@aws/ml-container-creator 0.6.1 → 0.8.0

package/templates/diffusors/serve

@@ -9,10 +9,10 @@ echo "Starting vLLM-Omni server (diffusion model serving)"
 
 # Resolve model URI prefixes that engines cannot handle natively.
 # The generator's model-picker may store provider-specific URIs
-# (e.g. jumpstart://model-txt2img-stabilityai-stable-diffusion-v2-1-base)
-# as the model identifier. vLLM expects a HuggingFace repo ID or local path.
+# (e.g. registry://my-model-group/1) as the model identifier.
+# vLLM expects a HuggingFace repo ID or local path.
 _RAW_MODEL="${VLLM_MODEL:-}"
-if [[ "$_RAW_MODEL" == jumpstart://* ]] || [[ "$_RAW_MODEL" == jumpstart-hub://* ]] || [[ "$_RAW_MODEL" == registry://* ]]; then
+if [[ "$_RAW_MODEL" == registry://* ]]; then
     if [ -d /opt/ml/model ] && [ "$(ls -A /opt/ml/model 2>/dev/null)" ]; then
         echo "Resolved VLLM_MODEL='${_RAW_MODEL}' → /opt/ml/model (local artifacts found)"
         export VLLM_MODEL="/opt/ml/model"

package/templates/do/.tune_helper.py

@@ -176,7 +176,7 @@ def cmd_submit(args):
             )
         elif "ValidationException" in error_msg and "license" in error_msg.lower():
             _error_exit(
-                f"Model license not accepted. Accept the license in JumpStart before "
+                f"Model license not accepted. Accept the model license before "
                 f"using this model for customization. Details: {error_msg}"
             )
         else:
@@ -660,7 +660,7 @@ def main():
 
     # ── submit ────────────────────────────────────────────────────────────────
     submit_parser = subparsers.add_parser("submit", help="Submit a customization job")
-    submit_parser.add_argument("--model-id", required=True, help="JumpStart model ID")
+    submit_parser.add_argument("--model-id", required=True, help="Model ID")
     submit_parser.add_argument("--technique", required=True,
                                choices=["sft", "dpo", "rlaif", "rlvr"],
                                help="Customization technique")

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}}"
 

package/templates/do/register

@@ -191,8 +191,14 @@ fi
 # ============================================================
 
 # DEPLOYMENT_CONFIG format: <architecture>-<backend> (e.g., transformers-vllm, http-flask, triton-fil)
-ARCHITECTURE="${DEPLOYMENT_CONFIG%%-*}"
-BACKEND="${DEPLOYMENT_CONFIG#*-}"
+# Special case: marketplace has no backend
+if [ "${DEPLOYMENT_CONFIG}" = "marketplace" ]; then
+    ARCHITECTURE="marketplace"
+    BACKEND=""
+else
+    ARCHITECTURE="${DEPLOYMENT_CONFIG%%-*}"
+    BACKEND="${DEPLOYMENT_CONFIG#*-}"
+fi
 
 echo "📋 Registering deployment to registry"
 echo "   Project: ${PROJECT_NAME}"

package/templates/do/test

@@ -103,9 +103,9 @@ case "${FRAMEWORK}" in
         case "${MODEL_SERVER}" in
             vllm|sglang)
                 # OpenAI-compatible chat completions format
-                # For S3/JumpStart models, vLLM registers the model under the local path
+                # For S3/registry 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://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                if [[ "${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}'
@@ -431,7 +431,7 @@ case "${FRAMEWORK}" in
         case "${MODEL_SERVER}" in
             vllm|sglang)
                 VLLM_MODEL_NAME="${MODEL_NAME}"
-                if [[ "${MODEL_NAME}" == jumpstart://* ]] || [[ "${MODEL_NAME}" == jumpstart-hub://* ]] || [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                if [[ "${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}'
@@ -808,7 +808,7 @@ case "${FRAMEWORK}" in
             vllm|sglang)
                 # OpenAI-compatible chat completions format
                 VLLM_MODEL_NAME="${MODEL_NAME}"
-                if [[ "${MODEL_NAME}" == jumpstart://* ]] || [[ "${MODEL_NAME}" == jumpstart-hub://* ]] || [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
+                if [[ "${MODEL_NAME}" == s3://* ]] || [[ "${MODEL_NAME}" == /opt/ml/* ]]; then
                     VLLM_MODEL_NAME="/opt/ml/model"
                 fi
                 TEST_PAYLOAD='{"model": "'"${VLLM_MODEL_NAME}"'", "messages": [{"role": "user", "content": "What is machine learning?"}], "max_tokens": 50, "temperature": 0.7}'
@@ -1095,7 +1095,7 @@ case "${FRAMEWORK}" in
         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
+                if [[ "${MODEL_NAME}" == s3://* ]] || [[ "${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}'

