@aws/ml-container-creator 0.4.0 → 0.6.0

package/bin/cli.js

@@ -98,6 +98,9 @@ program
     .addOption(new Option('--include-sample', 'Include sample model code'))
     .addOption(new Option('--include-testing', 'Include test suite'))
     .addOption(new Option('--test-types <types>', 'Comma-separated test types'))
+    .addOption(new Option('--enable-lora', 'Enable LoRA adapter serving (transformers with vllm/sglang/djl-lmi only)'))
+    .addOption(new Option('--max-loras <n>', 'Maximum concurrent LoRA adapters in GPU memory (default: 30)'))
+    .addOption(new Option('--max-lora-rank <n>', 'Maximum LoRA rank (default: 64)'))
 
     // --- MCP & Discovery ---
     .addOption(new Option('--smart', 'Enable Bedrock-powered smart mode on MCP servers'))
@@ -190,7 +193,7 @@ program.configureHelp({
                 groups.env.push(opt);
             } else if (['--hf-token', '--hf-token-arn', '--ngc-token', '--ngc-token-arn'].includes(long)) {
                 groups.auth.push(opt);
-            } else if (['--include-sample', '--include-testing', '--test-types'].includes(long)) {
+            } else if (['--include-sample', '--include-testing', '--test-types', '--enable-lora', '--max-loras', '--max-lora-rank'].includes(long)) {
                 groups.features.push(opt);
             } else if (['--smart', '--discover'].includes(long)) {
                 groups.mcp.push(opt);
@@ -307,7 +310,6 @@ program
 program
     .command('registry')
     .description('Registry operations (list, get, remove, replay, export, import, search) — experimental, may be reconciled with do/register')
-    .passThroughOptions()
     .argument('<action>', 'Registry action (log, list, get, remove, replay, export, import, search)')
     .argument('[args...]', 'Additional arguments')
     .option('--backend <backend>', 'Filter by backend')
@@ -328,6 +330,7 @@ program
     .option('--notes <text>', 'Deployment notes')
     .option('--project', 'Use project-level registry')
     .option('--parameters <json>', 'Parameters JSON string')
+    .option('--ic-list <json>', 'IC list JSON string')
     .option('--generator-version <version>', 'Generator version')
     // Options used by `registry list-architectures`
     .option('--server <name>', 'Filter by server name (for list-architectures)')

package/config/bootstrap-stack.json

@@ -7,7 +7,7 @@
       "Type": "String",
       "Default": "false",
       "AllowedValues": ["true", "false"],
-      "Description": "Whether to create S3 buckets for async inference and batch transform"
+      "Description": "Whether to create S3 buckets for async inference, batch transform, adapters, and benchmarks"
     },
     "UseExistingRoleArn": {
       "Type": "String",
@@ -134,8 +134,8 @@
                     "s3:ListBucket"
                   ],
                   "Resource": [
-                    "arn:aws:s3:::ml-container-creator-*",
-                    "arn:aws:s3:::ml-container-creator-*/*"
+                    "arn:aws:s3:::mlcc-*",
+                    "arn:aws:s3:::mlcc-*/*"
                   ]
                 },
                 {
@@ -209,7 +209,7 @@
       "DeletionPolicy": "Retain",
       "UpdateReplacePolicy": "Retain",
       "Properties": {
-        "BucketName": { "Fn::Sub": "${AWS::AccountId}-${AWS::Region}-ml-container-creator-async" },
+        "BucketName": { "Fn::Sub": "mlcc-async-${AWS::AccountId}-${AWS::Region}" },
         "VersioningConfiguration": { "Status": "Enabled" },
         "BucketEncryption": {
           "ServerSideEncryptionConfiguration": [
@@ -218,7 +218,8 @@
         },
         "Tags": [
           { "Key": "mlcc:managed-by", "Value": "ml-container-creator" },
-          { "Key": "mlcc:created-by", "Value": "bootstrap" }
+          { "Key": "mlcc:created-by", "Value": "bootstrap" },
+          { "Key": "mlcc:purpose", "Value": "async-inference-output" }
         ]
       }
     },
@@ -229,7 +230,7 @@
       "DeletionPolicy": "Retain",
       "UpdateReplacePolicy": "Retain",
       "Properties": {
-        "BucketName": { "Fn::Sub": "${AWS::AccountId}-${AWS::Region}-ml-container-creator-batch" },
+        "BucketName": { "Fn::Sub": "mlcc-batch-${AWS::AccountId}-${AWS::Region}" },
         "VersioningConfiguration": { "Status": "Enabled" },
         "BucketEncryption": {
           "ServerSideEncryptionConfiguration": [
@@ -238,17 +239,40 @@
         },
         "Tags": [
           { "Key": "mlcc:managed-by", "Value": "ml-container-creator" },
-          { "Key": "mlcc:created-by", "Value": "bootstrap" }
+          { "Key": "mlcc:created-by", "Value": "bootstrap" },
+          { "Key": "mlcc:purpose", "Value": "batch-transform-io" }
+        ]
+      }
+    },
+
+    "AdapterS3Bucket": {
+      "Type": "AWS::S3::Bucket",
+      "Condition": "ShouldCreateS3Buckets",
+      "DeletionPolicy": "Retain",
+      "UpdateReplacePolicy": "Retain",
+      "Properties": {
+        "BucketName": { "Fn::Sub": "mlcc-adapters-${AWS::AccountId}-${AWS::Region}" },
+        "VersioningConfiguration": { "Status": "Enabled" },
+        "BucketEncryption": {
+          "ServerSideEncryptionConfiguration": [
+            { "ServerSideEncryptionByDefault": { "SSEAlgorithm": "AES256" } }
+          ]
+        },
+        "Tags": [
+          { "Key": "mlcc:managed-by", "Value": "ml-container-creator" },
+          { "Key": "mlcc:created-by", "Value": "bootstrap" },
+          { "Key": "mlcc:purpose", "Value": "lora-adapter-storage" }
         ]
       }
     },
 
