0]:
+    img_data = base64.b64decode(resp['data'][0]['b64_json'])
+    with open('${OUTPUT_IMAGE}', 'wb') as f:
+        f.write(img_data)
+    print('ok')
+else:
+    print('skip')
+" 2>/dev/null | grep -q "ok"; then
+                            echo "   🖼️  Image saved to: ${OUTPUT_IMAGE}"
+                        fi
+                    fi
+                else
+                    rm -f "${TEMP_RESULT}"
+                    echo "⚠️  Result exists but could not be downloaded"
+                fi
+                break
+            fi
+            
+            sleep ${POLL_INTERVAL}
+            ELAPSED=$((ELAPSED + POLL_INTERVAL))
+            echo "   ⏳ Waiting... (${ELAPSED}s / ${POLL_TIMEOUT}s)"
+        done
+        
+        if [ ${ELAPSED} -ge ${POLL_TIMEOUT} ]; then
+            echo "❌ Async inference timed out after ${POLL_TIMEOUT}s"
+            echo ""
+            echo "   The request may still be processing. Check:"
+            echo "   • S3 output path: ${OUTPUT_LOCATION}"
+            echo "   • CloudWatch Logs: /aws/sagemaker/Endpoints/${ENDPOINT_NAME}"
+            echo "   • Endpoint status: aws sagemaker describe-endpoint --endpoint-name ${ENDPOINT_NAME} --region ${AWS_REGION}"
+            exit 1
+        fi
+    fi
+fi
+
+echo ""
+echo "✅ All tests passed!"
+echo ""
+
+if [ "${TEST_MODE}" = "local" ]; then
+    echo "Next steps:"
+    echo "  • Push to ECR: ./do/push"
+    echo "  • Deploy to SageMaker: ./do/deploy"
+else
+    echo "Async endpoint is ready for production use!"
+    echo "  • Endpoint name: ${ENDPOINT_NAME}"
+    echo "  • Region: ${AWS_REGION}"
+    echo "  • S3 output: ${ASYNC_S3_OUTPUT_PATH}"
+    echo ""
+    echo "📝 Register this deployment:"
+    echo "  ./do/register"
+fi
+
+<% } else if (deploymentTarget === 'hyperpod-eks') { %>
+# ============================================================
+# HyperPod EKS Testing
+# ============================================================
+
+# Parse arguments: local or hyperpod test mode
+# Default to hyperpod if no argument given (deployment target is hyperpod-eks)
+TEST_TARGET="${1:-hyperpod}"
+
+case "${TEST_TARGET}" in
+    local)
+        echo "🧪 Testing local container at localhost:8080"
+        echo "   Project: ${PROJECT_NAME}"
+        echo "   Framework: ${FRAMEWORK}"
+        echo "   Model server: ${MODEL_SERVER}"
+        TARGET_URL="http://localhost:8080"
+        ;;
+    hyperpod)
+        echo "🧪 Testing HyperPod EKS deployment"
+        echo "   Project: ${PROJECT_NAME}"
+        echo "   Framework: ${FRAMEWORK}"
+        echo "   Model server: ${MODEL_SERVER}"
+        echo "   Cluster: ${HYPERPOD_CLUSTER_NAME}"
+        echo "   Namespace: ${HYPERPOD_NAMESPACE}"
+        echo "   Region: ${AWS_REGION}"
+        echo ""
+        
+        # Get kubeconfig for HyperPod cluster
+        echo "🔑 Configuring kubectl for HyperPod cluster..."
