@aws/ml-container-creator 0.6.0 → 0.7.1

package/templates/do/adapter

@@ -29,6 +29,7 @@ _usage() {
     echo "Commands:"
     echo "  add <name> --weights <s3-uri>        Add a new LoRA adapter from S3"
     echo "  add <name> --from-hub <hf-repo-id>   Add a new LoRA adapter from HuggingFace Hub"
+    echo "  add <name> --from-tune [technique]   Add adapter from do/tune output"
     echo "  list                                  List all adapters on the endpoint"
     echo "  remove <name>                         Remove an adapter"
     echo "  update <name> --weights <new-s3-uri>  Update adapter weights from S3"
@@ -41,6 +42,8 @@ _usage() {
     echo "Examples:"
     echo "  ./do/adapter add ectsum --weights s3://my-bucket/adapters/ectsum/adapter.tar.gz"
     echo "  ./do/adapter add ectsum --from-hub predibase/llama-3.1-8b-ectsum"
+    echo "  ./do/adapter add tuned-sft --from-tune"
+    echo "  ./do/adapter add tuned-sft --from-tune sft"
     echo "  ./do/adapter list"
     echo "  ./do/adapter remove ectsum"
     echo "  ./do/adapter update ectsum --weights s3://my-bucket/adapters/ectsum-v2/adapter.tar.gz"
@@ -48,7 +51,7 @@ _usage() {
     echo ""
     echo "Adapter metadata is stored in do/adapters/<name>.conf"
     echo ""
-    echo "Note: --weights and --from-hub are mutually exclusive."
+    echo "Note: --weights, --from-hub, and --from-tune are mutually exclusive."
 }
 
 # ── Validate LoRA is enabled ──────────────────────────────────────────────────
@@ -357,6 +360,8 @@ _adapter_add() {
     local adapter_name=""
     local weights_uri=""
     local from_hub=""
+    local from_tune=""
+    local from_tune_technique=""
 
     # Parse add arguments
     shift  # remove 'add' from args
@@ -380,28 +385,45 @@ _adapter_add() {
                 from_hub="$2"
                 shift 2
                 ;;
+            --from-tune)
+                from_tune="true"
+                # Check if next argument is a technique (not another flag and not empty)
+                if [ -n "${2:-}" ] && [[ "${2}" != -* ]]; then
+                    from_tune_technique="$2"
+                    shift 2
+                else
+                    shift
+                fi
+                ;;
             --help|-h)
                 echo "Usage: ./do/adapter add <name> --weights <s3-uri>"
                 echo "       ./do/adapter add <name> --from-hub <hf-repo-id>"
+                echo "       ./do/adapter add <name> --from-tune [technique]"
                 echo ""
                 echo "Add a new LoRA adapter to the endpoint."
                 echo ""
                 echo "Arguments:"
-                echo "  <name>                    Adapter name (lowercase alphanumeric + hyphens, 1-50 chars)"
-                echo "  --weights <s3-uri>        S3 URI to adapter weights (.tar.gz)"
-                echo "  --from-hub <hf-repo-id>   Download adapter from HuggingFace Hub"
+                echo "  <name>                      Adapter name (lowercase alphanumeric + hyphens, 1-50 chars)"
+                echo "  --weights <s3-uri>          S3 URI to adapter weights (.tar.gz)"
+                echo "  --from-hub <hf-repo-id>     Download adapter from HuggingFace Hub"
+                echo "  --from-tune [technique]     Use adapter output from do/tune"
+                echo "                              Without technique: uses latest tune output"
+                echo "                              With technique (e.g., sft, dpo): uses technique-specific output"
                 echo ""
-                echo "Note: --weights and --from-hub are mutually exclusive."
+                echo "Note: --weights, --from-hub, and --from-tune are mutually exclusive."
                 echo ""
                 echo "Examples:"
                 echo "  ./do/adapter add ectsum --weights s3://bucket/adapters/ectsum/adapter.tar.gz"
                 echo "  ./do/adapter add ectsum --from-hub predibase/llama-3.1-8b-ectsum"
+                echo "  ./do/adapter add tuned-sft --from-tune"
+                echo "  ./do/adapter add tuned-sft --from-tune sft"
                 exit 0
                 ;;
             -*)
                 echo "❌ Unknown option: $1"
                 echo "   Usage: ./do/adapter add <name> --weights <s3-uri>"
                 echo "          ./do/adapter add <name> --from-hub <hf-repo-id>"
+                echo "          ./do/adapter add <name> --from-tune [technique]"
                 exit 1
                 ;;
             *)
@@ -411,6 +433,7 @@ _adapter_add() {
                     echo "❌ Unexpected argument: $1"
                     echo "   Usage: ./do/adapter add <name> --weights <s3-uri>"
                     echo "          ./do/adapter add <name> --from-hub <hf-repo-id>"
+                    echo "          ./do/adapter add <name> --from-tune [technique]"
                     exit 1
                 fi
                 shift
@@ -423,26 +446,85 @@ _adapter_add() {
         echo "❌ Adapter name is required"
         echo "   Usage: ./do/adapter add <name> --weights <s3-uri>"
         echo "          ./do/adapter add <name> --from-hub <hf-repo-id>"
+        echo "          ./do/adapter add <name> --from-tune [technique]"
         exit 1
     fi
 
