@aws/ml-container-creator 0.3.0 → 0.5.0

package/templates/do/register

@@ -89,6 +89,103 @@ case "${STATUS}" in
         ;;
 esac
 
+# ============================================================
+# Build IC list from do/ic/ directory (multi-IC support)
+# ============================================================
+
+<% if (deploymentTarget === 'realtime-inference') { %>
+IC_LIST_JSON="[]"
+if [ -d "${SCRIPT_DIR}/ic" ]; then
+    # Build IC list from all conf files (alphabetical order)
+    IC_ENTRIES=""
+    IC_COUNT=0
+    for conf in "${SCRIPT_DIR}"/ic/*.conf; do
+        [ -f "${conf}" ] || continue
+
+        # Source the IC config to get its variables
+        (
+            # Subshell to avoid polluting current environment
+            source "${conf}" 2>/dev/null
+            echo "${IC_DEPLOYED_NAME:-}|${IC_IMAGE_TAG:-}|${IC_GPU_COUNT:-1}|${IC_COPY_COUNT:-1}"
+        ) > /tmp/ic_entry_$$ 2>/dev/null
+
+        IC_ENTRY=$(cat /tmp/ic_entry_$$ 2>/dev/null || echo "|||")
+        rm -f /tmp/ic_entry_$$
+
+        IC_BASENAME=$(basename "${conf}" .conf)
+        IC_ENTRY_IMAGE=$(echo "${IC_ENTRY}" | cut -d'|' -f2)
+        IC_ENTRY_GPU=$(echo "${IC_ENTRY}" | cut -d'|' -f3)
+        IC_ENTRY_COPY=$(echo "${IC_ENTRY}" | cut -d'|' -f4)
+
+        if [ -n "${IC_ENTRIES}" ]; then
+            IC_ENTRIES="${IC_ENTRIES},"
+        fi
+        IC_ENTRIES="${IC_ENTRIES}{\"name\":\"${IC_BASENAME}\",\"image\":\"${IC_ENTRY_IMAGE}\",\"gpuCount\":${IC_ENTRY_GPU:-1},\"copyCount\":${IC_ENTRY_COPY:-1}}"
+        IC_COUNT=$((IC_COUNT + 1))
+    done
+
+    if [ "${CI_MODE}" = true ] && [ ${IC_COUNT} -gt 1 ]; then
+        # CI mode: only include the first IC (alphabetically) to keep CI costs down
+        FIRST_CONF=$(ls "${SCRIPT_DIR}"/ic/*.conf 2>/dev/null | head -1)
+        if [ -n "${FIRST_CONF}" ]; then
+            (
+                source "${FIRST_CONF}" 2>/dev/null
+                echo "${IC_DEPLOYED_NAME:-}|${IC_IMAGE_TAG:-}|${IC_GPU_COUNT:-1}|${IC_COPY_COUNT:-1}"
+            ) > /tmp/ic_first_$$ 2>/dev/null
+
+            FIRST_ENTRY=$(cat /tmp/ic_first_$$ 2>/dev/null || echo "|||")
+            rm -f /tmp/ic_first_$$
+
+            FIRST_BASENAME=$(basename "${FIRST_CONF}" .conf)
+            FIRST_IMAGE=$(echo "${FIRST_ENTRY}" | cut -d'|' -f2)
+            FIRST_GPU=$(echo "${FIRST_ENTRY}" | cut -d'|' -f3)
+            FIRST_COPY=$(echo "${FIRST_ENTRY}" | cut -d'|' -f4)
+
+            IC_LIST_JSON="[{\"name\":\"${FIRST_BASENAME}\",\"image\":\"${FIRST_IMAGE}\",\"gpuCount\":${FIRST_GPU:-1},\"copyCount\":${FIRST_COPY:-1}}]"
+        fi
+    else
+        IC_LIST_JSON="[${IC_ENTRIES}]"
+    fi
+else
+    # Legacy: single IC from do/config
+    IC_LIST_JSON="[{\"name\":\"default\",\"image\":\"${PROJECT_NAME}-latest\",\"gpuCount\":${IC_GPU_COUNT:-1},\"copyCount\":${IC_COPY_COUNT:-1}}]"
+fi
+
+# Append adapter entries from do/adapters/*.conf
+ADAPTER_COUNT=0
+if [ -d "${SCRIPT_DIR}/adapters" ]; then
+    ADAPTER_ENTRIES=""
+    for conf in "${SCRIPT_DIR}"/adapters/*.conf; do
+        [ -f "${conf}" ] || continue
+        [[ "$(basename "${conf}")" == ".gitkeep" ]] && continue
+
+        ADAPTER_NAME_VAL=""
+        ADAPTER_WEIGHTS_VAL=""
+        ADAPTER_IC_VAL=""
+        eval "$(grep '^export ADAPTER_NAME=' "${conf}" 2>/dev/null)"
+        eval "$(grep '^export ADAPTER_WEIGHTS_URI=' "${conf}" 2>/dev/null)"
+        eval "$(grep '^export ADAPTER_IC_NAME=' "${conf}" 2>/dev/null)"
+        ADAPTER_NAME_VAL="${ADAPTER_NAME:-$(basename "${conf}" .conf)}"
+        ADAPTER_WEIGHTS_VAL="${ADAPTER_WEIGHTS_URI:-}"
+        ADAPTER_IC_VAL="${ADAPTER_IC_NAME:-}"
+
+        if [ -n "${ADAPTER_ENTRIES}" ]; then
+            ADAPTER_ENTRIES="${ADAPTER_ENTRIES},"
+        fi
+        ADAPTER_ENTRIES="${ADAPTER_ENTRIES}{\"name\":\"${ADAPTER_NAME_VAL}\",\"isAdapter\":true,\"baseIcName\":\"${ADAPTER_IC_VAL}\",\"artifactUrl\":\"${ADAPTER_WEIGHTS_VAL}\",\"gpuCount\":0,\"copyCount\":1}"
+        ADAPTER_COUNT=$((ADAPTER_COUNT + 1))
+        unset ADAPTER_NAME ADAPTER_WEIGHTS_URI ADAPTER_IC_NAME
+    done
+
+    if [ -n "${ADAPTER_ENTRIES}" ] && [ "${IC_LIST_JSON}" != "[]" ]; then
+        # Append adapters to existing IC list
+        IC_LIST_JSON="${IC_LIST_JSON%]},${ADAPTER_ENTRIES}]"
+    elif [ -n "${ADAPTER_ENTRIES}" ]; then
+        IC_LIST_JSON="[${ADAPTER_ENTRIES}]"
+    fi
+fi
+<% } %>
+
 # ============================================================
 # Derive architecture and backend from DEPLOYMENT_CONFIG
 # ============================================================
@@ -293,7 +390,7 @@ echo ""
 # ============================================================
 