+        KUBECONFIG_PATH="${HOME}/.kube/hyperpod-${HYPERPOD_CLUSTER_NAME}"
+        
+        EKS_CLUSTER_ARN=$(aws sagemaker describe-cluster \
+            --cluster-name "${HYPERPOD_CLUSTER_NAME}" \
+            --region "${AWS_REGION}" \
+            --query "Orchestrator.Eks.ClusterArn" \
+            --output text 2>&1) || {
+            echo "❌ Failed to describe HyperPod cluster: ${HYPERPOD_CLUSTER_NAME}"
+            echo ""
+            echo "   Check that:"
+            echo "   • The cluster name is correct"
+            echo "   • The cluster exists in region: ${AWS_REGION}"
+            echo "   • Your IAM user/role has permission to access the cluster"
+            exit 4
+        }
+        
+        EKS_CLUSTER_NAME=$(echo "${EKS_CLUSTER_ARN}" | awk -F'/' '{print $NF}')
+        
+        if ! aws eks update-kubeconfig \
+            --name "${EKS_CLUSTER_NAME}" \
+            --region "${AWS_REGION}" \
+            --kubeconfig "${KUBECONFIG_PATH}" 2>&1; then
+            echo "❌ Failed to configure kubectl for EKS cluster: ${EKS_CLUSTER_NAME}"
+            exit 4
+        fi
+        
+        export KUBECONFIG="${KUBECONFIG_PATH}"
+        
+        # Verify cluster connectivity
+        if ! kubectl cluster-info &> /dev/null; then
+            echo "❌ Cannot connect to HyperPod cluster"
+            exit 4
+        fi
+        echo "✅ Connected to HyperPod cluster"
+        
+        # Port-forward the service to a local port
+        LOCAL_PORT=8080
+        echo ""
+        echo "🔌 Port-forwarding svc/${PROJECT_NAME} to localhost:${LOCAL_PORT}..."
+        kubectl port-forward "svc/${PROJECT_NAME}" "${LOCAL_PORT}:8080" \
+            -n "${HYPERPOD_NAMESPACE}" &
+        PF_PID=$!
+        
+        # Wait for port-forward to establish
+        sleep 3
+        
+        # Ensure cleanup on exit
+        trap "kill ${PF_PID} 2>/dev/null || true" EXIT
+        
+        # Verify port-forward is running
+        if ! kill -0 ${PF_PID} 2>/dev/null; then
+            echo "❌ Port-forward failed to start"
+            echo ""
+            echo "   Check that:"
+            echo "   • The service exists: kubectl get svc ${PROJECT_NAME} -n ${HYPERPOD_NAMESPACE}"
+            echo "   • The deployment is running: kubectl get pods -n ${HYPERPOD_NAMESPACE}"
+            exit 1
+        fi
+        echo "✅ Port-forward established"
+        
+        TARGET_URL="http://localhost:${LOCAL_PORT}"
+        ;;
+    *)
+        echo "Usage: ./do/test [local|hyperpod]"
+        echo ""
+        echo "Test modes:"
+        echo "  local    - Test local container at localhost:8080"
+        echo "  hyperpod - Test HyperPod EKS deployment via port-forward"
+        exit 1
+        ;;
+esac
+
+echo ""
+
+# Test 1: Health check (/ping)
+echo "🔍 Test 1: Health check"
+echo "   Sending GET request to ${TARGET_URL}/ping"
+
+if ! PING_RESPONSE=$(curl -s -w "\n%{http_code}" -X GET "${TARGET_URL}/ping" 2>&1); then
+    echo "❌ Health check failed: Could not connect"
+    if [ "${TEST_TARGET}" = "local" ]; then
+        echo "   Make sure the container is running: ./do/run"
+    else
+        echo "   Check that the port-forward is working and pods are running"
+    fi
+    exit 1
+fi
+
+HTTP_CODE=$(echo "${PING_RESPONSE}" | tail -n1)
+RESPONSE_BODY=$(echo "${PING_RESPONSE}" | sed '$d')
+
+if [ "${HTTP_CODE}" = "200" ]; then
+    echo "✅ Health check passed (HTTP ${HTTP_CODE})"
+else
+    echo "❌ Health check failed (HTTP ${HTTP_CODE})"
+    echo "   Response: ${RESPONSE_BODY}"
+    exit 1
+fi
+
+echo ""
+
+# Test 2: Inference request (/invocations)
+echo "🔍 Test 2: Inference request"
+
+# Create framework-specific test payload
+case "${FRAMEWORK}" in
+    sklearn|xgboost)
+        # Traditional ML: JSON with instances array
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    tensorflow)
+        # TensorFlow: JSON with instances array
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    transformers)
+        # Transformers: payload format depends on model server
+        case "${MODEL_SERVER}" in
+            vllm|sglang)
+                # OpenAI-compatible chat completions format