     # ── Mutual exclusivity check ─────────────────────────────────────────
-    if [ -n "${weights_uri}" ] && [ -n "${from_hub}" ]; then
-        echo "❌ --weights and --from-hub are mutually exclusive"
+    local source_count=0
+    [ -n "${weights_uri}" ] && source_count=$((source_count + 1))
+    [ -n "${from_hub}" ] && source_count=$((source_count + 1))
+    [ -n "${from_tune}" ] && source_count=$((source_count + 1))
+
+    if [ "${source_count}" -gt 1 ]; then
+        echo "❌ --weights, --from-hub, and --from-tune are mutually exclusive"
         echo ""
-        echo "   Use one or the other:"
+        echo "   Use one of:"
         echo "   ./do/adapter add ${adapter_name} --weights <s3-uri>"
         echo "   ./do/adapter add ${adapter_name} --from-hub <hf-repo-id>"
+        echo "   ./do/adapter add ${adapter_name} --from-tune [technique]"
         exit 1
     fi
 
-    if [ -z "${weights_uri}" ] && [ -z "${from_hub}" ]; then
-        echo "❌ Either --weights or --from-hub is required"
+    if [ "${source_count}" -eq 0 ]; then
+        echo "❌ One of --weights, --from-hub, or --from-tune is required"
         echo "   Usage: ./do/adapter add <name> --weights <s3-uri>"
         echo "          ./do/adapter add <name> --from-hub <hf-repo-id>"
+        echo "          ./do/adapter add <name> --from-tune [technique]"
         exit 1
     fi
 