 compute_config_id() {
-    local input="${DEPLOYMENT_CONFIG}:${MODEL_NAME:-none}:${INSTANCE_TYPE}:${AWS_REGION}:${DEPLOYMENT_TARGET}"
+    local input="${DEPLOYMENT_CONFIG}:${MODEL_NAME:-none}:${INSTANCE_TYPE}:${AWS_REGION}:${DEPLOYMENT_TARGET}:ic${IC_COUNT:-1}:adapt${ADAPTER_COUNT:-0}"
     # Use sha256sum (Linux) with fallback to shasum (macOS)
     if command -v sha256sum &> /dev/null; then
         echo -n "$input" | sha256sum | cut -c1-16
@@ -373,6 +470,9 @@ write_ci_record() {
     "modelWeight": ${IC_MODEL_WEIGHT}
 <% } %>
   },
+<% } %>
+<% if (deploymentTarget === 'realtime-inference') { %>
+  "icList": ${IC_LIST_JSON},
 <% } %>
   "parameters": ${PARAMETERS}
 }
@@ -393,6 +493,7 @@ CJEOF
 
     # Try put-item with condition (new record)
     if aws dynamodb put-item \
+        --region "${AWS_REGION}" \
         --table-name "${CI_TABLE_NAME}" \
         --item "{
             \"configId\": {\"S\": \"${config_id}\"},
@@ -412,6 +513,7 @@ CJEOF
     else
         # Record already exists — update it (reset testStatus, update configJson, preserve createdAt)
         if aws dynamodb update-item \
+            --region "${AWS_REGION}" \
             --table-name "${CI_TABLE_NAME}" \
             --key "{\"configId\": {\"S\": \"${config_id}\"}}" \
             --update-expression "SET configJson = :cj, testStatus = :ts, deploymentConfig = :dc, baseImage = :bi, baseImageVersion = :bv, buildStrategy = :bs, projectName = :pn, schemaVersion = :sv" \
@@ -486,6 +588,9 @@ if [ "${JSON_OUTPUT}" = true ] || [ "${CI_MODE}" = true ]; then
     "modelWeight": ${IC_MODEL_WEIGHT}
 <% } %>
   },
+<% } %>
+<% if (deploymentTarget === 'realtime-inference') { %>
+  "icList": ${IC_LIST_JSON},
 <% } %>
   "parameters": ${PARAMETERS}
 }
@@ -496,6 +601,11 @@ DJEOF
     echo "${DEPLOYMENT_JSON}" | python3 -c "import sys,json; print(json.dumps(json.load(sys.stdin), indent=2))" 2>/dev/null || echo "${DEPLOYMENT_JSON}"
 