+                VLLM_MODEL_NAME="${MODEL_NAME}"
+                if [[ "${MODEL_NAME}" == jumpstart://* ]] || [[ "${MODEL_NAME}" == jumpstart-hub://* ]] || [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                    VLLM_MODEL_NAME="/opt/ml/model"
+                fi
+                TEST_PAYLOAD='{"model": "'"${VLLM_MODEL_NAME}"'", "messages": [{"role": "user", "content": "What is machine learning?"}], "max_tokens": 50, "temperature": 0.7}'
+                echo "   Payload: OpenAI-compatible chat completion request"
+                echo "   Model: ${VLLM_MODEL_NAME}"
+                ;;
+            *)
+                # HuggingFace-style format for LMI, DJL, TensorRT-LLM
+                TEST_PAYLOAD='{"inputs": "What is machine learning?", "parameters": {"max_new_tokens": 50, "temperature": 0.7}}'
+                echo "   Payload: HuggingFace-style text generation request"
+                ;;
+        esac
+        ;;
+    diffusors)
+        # Diffusors: OpenAI DALL-E compatible image generation request
+        TEST_PAYLOAD='{"prompt": "A white cat", "n": 1, "size": "512x512"}'
+        echo "   Payload: OpenAI DALL-E compatible image generation request"
+        ;;
+    *)
+        echo "❌ Unknown framework: ${FRAMEWORK}"
+        exit 3
+        ;;
+esac
+
+if [ "${FRAMEWORK}" = "diffusors" ]; then
+    echo "   Sending POST request to ${TARGET_URL}/v1/images/generations"
+
+    if ! INVOKE_RESPONSE=$(curl -s -m 120 -w "\n%{http_code}" -X POST "${TARGET_URL}/v1/images/generations" \
+        -H "Content-Type: application/json" \
+        -d "${TEST_PAYLOAD}" 2>&1); then
+        echo "❌ Inference request failed: Could not connect"
+        exit 1
+    fi
+
+    HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+    RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+
+    if [ "${HTTP_CODE}" = "200" ]; then
+        # Validate response contains data array with b64_json
+        if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json
+resp = json.load(sys.stdin)
+assert 'data' in resp, 'Missing data array'
+assert len(resp['data']) > 0, 'Empty data array'
+assert 'b64_json' in resp['data'][0], 'Missing b64_json in data entry'
+" 2>/dev/null; then
+            echo "✅ Image generation successful (HTTP ${HTTP_CODE})"
+            echo "   Response contains valid data array with b64_json image"
+            
+            # Save generated image to file
+            OUTPUT_IMAGE="${SCRIPT_DIR}/../test_output.png"
+            if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json, base64
+resp = json.load(sys.stdin)
+img_data = base64.b64decode(resp['data'][0]['b64_json'])
+with open('${OUTPUT_IMAGE}', 'wb') as f:
+    f.write(img_data)
+" 2>/dev/null; then
+                echo "   🖼️  Image saved to: ${OUTPUT_IMAGE}"
+            fi
+        else
+            echo "⚠️  Image generation returned HTTP ${HTTP_CODE} but response format unexpected"
+            echo "   Response preview: ${RESPONSE_BODY:0:200}"
+        fi
+    else
+        echo "❌ Image generation failed (HTTP ${HTTP_CODE})"
+        echo "   Response: ${RESPONSE_BODY}"
+        exit 1
+    fi
+else
+    echo "   Sending POST request to ${TARGET_URL}/invocations"
+
+    if ! INVOKE_RESPONSE=$(curl -s -w "\n%{http_code}" -X POST "${TARGET_URL}/invocations" \
+        -H "Content-Type: application/json" \
+        -d "${TEST_PAYLOAD}" 2>&1); then
+        echo "❌ Inference request failed: Could not connect"
+        exit 1
+    fi
+
+    HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+    RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+
+    if [ "${HTTP_CODE}" = "200" ]; then
+        echo "✅ Inference request successful (HTTP ${HTTP_CODE})"
+        echo "   Response preview: ${RESPONSE_BODY:0:200}"
+        if [ ${#RESPONSE_BODY} -gt 200 ]; then
+            echo "   (truncated, full response is ${#RESPONSE_BODY} characters)"
+        fi
+    else
+        echo "❌ Inference request failed (HTTP ${HTTP_CODE})"
+        echo "   Response: ${RESPONSE_BODY}"
+        exit 1
+    fi
+fi
+
+echo ""
+echo "✅ All tests passed!"