+    # ── Resolve --from-tune to weights_uri ────────────────────────────────
+    if [ -n "${from_tune}" ]; then
+        if [ -n "${from_tune_technique}" ]; then
+            # Technique-specific: read TUNE_ADAPTER_PATH_<TECHNIQUE>
+            local technique_upper
+            technique_upper=$(echo "${from_tune_technique}" | tr '[:lower:]' '[:upper:]')
+            local tune_var="TUNE_ADAPTER_PATH_${technique_upper}"
+            local tune_path="${!tune_var:-}"
+
+            if [ -z "${tune_path}" ]; then
+                echo "❌ No adapter output found for technique: ${from_tune_technique}"
+                echo ""
+                echo "   ${tune_var} is not set in do/config."
+                echo ""
+                echo "   Run a tune job first:"
+                echo "   ./do/tune --technique ${from_tune_technique} --dataset <source>"
+                exit 1
+            fi
+
+            weights_uri="${tune_path}"
+            echo "📦 Using tune adapter output for technique '${from_tune_technique}': ${weights_uri}"
+        else
+            # No technique: read TUNE_OUTPUT_PATH_LATEST and verify type
+            if [ -z "${TUNE_OUTPUT_PATH_LATEST:-}" ]; then
+                echo "❌ No tune output found."
+                echo ""
+                echo "   TUNE_OUTPUT_PATH_LATEST is not set in do/config."
+                echo ""
+                echo "   Run a tune job first:"
+                echo "   ./do/tune --technique <technique> --dataset <source>"
+                exit 1
+            fi
+
+            # Verify output type is adapter (not full-model)
+            if [ "${TUNE_OUTPUT_TYPE_LATEST:-}" = "full-model" ]; then
+                echo "❌ Latest tune output is a full model, not an adapter."
+                echo ""
+                echo "   TUNE_OUTPUT_TYPE_LATEST=full-model"
+                echo ""
+                echo "   Full model outputs cannot be added as adapters."
+                echo "   Use do/add-ic instead:"
+                echo "   ./do/add-ic ${adapter_name} --from-tune"
+                exit 1
+            fi
+
+            weights_uri="${TUNE_OUTPUT_PATH_LATEST}"
+            echo "📦 Using latest tune adapter output: ${weights_uri}"
+        fi
+        echo ""
+    fi
+
     # ── Validate HF repo ID format (if --from-hub) ───────────────────────
     if [ -n "${from_hub}" ]; then
         # Valid formats: "org/name" or "name" (alphanumeric, hyphens, underscores, dots)
@@ -468,8 +550,8 @@ _adapter_add() {
         exit 1
     fi
 
-    # ── Validate S3 URI format (only when --weights is used) ─────────────
-    if [ -n "${weights_uri}" ]; then
+    # ── Validate S3 URI format (only when --weights is explicitly used) ──
+    if [ -n "${weights_uri}" ] && [ -z "${from_hub}" ] && [ -z "${from_tune}" ]; then
         if ! echo "${weights_uri}" | grep -qE '^s3://.*\.tar\.gz$'; then
             echo "❌ Invalid S3 URI: ${weights_uri}"
             echo ""
@@ -495,6 +577,9 @@ _adapter_add() {
     echo "🔌 Adding adapter: ${adapter_name}"
     if [ -n "${from_hub}" ]; then
         echo "   Source: HuggingFace Hub (${from_hub})"
+    elif [ -n "${from_tune}" ]; then
+        echo "   Source: do/tune output"
+        echo "   Weights: ${weights_uri}"
     else
         echo "   Weights: ${weights_uri}"
     fi
@@ -593,6 +678,14 @@ export ADAPTER_HF_REPO="${from_hub}"
 EOF
     fi
 
+    # Add tune-specific metadata if --from-tune was used
+    if [ -n "${from_tune}" ]; then
+        cat >> "${SCRIPT_DIR}/adapters/${adapter_name}.conf" <<EOF
+export ADAPTER_SOURCE="tune"
+export ADAPTER_TUNE_TECHNIQUE="${from_tune_technique:-latest}"
+EOF
+    fi
+
     echo ""
     echo "✅ Adapter added successfully!"
     echo ""
@@ -602,6 +695,8 @@ EOF
     echo "   Weights: ${weights_uri}"
     if [ -n "${from_hub}" ]; then
         echo "   Source: HuggingFace Hub (${from_hub})"
+    elif [ -n "${from_tune}" ]; then
+        echo "   Source: do/tune (${from_tune_technique:-latest})"
     fi
     echo "   Created: ${created_at}"
     echo ""

package/templates/do/add-ic

@@ -4,6 +4,10 @@
 #
 # Add a new inference component to this project.
 # Creates a new IC config file in do/ic/ and deploys it immediately.
+#
+# Usage:
+#   ./do/add-ic [name] [--from-tune] [--model-data <s3-uri>]
+#   ./do/add-ic --help
 
 set -e
 set -u
@@ -12,38 +16,157 @@ set -o pipefail
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 source "${SCRIPT_DIR}/config"
 