     if [ "${CI_MODE}" = true ]; then
+        # Strip capacity reservation ARN for CI — force on-demand deployment
+        # CI projects must never use reserved capacity (reservations are account-specific
+        # and time-bound; CI replay should always target on-demand instances)
+        unset CAPACITY_RESERVATION_ARN 2>/dev/null || true
+
         echo ""
         echo "⚠️  CI Integration is experimental and currently only tested for"
         echo "   SageMaker Real-Time Inference endpoints."
@@ -507,7 +617,7 @@ DJEOF
         echo "🔑 configId: ${CONFIG_ID}"
 
         # Check if CI_Table exists before writing
-        if ! aws dynamodb describe-table --table-name "${CI_TABLE_NAME}" &>/dev/null; then
+        if ! aws dynamodb describe-table --table-name "${CI_TABLE_NAME}" --region "${AWS_REGION}" &>/dev/null; then
             echo ""
             echo "⚠️  CI infrastructure not provisioned. Run 'ml-container-creator bootstrap' with CI enabled."
             echo "   Skipping CI table write."
@@ -567,6 +677,13 @@ fi
 # Pass parameters as JSON string
 CMD_ARGS+=("--parameters" "${PARAMETERS}")
 
+# Pass IC list as JSON string
+<% if (deploymentTarget === 'realtime-inference') { %>
+if [ "${IC_LIST_JSON}" != "[]" ]; then
+    CMD_ARGS+=("--ic-list" "${IC_LIST_JSON}")
+fi
+<% } %>
+
 # Pass generator version from package.json if available
 GENERATOR_VERSION=""
 if command -v node &> /dev/null; then