+echo ""
+
+if [ "${TEST_TARGET}" = "local" ]; then
+    echo "Next steps:"
+    echo "  • Push to ECR: ./do/push"
+    echo "  • Deploy to HyperPod: ./do/deploy"
+else
+    echo "HyperPod deployment is ready for production use!"
+    echo "  • Cluster: ${HYPERPOD_CLUSTER_NAME}"
+    echo "  • Namespace: ${HYPERPOD_NAMESPACE}"
+    echo "  • Service: ${PROJECT_NAME}"
+    echo ""
+    echo "📝 Register this deployment:"
+    echo "  ./do/register"
+fi
+
+<% } else if (deploymentTarget === 'batch-transform') { %>
+# ============================================================
+# SageMaker Managed Inference - Batch Testing
+# ============================================================
+
+# Parse arguments: local or batch test mode
+# Default to batch if no argument given (deployment target is batch-transform)
+TEST_TARGET="${1:-batch}"
+
+case "${TEST_TARGET}" in
+    local)
+        echo "🧪 Testing local container at localhost:8080"
+        echo "   Project: ${PROJECT_NAME}"
+        echo "   Framework: ${FRAMEWORK}"
+        echo "   Model server: ${MODEL_SERVER}"
+        TARGET_URL="http://localhost:8080"
+        ;;
+    batch)
+        echo "🧪 Checking batch transform job status"
+        echo "   Project: ${PROJECT_NAME}"
+        echo "   Framework: ${FRAMEWORK}"
+        echo "   Region: ${AWS_REGION}"
+        echo "   S3 input: ${BATCH_INPUT_PATH}"
+        echo "   S3 output: ${BATCH_OUTPUT_PATH}"
+        echo ""
+
+        # Get transform job name from config
+        TRANSFORM_JOB_NAME="${TRANSFORM_JOB_NAME:-}"
+        if [ -z "${TRANSFORM_JOB_NAME}" ]; then
+            echo "❌ No transform job name found"
+            echo "   Run ./do/deploy first to create a transform job"
+            exit 1
+        fi
+
+        echo "🔍 Checking transform job: ${TRANSFORM_JOB_NAME}"
+
+        if ! JOB_STATUS_JSON=$(aws sagemaker describe-transform-job \
+            --transform-job-name "${TRANSFORM_JOB_NAME}" \
+            --region "${AWS_REGION}" 2>&1); then
+            echo "❌ Failed to describe transform job"
+            echo "   Error: ${JOB_STATUS_JSON}"
+            exit 1
+        fi
+
+        JOB_STATUS=$(echo "${JOB_STATUS_JSON}" | python3 -c "import sys,json; print(json.load(sys.stdin).get('TransformJobStatus','Unknown'))" 2>/dev/null || echo "Unknown")
+
+        case "${JOB_STATUS}" in
+            Completed)
+                echo "✅ Transform job completed successfully"
+                echo ""
+
+                # Download results locally
+                LOCAL_OUTPUT_DIR="${SCRIPT_DIR}/../batch-output"
+                mkdir -p "${LOCAL_OUTPUT_DIR}"
+                echo "📥 Downloading results to ${LOCAL_OUTPUT_DIR}/"
+                if aws s3 sync "${BATCH_OUTPUT_PATH}" "${LOCAL_OUTPUT_DIR}/" --region "${AWS_REGION}"; then
+                    DOWNLOADED=$(ls -1 "${LOCAL_OUTPUT_DIR}" 2>/dev/null | wc -l | tr -d ' ')
+                    echo "✅ Downloaded ${DOWNLOADED} file(s) to ${LOCAL_OUTPUT_DIR}/"
+                    echo ""
+
+                    # Display first output file preview
+                    FIRST_FILE=$(ls -1 "${LOCAL_OUTPUT_DIR}" 2>/dev/null | head -1)
+                    if [ -n "${FIRST_FILE}" ]; then
+                        echo "📄 Sample output (${FIRST_FILE}):"
+                        head -5 "${LOCAL_OUTPUT_DIR}/${FIRST_FILE}"
+                        LINES=$(wc -l < "${LOCAL_OUTPUT_DIR}/${FIRST_FILE}" | tr -d ' ')
+                        if [ "${LINES}" -gt 5 ]; then
+                            echo "   ... (${LINES} total lines)"
+                        fi
+                    fi
+                else
+                    echo "⚠️  Could not download output files"
+                fi
+
+                echo ""
+                echo "✅ All tests passed!"