package/templates/do/tune

@@ -67,7 +67,7 @@ _parse_args() {
                 ARG_TRAINING_TYPE="$2"; shift 2 ;;
             --model)
                 if [ -z "${2:-}" ]; then
-                    echo "❌ --model requires a JumpStart model ID"
+                    echo "❌ --model requires a model ID"
                     exit 1
                 fi
                 ARG_MODEL="$2"; shift 2 ;;
@@ -287,7 +287,7 @@ for family in sorted(families.keys()):
     for entry in entries:
         techniques = list(entry.get('techniques', {}).keys())
         print(f'    • {entry[\"displayName\"]}')
-        print(f'      ID: {entry[\"jumpStartModelId\"]}')
+        print(f'      ID: {entry[\"modelId\"]}')
         for t in techniques:
             tc = entry['techniques'][t]
             types = ', '.join(tc.get('trainingTypes', []))

package/templates/marketplace/config

@@ -0,0 +1,118 @@
+#!/bin/bash
+# do-framework configuration (marketplace)
+# This file is sourced by all do scripts
+
+# Project identification
+export PROJECT_NAME="<%= projectName %>"
+export DEPLOYMENT_CONFIG="marketplace"
+
+# Marketplace model package
+export MODEL_PACKAGE_ARN="<%= modelPackageArn %>"
+
+# AWS configuration
+export AWS_REGION="<%= awsRegion %>"
+
+# Deployment configuration
+export DEPLOYMENT_TARGET="<%= deploymentTarget %>"
+export INSTANCE_TYPE="<%= instanceType %>"
+
+<% if (roleArn) { %>
+export ROLE_ARN="<%= roleArn %>"
+<% } %>
+
+<% if (deploymentTarget === 'async-inference') { %>
+# Async-specific configuration
+ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text 2>/dev/null || echo "UNKNOWN")
+
+<% if (asyncS3OutputPath) { %>
+export ASYNC_S3_OUTPUT_PATH="<%= asyncS3OutputPath %>"
+<% } else { %>
+export ASYNC_S3_OUTPUT_PATH="s3://mlcc-async-${ACCOUNT_ID}-${AWS_REGION}/${PROJECT_NAME}/output/"
+<% } %>
+
+<% if (asyncSnsSuccessTopic) { %>
+export ASYNC_SNS_SUCCESS_TOPIC="<%= asyncSnsSuccessTopic %>"
+<% } else { %>
+export ASYNC_SNS_SUCCESS_TOPIC="arn:aws:sns:${AWS_REGION}:${ACCOUNT_ID}:ml-container-creator-${PROJECT_NAME}-async-success"
+<% } %>
+
+<% if (asyncSnsErrorTopic) { %>
+export ASYNC_SNS_ERROR_TOPIC="<%= asyncSnsErrorTopic %>"
+<% } else { %>
+export ASYNC_SNS_ERROR_TOPIC="arn:aws:sns:${AWS_REGION}:${ACCOUNT_ID}:ml-container-creator-${PROJECT_NAME}-async-error"
+<% } %>
+
+<% if (asyncMaxConcurrentInvocations) { %>
+export ASYNC_MAX_CONCURRENT_INVOCATIONS="<%= asyncMaxConcurrentInvocations %>"
+<% } %>
+<% } %>
+
+<% if (deploymentTarget === 'batch-transform') { %>
+# Batch Transform configuration
+ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text 2>/dev/null || echo "UNKNOWN")
+
+<% if (batchInputPath) { %>