package/templates/do/status

@@ -0,0 +1,337 @@
+#!/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"
+
+# ============================================================
+# SageMaker Real-Time Inference Status
+# ============================================================
+
+# Validate AWS credentials
+if ! aws sts get-caller-identity &> /dev/null; then
+    echo "❌ AWS credentials not configured"
+    echo "   Run: aws configure"
+    exit 4
+fi
+
+# Check that we have an endpoint to query
+if [ -z "${ENDPOINT_NAME:-}" ]; then
+    echo "❌ No endpoint configured"
+    echo "   Run ./do/deploy first to create an endpoint."
+    exit 1
+fi
+
+# ============================================================
+# Describe Endpoint
+# ============================================================
+ENDPOINT_JSON=$(aws sagemaker describe-endpoint \
+    --endpoint-name "${ENDPOINT_NAME}" \
+    --region "${AWS_REGION}" 2>/dev/null) || {
+    echo "❌ Endpoint not found: ${ENDPOINT_NAME}"
+    echo "   The endpoint may have been deleted. Run ./do/deploy to create a new one."
+    exit 1
+}
+
+EP_STATUS=$(echo "${ENDPOINT_JSON}" | grep -o '"EndpointStatus":"[^"]*"' | head -1 | cut -d'"' -f4)
+EP_INSTANCE_TYPE=$(echo "${ENDPOINT_JSON}" | grep -o '"InstanceType":"[^"]*"' | head -1 | cut -d'"' -f4)
+EP_INSTANCE_COUNT=$(echo "${ENDPOINT_JSON}" | grep -o '"CurrentInstanceCount":[0-9]*' | head -1 | cut -d':' -f2)
+
+# Fallback for instance count if not available
+if [ -z "${EP_INSTANCE_COUNT}" ]; then
+    EP_INSTANCE_COUNT=$(echo "${ENDPOINT_JSON}" | grep -o '"InitialInstanceCount":[0-9]*' | head -1 | cut -d':' -f2)
+fi
+EP_INSTANCE_COUNT="${EP_INSTANCE_COUNT:-1}"
+
+# Use INSTANCE_TYPE from config as fallback if not in describe response
+EP_INSTANCE_TYPE="${EP_INSTANCE_TYPE:-${INSTANCE_TYPE:-unknown}}"
+
+# GPU count lookup for the instance type
+_get_instance_gpus() {
+    local itype="$1"
+    case "${itype}" in
+        ml.g4dn.xlarge)     echo 1 ;;
+        ml.g4dn.12xlarge)   echo 4 ;;
+        ml.g5.xlarge)       echo 1 ;;
+        ml.g5.2xlarge)      echo 1 ;;
+        ml.g5.4xlarge)      echo 1 ;;
+        ml.g5.8xlarge)      echo 1 ;;
+        ml.g5.12xlarge)     echo 4 ;;
+        ml.g5.48xlarge)     echo 8 ;;
+        ml.g6.xlarge)       echo 1 ;;
+        ml.g6.12xlarge)     echo 4 ;;
+        ml.g6.48xlarge)     echo 8 ;;
+        ml.g6e.xlarge)      echo 1 ;;
+        ml.g6e.2xlarge)     echo 1 ;;
+        ml.g6e.4xlarge)     echo 1 ;;
+        ml.g6e.8xlarge)     echo 1 ;;
+        ml.g6e.12xlarge)    echo 4 ;;
+        ml.g6e.48xlarge)    echo 8 ;;
+        ml.g7e.xlarge)      echo 1 ;;
+        ml.g7e.2xlarge)     echo 1 ;;
+        ml.g7e.4xlarge)     echo 1 ;;
+        ml.g7e.8xlarge)     echo 1 ;;
+        ml.g7e.12xlarge)    echo 4 ;;
+        ml.g7e.48xlarge)    echo 8 ;;
+        ml.p3.2xlarge)      echo 1 ;;
+        ml.p3.8xlarge)      echo 4 ;;
+        ml.p3.16xlarge)     echo 8 ;;
+        ml.p4d.24xlarge)    echo 8 ;;
+        ml.p4de.24xlarge)   echo 8 ;;
+        ml.p5.48xlarge)     echo 8 ;;
+        *)                  echo "" ;;
+    esac
+}
+
+INSTANCE_GPUS=$(_get_instance_gpus "${EP_INSTANCE_TYPE}")
+TOTAL_GPUS=""
+if [ -n "${INSTANCE_GPUS}" ]; then
+    TOTAL_GPUS=$(( INSTANCE_GPUS * EP_INSTANCE_COUNT ))
+fi
+
+# ============================================================
+# Detect Instance Pools
+# ============================================================
+HAS_INSTANCE_POOLS=false