+                echo ""
+                echo "📝 Register this deployment:"
+                echo "  ./do/register"
+                ;;
+            InProgress)
+                echo "⏳ Transform job is still in progress"
+
+                # Extract progress details
+                CREATION_TIME=$(echo "${JOB_STATUS_JSON}" | python3 -c "import sys,json; print(json.load(sys.stdin).get('CreationTime','Unknown'))" 2>/dev/null || echo "Unknown")
+                echo "   Started: ${CREATION_TIME}"
+                echo "   Status: InProgress"
+                echo ""
+                echo "   The job is still running. Check again later:"
+                echo "   ./do/test"
+                echo ""
+                echo "   View logs:"
+                echo "   ./do/logs"
+                ;;
+            Failed)
+                echo "❌ Transform job failed"
+
+                FAILURE_REASON=$(echo "${JOB_STATUS_JSON}" | python3 -c "import sys,json; print(json.load(sys.stdin).get('FailureReason','Unknown'))" 2>/dev/null || echo "Unknown")
+                echo "   Reason: ${FAILURE_REASON}"
+                echo ""
+                echo "   View logs for more details:"
+                echo "   ./do/logs"
+                exit 1
+                ;;
+            Stopped)
+                echo "⚠️  Transform job was stopped"
+                echo "   The job was manually stopped before completion"
+                echo ""
+                echo "   To start a new job, run:"
+                echo "   ./do/deploy"
+                ;;
+            *)
+                echo "⚠️  Transform job status: ${JOB_STATUS}"
+                echo "   Check again later: ./do/test"
+                ;;
+        esac
+        exit 0
+        ;;
+    *)
+        echo "Usage: ./do/test [local|batch]"
+        echo ""
+        echo "Test modes:"
+        echo "  local - Test local container at localhost:8080"
+        echo "  batch - Check transform job status and view results"
+        exit 1
+        ;;
+esac
+
+echo ""
+
+# Test 1: Health check (/ping)
+echo "🔍 Test 1: Health check"
+echo "   Sending GET request to ${TARGET_URL}/ping"
+
+if ! PING_RESPONSE=$(curl -s -w "\n%{http_code}" -X GET "${TARGET_URL}/ping" 2>&1); then
+    echo "❌ Health check failed: Could not connect to local container"
+    echo "   Make sure the container is running: ./do/run"
+    exit 1
+fi
+
+HTTP_CODE=$(echo "${PING_RESPONSE}" | tail -n1)
+RESPONSE_BODY=$(echo "${PING_RESPONSE}" | sed '$d')
+
+if [ "${HTTP_CODE}" = "200" ]; then
+    echo "✅ Health check passed (HTTP ${HTTP_CODE})"
+else
+    echo "❌ Health check failed (HTTP ${HTTP_CODE})"
+    echo "   Response: ${RESPONSE_BODY}"
+    exit 1
+fi
+
+echo ""
+
+# Test 2: Inference request (/invocations)
+echo "🔍 Test 2: Inference request"
+
+# Create framework-specific test payload
+case "${FRAMEWORK}" in
+    sklearn|xgboost)
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    tensorflow)
+        TEST_PAYLOAD='{"instances": [[1.0, 2.0, 3.0, 4.0]]}'
+        echo "   Payload: Sample feature vector"
+        ;;
+    transformers)
+        case "${MODEL_SERVER}" in
+            vllm|sglang)
+                VLLM_MODEL_NAME="${MODEL_NAME}"
+                if [[ "${MODEL_NAME}" == jumpstart://* ]] || [[ "${MODEL_NAME}" == jumpstart-hub://* ]] || [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                    VLLM_MODEL_NAME="/opt/ml/model"
+                fi
+                TEST_PAYLOAD='{"model": "'"${VLLM_MODEL_NAME}"'", "messages": [{"role": "user", "content": "What is machine learning?"}], "max_tokens": 50, "temperature": 0.7}'