+# ============================================================
+# Usage
+# ============================================================
+_usage() {
+    echo "Usage: ./do/add-ic [name] [options]"
+    echo ""
+    echo "Add a new inference component to this project."
+    echo ""
+    echo "Arguments:"
+    echo "  [name]                     IC name (optional, prompted if not provided)"
+    echo ""
+    echo "Options:"
+    echo "  --from-tune                Use model output from do/tune (reads TUNE_OUTPUT_PATH_LATEST)"
+    echo "  --model-data <s3-uri>      S3 URI to model data (tar.gz or model directory)"
+    echo "  --help, -h                 Show this help message"
+    echo ""
+    echo "Note: --from-tune and --model-data are mutually exclusive."
+    echo ""
+    echo "Examples:"
+    echo "  ./do/add-ic                                    # Interactive mode"
+    echo "  ./do/add-ic tuned-v1 --from-tune               # Use latest tune output"
+    echo "  ./do/add-ic tuned-v1 --model-data s3://bucket/model.tar.gz"
+    echo ""
+}
+
+# ============================================================
+# Parse arguments
+# ============================================================
+IC_NAME=""
+FROM_TUNE=""
+MODEL_DATA=""
+
+while [ $# -gt 0 ]; do
+    case "$1" in
+        --from-tune)
+            FROM_TUNE="true"
+            shift
+            ;;
+        --model-data)
+            if [ -z "${2:-}" ]; then
+                echo "❌ --model-data requires an S3 URI argument"
+                echo "   Usage: ./do/add-ic <name> --model-data <s3-uri>"
+                exit 1
+            fi
+            MODEL_DATA="$2"
+            shift 2
+            ;;
+        --help|-h)
+            _usage
+            exit 0
+            ;;
+        -*)
+            echo "❌ Unknown option: $1"
+            _usage
+            exit 1
+            ;;
+        *)
+            if [ -z "${IC_NAME}" ]; then
+                IC_NAME="$1"
+            else
+                echo "❌ Unexpected argument: $1"
+                _usage
+                exit 1
+            fi
+            shift
+            ;;
+    esac
+done
+
+# ============================================================
+# Mutual exclusivity check
+# ============================================================
+if [ -n "${FROM_TUNE}" ] && [ -n "${MODEL_DATA}" ]; then
+    echo "❌ --from-tune and --model-data are mutually exclusive"
+    echo ""
+    echo "   Use one of:"
+    echo "   ./do/add-ic <name> --from-tune"
+    echo "   ./do/add-ic <name> --model-data <s3-uri>"
+    exit 1
+fi
+
+# ============================================================
+# Resolve --from-tune to MODEL_DATA
+# ============================================================
+if [ -n "${FROM_TUNE}" ]; then
+    if [ -z "${TUNE_OUTPUT_PATH_LATEST:-}" ]; then
+        echo "❌ No tune output found."
+        echo ""
+        echo "   TUNE_OUTPUT_PATH_LATEST is not set in do/config."
+        echo ""
+        echo "   Run a tune job first:"
+        echo "   ./do/tune --technique <technique> --dataset <source>"
+        exit 1
+    fi
+
+    MODEL_DATA="${TUNE_OUTPUT_PATH_LATEST}"
+    echo "📦 Using tune output: ${MODEL_DATA}"
+    echo ""
+fi
+
 echo "➕ Add New Inference Component"
 echo "   Project: ${PROJECT_NAME}"
+if [ -n "${MODEL_DATA}" ]; then
+    echo "   Model data: ${MODEL_DATA}"
+fi
 echo ""
 
 # ============================================================
-# Prompt for IC name
+# Prompt for IC name (if not provided as argument)
 # ============================================================
-while true; do
-    read -p "IC name (lowercase alphanumeric + hyphens): " IC_NAME
+if [ -z "${IC_NAME}" ]; then
+    while true; do
+        read -p "IC name (lowercase alphanumeric + hyphens): " IC_NAME
 