+if echo "${ENDPOINT_JSON}" | grep -q '"InstancePools"'; then
+    HAS_INSTANCE_POOLS=true
+fi
+
+# ============================================================
+# Print Endpoint Status
+# ============================================================
+echo ""
+if [ "${ENDPOINT_EXTERNAL:-false}" = "true" ]; then
+    echo "Endpoint: ${ENDPOINT_NAME} (external) [${EP_STATUS}]"
+else
+    echo "Endpoint: ${ENDPOINT_NAME} [${EP_STATUS}]"
+fi
+
+if [ "${HAS_INSTANCE_POOLS}" = "true" ]; then
+    # Instance pools path: show per-pool information
+    echo "Instance Pools:"
+
+    # Extract pool entries from the DescribeEndpoint response
+    # Each pool has: InstanceType, Priority, and CurrentInstanceCount (from the running endpoint)
+    # Parse using grep/sed — pools are in ProductionVariants[0].InstancePools array
+    pool_entries=$(echo "${ENDPOINT_JSON}" | grep -oE '"InstancePools"\s*:\s*\[[^]]*\]' | head -1 | sed 's/"InstancePools"\s*:\s*//')
+
+    if [ -n "${pool_entries}" ]; then
+        # Extract individual pool objects
+        # Each pool: {"InstanceType":"ml.xxx","Priority":N} — may also have CurrentInstanceCount
+        pool_types=$(echo "${pool_entries}" | grep -oE '"InstanceType"\s*:\s*"[^"]+"' | sed 's/"InstanceType"\s*:\s*"//;s/"$//')
+        pool_priorities=$(echo "${pool_entries}" | grep -oE '"Priority"\s*:\s*[0-9]+' | sed 's/"Priority"\s*:\s*//')
+
+        # CurrentInstanceCount may appear per-pool in the response
+        # If not per-pool, fall back to the endpoint-level CurrentInstanceCount
+        pool_instance_counts=$(echo "${pool_entries}" | grep -oE '"CurrentInstanceCount"\s*:\s*[0-9]+' | sed 's/"CurrentInstanceCount"\s*:\s*//')
+
+        # Convert to arrays
+        IFS=$'\n' read -r -d '' -a types_arr <<< "${pool_types}" || true
+        IFS=$'\n' read -r -d '' -a priorities_arr <<< "${pool_priorities}" || true
+        IFS=$'\n' read -r -d '' -a counts_arr <<< "${pool_instance_counts}" || true
+
+        for i in "${!types_arr[@]}"; do
+            local_type="${types_arr[$i]}"
+            local_priority="${priorities_arr[$i]:-$((i+1))}"
+            local_count="${counts_arr[$i]:-0}"
+
+            # Mark pools with instances > 0 as active
+            if [ "${local_count}" -gt 0 ] 2>/dev/null; then
+                printf "  Priority %s: %-20s (%s instances) ← active\n" "${local_priority}" "${local_type}" "${local_count}"
+            else
+                printf "  Priority %s: %-20s (%s instances)\n" "${local_priority}" "${local_type}" "${local_count}"
+            fi
+        done
+    fi
+else
+    # Standard single instance type path
+    if [ -n "${TOTAL_GPUS}" ]; then
+        echo "Instance: ${EP_INSTANCE_TYPE} (${EP_INSTANCE_COUNT} instance, ${TOTAL_GPUS} GPUs)"
+    else
+        echo "Instance: ${EP_INSTANCE_TYPE} (${EP_INSTANCE_COUNT} instance)"
+    fi
+fi
+echo ""
+
+# ============================================================
+# Describe Inference Components
+# ============================================================
+TOTAL_GPU_USED=0
+IC_ROWS=""
+
+if [ -d "${SCRIPT_DIR}/ic" ]; then
+    # Multi-IC path: iterate do/ic/*.conf files
+    # NOTE: Only base ICs in do/ic/ are counted toward GPU usage.
+    # Adapter ICs (do/adapters/*.conf) share the base IC's GPU resources
+    # and do not have their own ComputeResourceRequirements, so they are
+    # intentionally excluded from GPU capacity calculations.
+    HAS_ICS=false
+
+    for conf in "${SCRIPT_DIR}"/ic/*.conf; do
+        [ -f "${conf}" ] || continue