+                echo "   Payload: OpenAI-compatible chat completion request"
+                echo "   Model: ${VLLM_MODEL_NAME}"
+                ;;
+            *)
+                TEST_PAYLOAD='{"inputs": "What is machine learning?", "parameters": {"max_new_tokens": 50, "temperature": 0.7}}'
+                echo "   Payload: HuggingFace-style text generation request"
+                ;;
+        esac
+        ;;
+    diffusors)
+        TEST_PAYLOAD='{"prompt": "A white cat", "n": 1, "size": "512x512"}'
+        echo "   Payload: OpenAI DALL-E compatible image generation request"
+        ;;
+    *)
+        echo "❌ Unknown framework: ${FRAMEWORK}"
+        exit 3
+        ;;
+esac
+
+if [ "${FRAMEWORK}" = "diffusors" ]; then
+    echo "   Sending POST request to ${TARGET_URL}/v1/images/generations"
+
+    if ! INVOKE_RESPONSE=$(curl -s -m 120 -w "\n%{http_code}" -X POST "${TARGET_URL}/v1/images/generations" \
+        -H "Content-Type: application/json" \
+        -d "${TEST_PAYLOAD}" 2>&1); then
+        echo "❌ Inference request failed: Could not connect to local container"
+        exit 1
+    fi
+
+    HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+    RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+
+    if [ "${HTTP_CODE}" = "200" ]; then
+        if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json
+resp = json.load(sys.stdin)
+assert 'data' in resp, 'Missing data array'
+assert len(resp['data']) > 0, 'Empty data array'
+assert 'b64_json' in resp['data'][0], 'Missing b64_json in data entry'
+" 2>/dev/null; then
+            echo "✅ Image generation successful (HTTP ${HTTP_CODE})"
+            echo "   Response contains valid data array with b64_json image"
+
+            # Save generated image to file
+            OUTPUT_IMAGE="${SCRIPT_DIR}/../test_output.png"
+            if echo "${RESPONSE_BODY}" | python3 -c "
+import sys, json, base64
+resp = json.load(sys.stdin)
+img_data = base64.b64decode(resp['data'][0]['b64_json'])
+with open('${OUTPUT_IMAGE}', 'wb') as f:
+    f.write(img_data)
+" 2>/dev/null; then
+                echo "   🖼️  Image saved to: ${OUTPUT_IMAGE}"
+            fi
+        else
+            echo "⚠️  Image generation returned HTTP ${HTTP_CODE} but response format unexpected"
+            echo "   Response preview: ${RESPONSE_BODY:0:200}"
+        fi
+    else
+        echo "❌ Image generation failed (HTTP ${HTTP_CODE})"
+        echo "   Response: ${RESPONSE_BODY}"
+        exit 1
+    fi
+else
+    echo "   Sending POST request to ${TARGET_URL}/invocations"
+
+    if ! INVOKE_RESPONSE=$(curl -s -w "\n%{http_code}" -X POST "${TARGET_URL}/invocations" \
+        -H "Content-Type: application/json" \
+        -d "${TEST_PAYLOAD}" 2>&1); then
+        echo "❌ Inference request failed: Could not connect to local container"
+        exit 1
+    fi
+
+    HTTP_CODE=$(echo "${INVOKE_RESPONSE}" | tail -n1)
+    RESPONSE_BODY=$(echo "${INVOKE_RESPONSE}" | sed '$d')
+
+    if [ "${HTTP_CODE}" = "200" ]; then
+        echo "✅ Inference request successful (HTTP ${HTTP_CODE})"
+        echo "   Response preview: ${RESPONSE_BODY:0:200}"
+        if [ ${#RESPONSE_BODY} -gt 200 ]; then
+            echo "   (truncated, full response is ${#RESPONSE_BODY} characters)"
+        fi
+    else
+        echo "❌ Inference request failed (HTTP ${HTTP_CODE})"
+        echo "   Response: ${RESPONSE_BODY}"
+        exit 1
+    fi
+fi
+
+echo ""
+echo "✅ All tests passed!"
+echo ""
+
+echo "Next steps:"
+echo "  • Push to ECR: ./do/push"
+echo "  • Deploy batch transform: ./do/deploy"
+
+<% } %>