+export BATCH_INPUT_PATH="<%= batchInputPath %>"
+<% } else { %>
+export BATCH_INPUT_PATH="s3://mlcc-batch-${ACCOUNT_ID}-${AWS_REGION}/${PROJECT_NAME}/input/"
+<% } %>
+<% if (batchOutputPath) { %>
+export BATCH_OUTPUT_PATH="<%= batchOutputPath %>"
+<% } else { %>
+export BATCH_OUTPUT_PATH="s3://mlcc-batch-${ACCOUNT_ID}-${AWS_REGION}/${PROJECT_NAME}/output/"
+<% } %>
+export BATCH_INSTANCE_COUNT="<%= batchInstanceCount %>"
+export BATCH_SPLIT_TYPE="<%= batchSplitType %>"
+export BATCH_STRATEGY="<%= batchStrategy %>"
+export BATCH_JOIN_SOURCE="<%= batchJoinSource || 'None' %>"
+<% if (batchMaxConcurrentTransforms) { %>
+export BATCH_MAX_CONCURRENT_TRANSFORMS="<%= batchMaxConcurrentTransforms %>"
+<% } %>
+<% if (batchMaxPayloadInMB) { %>
+export BATCH_MAX_PAYLOAD_IN_MB="<%= batchMaxPayloadInMB %>"
+<% } %>
+<% } %>
+
+<% if (typeof includeBenchmark !== 'undefined' && includeBenchmark) { %>
+# SageMaker AI Benchmarking configuration
+export BENCHMARK_CONCURRENCY="<%= benchmarkConcurrency %>"
+export BENCHMARK_INPUT_TOKENS_MEAN="<%= benchmarkInputTokensMean %>"
+export BENCHMARK_OUTPUT_TOKENS_MEAN="<%= benchmarkOutputTokensMean %>"
+export BENCHMARK_STREAMING="<%= benchmarkStreaming %>"
+<% if (benchmarkRequestCount) { %>
+export BENCHMARK_REQUEST_COUNT="<%= benchmarkRequestCount %>"
+<% } else { %>
+export BENCHMARK_REQUEST_COUNT=""
+<% } %>
+<% if (benchmarkS3OutputPath) { %>
+export BENCHMARK_S3_OUTPUT_PATH="<%= benchmarkS3OutputPath %>"
+<% } else { %>
+export BENCHMARK_S3_OUTPUT_PATH="s3://mlcc-benchmark-$(aws sts get-caller-identity --query Account --output text)-${AWS_REGION}/${PROJECT_NAME}/"
+<% } %>
+export BENCHMARK_JOB_NAME=""
+export BENCHMARK_WORKLOAD_CONFIG_NAME=""
+<% } %>
+
+# Allow environment variable overrides
+export AWS_REGION=${AWS_REGION:-<%= awsRegion %>}
+export INSTANCE_TYPE=${INSTANCE_TYPE:-<%= instanceType %>}
+
+# Print configuration summary
+echo "⚙️  Configuration loaded"
+echo "   Project: ${PROJECT_NAME}"
+echo "   Config:  ${DEPLOYMENT_CONFIG}"
+echo "   Region:  ${AWS_REGION}"
+echo "   Model package: ${MODEL_PACKAGE_ARN}"
+echo "   Deployment target: ${DEPLOYMENT_TARGET}"
+echo "   Instance: ${INSTANCE_TYPE}"
+<% if (deploymentTarget === 'async-inference') { %>
+echo "   S3 output: ${ASYNC_S3_OUTPUT_PATH}"
+echo "   SNS success: ${ASYNC_SNS_SUCCESS_TOPIC}"
+echo "   SNS error: ${ASYNC_SNS_ERROR_TOPIC}"
+<% } else if (deploymentTarget === 'batch-transform') { %>
+echo "   Instance count: ${BATCH_INSTANCE_COUNT}"
+echo "   S3 input: ${BATCH_INPUT_PATH}"
+echo "   S3 output: ${BATCH_OUTPUT_PATH}"
+echo "   Split type: ${BATCH_SPLIT_TYPE}"
+echo "   Strategy: ${BATCH_STRATEGY}"
+<% } %>