+        HAS_ICS=true
+
+        ic_basename=$(basename "${conf}" .conf)
+
+        # Read IC_DEPLOYED_NAME from the conf file
+        ic_deployed_name=""
+        if grep -q "^export IC_DEPLOYED_NAME=" "${conf}" 2>/dev/null; then
+            ic_deployed_name=$(grep "^export IC_DEPLOYED_NAME=" "${conf}" | sed 's/^export IC_DEPLOYED_NAME="//' | sed 's/"$//')
+        fi
+
+        if [ -z "${ic_deployed_name}" ]; then
+            IC_ROWS="${IC_ROWS}$(printf "%-18s %-12s %-6s %-6s" "${ic_basename}" "Not Deployed" "-" "-")\n"
+            continue
+        fi
+
+        # Call DescribeInferenceComponent
+        IC_JSON=$(aws sagemaker describe-inference-component \
+            --inference-component-name "${ic_deployed_name}" \
+            --region "${AWS_REGION}" 2>/dev/null) || {
+            IC_ROWS="${IC_ROWS}$(printf "%-18s %-12s %-6s %-6s" "${ic_basename}" "Not Found" "-" "-")\n"
+            continue
+        }
+
+        ic_status=$(echo "${IC_JSON}" | grep -o '"InferenceComponentStatus":"[^"]*"' | head -1 | cut -d'"' -f4)
+        ic_status="${ic_status:-Unknown}"
+
+        ic_gpu_count=$(echo "${IC_JSON}" | grep -o '"NumberOfAcceleratorDevicesRequired":[0-9]*' | head -1 | cut -d':' -f2)
+        ic_gpu_count="${ic_gpu_count:-0}"
+
+        ic_copy_count=$(echo "${IC_JSON}" | grep -o '"DesiredCopyCount":[0-9]*' | head -1 | cut -d':' -f2)
+        if [ -z "${ic_copy_count}" ]; then
+            ic_copy_count=$(echo "${IC_JSON}" | grep -o '"CurrentCopyCount":[0-9]*' | head -1 | cut -d':' -f2)
+        fi
+        ic_copy_count="${ic_copy_count:-1}"
+
+        TOTAL_GPU_USED=$(( TOTAL_GPU_USED + ic_gpu_count ))
+
+        IC_ROWS="${IC_ROWS}$(printf "%-18s %-12s %-6s %-6s" "${ic_basename}" "${ic_status}" "${ic_gpu_count}" "${ic_copy_count}")\n"
+    done
+
+    if [ "${HAS_ICS}" = true ]; then
+        echo "Inference Components:"
+        printf "%-18s %-12s %-6s %-6s\n" "NAME" "STATUS" "GPUs" "COPIES"
+        echo -e "${IC_ROWS}" | head -n -1
+        echo "                                    ─────"
+        if [ -n "${TOTAL_GPUS}" ]; then
+            printf "Total GPU usage:                    %s/%s\n" "${TOTAL_GPU_USED}" "${TOTAL_GPUS}"
+        else
+            printf "Total GPU usage:                    %s\n" "${TOTAL_GPU_USED}"
+        fi
+    else
+        echo "No IC config files found in do/ic/"
+    fi
+else
+    # Legacy single-IC path: use INFERENCE_COMPONENT_NAME from config
+    ic_name="${INFERENCE_COMPONENT_NAME:-}"
+
+    if [ -z "${ic_name}" ]; then
+        echo "No inference component deployed."
+        echo "Run ./do/deploy to create one."
+    else
+        IC_JSON=$(aws sagemaker describe-inference-component \
+            --inference-component-name "${ic_name}" \
+            --region "${AWS_REGION}" 2>/dev/null) || {
+            echo "Inference Components:"
+            printf "%-18s %-12s %-6s %-6s\n" "NAME" "STATUS" "GPUs" "COPIES"
+            printf "%-18s %-12s %-6s %-6s\n" "default" "Not Found" "-" "-"
+            echo ""
+            exit 0
+        }
+
+        ic_status=$(echo "${IC_JSON}" | grep -o '"InferenceComponentStatus":"[^"]*"' | head -1 | cut -d'"' -f4)
+        ic_status="${ic_status:-Unknown}"
+
+        ic_gpu_count=$(echo "${IC_JSON}" | grep -o '"NumberOfAcceleratorDevicesRequired":[0-9]*' | head -1 | cut -d':' -f2)
+        ic_gpu_count="${ic_gpu_count:-0}"
+
+        ic_copy_count=$(echo "${IC_JSON}" | grep -o '"DesiredCopyCount":[0-9]*' | head -1 | cut -d':' -f2)