-    # Validate: non-empty
-    if [ -z "${IC_NAME}" ]; then
-        echo "   ❌ IC name cannot be empty."
-        continue
-    fi
+        # Validate: non-empty
+        if [ -z "${IC_NAME}" ]; then
+            echo "   ❌ IC name cannot be empty."
+            continue
+        fi
+
+        # Validate: lowercase alphanumeric + hyphens only
+        if ! echo "${IC_NAME}" | grep -qE '^[a-z0-9]([a-z0-9-]*[a-z0-9])?$'; then
+            echo "   ❌ IC name must be lowercase alphanumeric with hyphens (e.g., 'llama-70b')."
+            echo "      Must start and end with a letter or number."
+            continue
+        fi
 
-    # Validate: lowercase alphanumeric + hyphens only
+        # Validate: no collision with existing config
+        if [ -f "${SCRIPT_DIR}/ic/${IC_NAME}.conf" ]; then
+            echo "   ❌ IC config already exists: do/ic/${IC_NAME}.conf"
+            echo "      Choose a different name or edit the existing config."
+            continue
+        fi
+
+        break
+    done
+else
+    # Validate provided IC name
     if ! echo "${IC_NAME}" | grep -qE '^[a-z0-9]([a-z0-9-]*[a-z0-9])?$'; then
-        echo "   ❌ IC name must be lowercase alphanumeric with hyphens (e.g., 'llama-70b')."
-        echo "      Must start and end with a letter or number."
-        continue
+        echo "❌ Invalid IC name: ${IC_NAME}"
+        echo "   IC name must be lowercase alphanumeric with hyphens (e.g., 'llama-70b')."
+        echo "   Must start and end with a letter or number."
+        exit 1
     fi
 
-    # Validate: no collision with existing config
     if [ -f "${SCRIPT_DIR}/ic/${IC_NAME}.conf" ]; then
-        echo "   ❌ IC config already exists: do/ic/${IC_NAME}.conf"
-        echo "      Choose a different name or edit the existing config."
-        continue
+        echo "❌ IC config already exists: do/ic/${IC_NAME}.conf"
+        echo "   Choose a different name or edit the existing config."
+        exit 1
     fi
-
-    break
-done
+fi
 
 # ============================================================
 # Prompt for image tag
@@ -107,6 +230,16 @@ export IC_GPU_COUNT=${IC_GPU_COUNT}
 export IC_COPY_COUNT=${IC_COPY_COUNT}
 export IC_MIN_MEMORY_MB=${IC_MIN_MEMORY_MB}
 export IC_STARTUP_TIMEOUT=900
+EOF
+
+# Add model data if provided (from --from-tune or --model-data)
+if [ -n "${MODEL_DATA}" ]; then
+    cat >> "${IC_CONF_PATH}" <<EOF
+export IC_MODEL_DATA="${MODEL_DATA}"
+EOF
+fi
+
+cat >> "${IC_CONF_PATH}" <<EOF
 
 # Optional overrides:
 # export IC_MODEL_NAME="my-model-v2"
@@ -120,6 +253,9 @@ echo "   Image tag:  ${IC_IMAGE_TAG}"
 echo "   GPU count:  ${IC_GPU_COUNT}"
 echo "   Copy count: ${IC_COPY_COUNT}"
 echo "   Memory MB:  ${IC_MIN_MEMORY_MB}"
+if [ -n "${MODEL_DATA}" ]; then
+    echo "   Model data: ${MODEL_DATA}"
+fi
 echo ""
 