+        if [ -z "${ic_copy_count}" ]; then
+            ic_copy_count=$(echo "${IC_JSON}" | grep -o '"CurrentCopyCount":[0-9]*' | head -1 | cut -d':' -f2)
+        fi
+        ic_copy_count="${ic_copy_count:-1}"
+
+        TOTAL_GPU_USED=$(( TOTAL_GPU_USED + ic_gpu_count ))
+
+        echo "Inference Components:"
+        printf "%-18s %-12s %-6s %-6s\n" "NAME" "STATUS" "GPUs" "COPIES"
+        printf "%-18s %-12s %-6s %-6s\n" "default" "${ic_status}" "${ic_gpu_count}" "${ic_copy_count}"
+        echo "                                    ─────"
+        if [ -n "${TOTAL_GPUS}" ]; then
+            printf "Total GPU usage:                    %s/%s\n" "${TOTAL_GPU_USED}" "${TOTAL_GPUS}"
+        else
+            printf "Total GPU usage:                    %s\n" "${TOTAL_GPU_USED}"
+        fi
+    fi
+fi
+
+echo ""
+
+<% if (typeof enableLora !== 'undefined' && enableLora) { %>
+# ============================================================
+# Describe LoRA Adapters
+# ============================================================
+if [ "${ENABLE_LORA:-}" = "true" ]; then
+    # List all inference components on the endpoint
+    ADAPTER_IC_LIST=$(aws sagemaker list-inference-components \
+        --endpoint-name-equals "${ENDPOINT_NAME}" \
+        --region "${AWS_REGION}" 2>/dev/null) || ADAPTER_IC_LIST=""
+
+    if [ -n "${ADAPTER_IC_LIST}" ]; then
+        # Extract IC names
+        ADAPTER_IC_NAMES=$(echo "${ADAPTER_IC_LIST}" | jq -r '.InferenceComponents[].InferenceComponentName' 2>/dev/null)
+
+        # Filter to adapter ICs (those with BaseInferenceComponentName) and collect details
+        ADAPTER_ROWS=""
+        ADAPTER_COUNT=0
+
+        for ic_name in ${ADAPTER_IC_NAMES}; do
+            # Describe each IC to check if it's an adapter
+            ic_detail=$(aws sagemaker describe-inference-component \
+                --inference-component-name "${ic_name}" \
+                --region "${AWS_REGION}" 2>/dev/null) || continue
+
+            # Check if this IC has a BaseInferenceComponentName (adapter IC)
+            base_ic=$(echo "${ic_detail}" | jq -r '.Specification.BaseInferenceComponentName // empty' 2>/dev/null)
+
+            if [ -z "${base_ic}" ]; then
+                # Not an adapter IC — skip
+                continue
+            fi
+
+            # Extract status and artifact URL
+            adapter_status=$(echo "${ic_detail}" | jq -r '.InferenceComponentStatus // "Unknown"' 2>/dev/null)
+            adapter_weights=$(echo "${ic_detail}" | jq -r '.Specification.Container.ArtifactUrl // "N/A"' 2>/dev/null)
+
+            # Derive display name (strip project prefix if present)
+            display_name="${ic_name}"
+            if [[ "${ic_name}" == "${PROJECT_NAME}-adapter-"* ]]; then
+                display_name="${ic_name#${PROJECT_NAME}-adapter-}"
+            fi
+
+            ADAPTER_ROWS="${ADAPTER_ROWS}$(printf '%-14s%-12s%s' "${display_name}" "${adapter_status}" "${adapter_weights}")\n"
+            ADAPTER_COUNT=$((ADAPTER_COUNT + 1))
+        done
+
+        echo "Adapters (LoRA):  [max: <%= maxLoras %> GPU / 70 CPU]"
+        if [ "${ADAPTER_COUNT}" -eq 0 ]; then
+            echo "  No adapters deployed"
+        else
+            printf '%-14s%-12s%s\n' "NAME" "STATUS" "WEIGHTS"
+            echo -e "${ADAPTER_ROWS}" | head -n -1
+        fi
+    else
+        echo "Adapters (LoRA):  [max: <%= maxLoras %> GPU / 70 CPU]"
+        echo "  No adapters deployed"
+    fi
+    echo ""
+fi
+<% } %>

package/templates/do/test

@@ -15,10 +15,11 @@ source "${SCRIPT_DIR}/config"
 # SageMaker Real-Time Inference Testing
 # ============================================================
 
-# Parse arguments
-ENDPOINT_NAME="${1:-${ENDPOINT_NAME:-}}"
+# Parse arguments: ./do/test [<ic-name>]
+IC_ARG="${1:-}"
 
-if [ -z "${ENDPOINT_NAME}" ]; then
+# Determine test mode based on ENDPOINT_NAME in config
+if [ -z "${ENDPOINT_NAME:-}" ]; then
     echo "🧪 Testing local container at localhost:8080"
     echo "   Project: ${PROJECT_NAME}"
     echo "   Framework: ${FRAMEWORK}"
@@ -210,8 +211,53 @@ else
     # Create temporary file for response
     TEMP_RESPONSE=$(mktemp)
     
-    # Invoke endpoint via inference component
-    IC_NAME="${INFERENCE_COMPONENT_NAME:-}"
+    # Resolve inference component name
+    # Precedence: do/adapters/ → do/ic/ → legacy config
+    IC_NAME=""
+    if [ -n "${IC_ARG}" ] && [ -f "${SCRIPT_DIR}/adapters/${IC_ARG}.conf" ]; then
+        # Argument matches an adapter name — use adapter IC
+        ADAPTER_IC_NAME=""
+        source "${SCRIPT_DIR}/adapters/${IC_ARG}.conf"
+        if [ -z "${ADAPTER_IC_NAME}" ]; then
+            echo "❌ Adapter '${IC_ARG}' conf is missing ADAPTER_IC_NAME."
+            exit 1
+        fi
+        IC_NAME="${ADAPTER_IC_NAME}"
+    elif [ -n "${IC_ARG}" ]; then
+        # Explicit IC name provided as argument
+        IC_CONF="${SCRIPT_DIR}/ic/${IC_ARG}.conf"
+        if [ ! -f "${IC_CONF}" ]; then
+            echo "❌ IC config not found: do/ic/${IC_ARG}.conf"
+            exit 1
+        fi
+        IC_DEPLOYED_NAME=""
+        source "${IC_CONF}"
+        if [ -z "${IC_DEPLOYED_NAME}" ]; then
+            echo "❌ IC '${IC_ARG}' has not been deployed yet. Run ./do/deploy --ic ${IC_ARG} first."
+            exit 1
+        fi
+        IC_NAME="${IC_DEPLOYED_NAME}"
+    elif [ -d "${SCRIPT_DIR}/ic" ]; then
+        # No argument, but do/ic/ exists — use first IC alphabetically
+        IC_NAME=""
+        for conf in "${SCRIPT_DIR}"/ic/*.conf; do
+            [ -f "${conf}" ] || continue
+            IC_DEPLOYED_NAME=""
+            source "${conf}"
+            if [ -n "${IC_DEPLOYED_NAME}" ]; then
+                IC_NAME="${IC_DEPLOYED_NAME}"
+                break
+            fi
+        done
+        if [ -z "${IC_NAME}" ]; then
+            echo "❌ No ICs deployed. Run ./do/deploy first."
+            exit 1
+        fi
+    else
+        # Legacy: no do/ic/ directory, use INFERENCE_COMPONENT_NAME from do/config
+        IC_NAME="${INFERENCE_COMPONENT_NAME:-}"
+    fi
+
     INVOKE_ARGS=(
         --endpoint-name "${ENDPOINT_NAME}"
         --region "${AWS_REGION}"
@@ -285,12 +331,15 @@ if [ "${TEST_MODE}" = "local" ]; then
     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"
+    echo "📋 What's next?"
+<% if (typeof includeBenchmark !== 'undefined' && includeBenchmark) { %>
+    echo "   • Benchmark performance:      ./do/benchmark"
+<% } %>
+<% if (typeof enableLora !== 'undefined' && enableLora) { %>
+    echo "   • Add a LoRA adapter:         ./do/adapter add <name> --weights s3://..."
+<% } %>
+    echo "   • Register this deployment:   ./do/register"
+    echo "   • View logs:                  ./do/logs"
 fi
 
 <% } else if (deploymentTarget === 'async-inference') { %>
@@ -599,13 +648,13 @@ if [ "${TEST_MODE}" = "local" ]; then
     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"
+    echo "📋 What's next?"
+<% if (typeof includeBenchmark !== 'undefined' && includeBenchmark) { %>
+    echo "   • Benchmark performance:      ./do/benchmark"
+<% } %>
+    echo "   • Check async output:         aws s3 ls ${ASYNC_S3_OUTPUT_PATH}"
+    echo "   • Register this deployment:   ./do/register"
+    echo "   • View logs:                  ./do/logs"
 fi
 
 <% } else if (deploymentTarget === 'hyperpod-eks') { %>
@@ -864,13 +913,14 @@ if [ "${TEST_TARGET}" = "local" ]; then
     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"
+    echo "📋 What's next?"
+<% if (typeof includeBenchmark !== 'undefined' && includeBenchmark) { %>
+    echo "   • Benchmark performance:      ./do/benchmark"
+<% } %>
+    echo "   • Check pod status:           kubectl get pods -n ${HYPERPOD_NAMESPACE}"
+    echo "   • View pod logs:              kubectl logs -n ${HYPERPOD_NAMESPACE} -l app=${PROJECT_NAME}"
+    echo "   • Register this deployment:   ./do/register"
+    echo "   • View logs:                  ./do/logs"
 fi
 
 <% } else if (deploymentTarget === 'batch-transform') { %>
@@ -950,8 +1000,10 @@ case "${TEST_TARGET}" in
                 echo ""
                 echo "✅ All tests passed!"
                 echo ""
-                echo "📝 Register this deployment:"
-                echo "  ./do/register"
+                echo "📋 What's next?"
+                echo "   • View results:               cat batch-output/"
+                echo "   • Register this deployment:   ./do/register"
+                echo "   • View logs:                  ./do/logs"
                 ;;
             InProgress)
                 echo "⏳ Transform job is still in progress"
@@ -1144,4 +1196,4 @@ echo "Next steps:"
 echo "  • Push to ECR: ./do/push"
 echo "  • Deploy batch transform: ./do/deploy"
 
-<% } %>
+<% } %>