 # ============================================================

package/templates/do/config

@@ -188,6 +188,12 @@ export NGC_API_KEY_ARN="<%= ngcTokenArn %>"
 <% } else if (ngcApiKey) { %>
 export NGC_API_KEY="<%= ngcApiKey %>"
 <% } %>
+
+<% if (deploymentTarget !== 'batch-transform') { %>
+# Managed Model Customization (do/tune)
+export TUNE_SUPPORTED=<%= (typeof tuneSupported !== 'undefined' && tuneSupported) ? 'true' : 'false' %>
+export TUNE_S3_BUCKET="mlcc-tune-$(aws sts get-caller-identity --query Account --output text 2>/dev/null || echo 'UNKNOWN')-${AWS_REGION}"
+<% } %>
 <% } %>
 
 <% if (framework === 'diffusors') { %>

package/templates/do/export

@@ -2,16 +2,33 @@
 # Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
 # SPDX-License-Identifier: Apache-2.0
 
-# Export current configuration as a CLI command or JSON object
-# Usage: ./do/export [--json]
+# Export current configuration as a CLI command, JSON object, or Jupyter notebook
+# Usage: ./do/export [--json | --notebook]
 
 # Source configuration (suppress the summary output)
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 source "${SCRIPT_DIR}/config" > /dev/null 2>&1
 
+# ── Notebook output mode ──────────────────────────────────────────────────────
+
+if [ "${1:-}" = "--notebook" ]; then
+    # Ensure not combined with --json
+    if [ "${2:-}" = "--json" ]; then
+        echo "Error: --notebook and --json are mutually exclusive" >&2
+        exit 1
+    fi
+    python3 "${SCRIPT_DIR}/../deploy_notebook_generator.py"
+    exit 0
+fi
+
 # ── JSON output mode ─────────────────────────────────────────────────────────
 
 if [ "${1:-}" = "--json" ]; then
+    # Ensure not combined with --notebook
+    if [ "${2:-}" = "--notebook" ]; then
+        echo "Error: --notebook and --json are mutually exclusive" >&2
+        exit 1
+    fi
     # Build a JSON object with all configuration parameters.
     # Uses ConfigManager camelCase keys so the output can be fed directly
     # back into the generator via --config=<file>.

package/templates/do/lib/endpoint-config.sh

@@ -152,7 +152,9 @@ create_endpoint_config() {
         variant_json="${variant_json}}]"
     else
         # Standard path: single instance type
-        variant_json="[{\"VariantName\":\"AllTraffic\",\"InstanceType\":\"${INSTANCE_TYPE}\",\"InitialInstanceCount\":1"
+        # RoutingConfig is required for IC-based endpoints — without it the IC scheduler
+        # cannot place containers and the IC stays in Creating with no logs.
+        variant_json="[{\"VariantName\":\"AllTraffic\",\"InstanceType\":\"${INSTANCE_TYPE}\",\"InitialInstanceCount\":1,\"RoutingConfig\":{\"RoutingStrategy\":\"LEAST_OUTSTANDING_REQUESTS\"}"
 
         # Optional: AMI version
         if [ -n "${INFERENCE_AMI_VERSION:-}" ]; then

package/templates/do/lib/inference-component.sh

@@ -46,10 +46,14 @@ create_inference_component() {
 
     # Build container spec JSON
     local container_spec="{\"Image\":\"${ECR_REPOSITORY}:${IC_IMAGE_TAG:-${PROJECT_NAME}-latest}\""
+    # Always inject IC name for CW log forwarder
+    local ic_env="\"INFERENCE_COMPONENT_NAME\":\"${ic_name}\""
     if [ -n "${CONTAINER_ENV_JSON}${IC_CONTAINER_ENV_EXTRA:-}" ]; then
         local env_json="${CONTAINER_ENV_JSON}"
         [ -n "${IC_CONTAINER_ENV_EXTRA:-}" ] && env_json="${env_json:+${env_json},}${IC_CONTAINER_ENV_EXTRA}"
-        container_spec="${container_spec},\"Environment\":{${env_json}}"
+        container_spec="${container_spec},\"Environment\":{${ic_env},${env_json}}"
+    else
+        container_spec="${container_spec},\"Environment\":{${ic_env}}"
     fi
     container_spec="${container_spec}}"