     "BenchmarkS3Bucket": {
       "Type": "AWS::S3::Bucket",
+      "Condition": "ShouldCreateS3Buckets",
       "DeletionPolicy": "Retain",
       "UpdateReplacePolicy": "Retain",
       "Properties": {
-        "BucketName": { "Fn::Sub": "ml-container-creator-benchmark-${AWS::Region}-${AWS::AccountId}" },
+        "BucketName": { "Fn::Sub": "mlcc-benchmark-${AWS::AccountId}-${AWS::Region}" },
         "VersioningConfiguration": { "Status": "Enabled" },
         "BucketEncryption": {
           "ServerSideEncryptionConfiguration": [
@@ -257,7 +281,8 @@
         },
         "Tags": [
           { "Key": "mlcc:managed-by", "Value": "ml-container-creator" },
-          { "Key": "mlcc:created-by", "Value": "bootstrap" }
+          { "Key": "mlcc:created-by", "Value": "bootstrap" },
+          { "Key": "mlcc:purpose", "Value": "benchmark-results" }
         ]
       }
     }
@@ -292,7 +317,13 @@
       "Description": "S3 bucket for batch transform I/O",
       "Value": { "Ref": "BatchS3Bucket" }
     },
+    "AdapterS3BucketName": {
+      "Condition": "ShouldCreateS3Buckets",
+      "Description": "S3 bucket for LoRA adapter storage",
+      "Value": { "Ref": "AdapterS3Bucket" }
+    },
     "BenchmarkS3BucketName": {
+      "Condition": "ShouldCreateS3Buckets",
       "Description": "S3 bucket for benchmark results output",
       "Value": { "Ref": "BenchmarkS3Bucket" }
     },

package/infra/ci-harness/buildspec.yml

@@ -40,6 +40,10 @@ phases:
       - REGISTER_DURATION=0
       - REGISTER_LOG_POINTER=""
       - REGISTER_ERROR_SUMMARY=""
+      - ADAPTER_TEST_STATUS="skip"
+      - ADAPTER_TEST_DURATION=0
+      - ADAPTER_TEST_LOG_POINTER=""
+      - ADAPTER_TEST_ERROR_SUMMARY=""
       - TEARDOWN_STATUS="skip"
       - TEARDOWN_DURATION=0
       - TEARDOWN_LOG_POINTER=""
@@ -182,6 +186,54 @@ phases:
         fi
       - rm -f "$STAGE_STDERR_FILE"
 
+      # --- Stage: Adapter_Test (only if do/adapters/ has .conf files) ---
+      - echo "=== Stage: Adapter_Test ==="
+      - STAGE_START=$(date +%s)
+      - ADAPTER_TEST_LOG_POINTER="$LOG_POINTER_PREFIX"
+      - STAGE_STDERR_FILE=$(mktemp)
+      - |
+        if [ -n "$FIRST_FAILURE" ]; then
+          echo "Skipping Adapter_Test stage due to prior failure in $FIRST_FAILURE"
+          ADAPTER_TEST_STATUS="skip"
+          ADAPTER_TEST_DURATION=0
+        else
+          cd /tmp/ci-project
+          ADAPTER_CONFS=$(find do/adapters -name '*.conf' 2>/dev/null | grep -v '.gitkeep' || true)
+          if [ -z "$ADAPTER_CONFS" ]; then
+            echo "No adapter configs found in do/adapters/ — skipping"
+            ADAPTER_TEST_STATUS="skip"
+            ADAPTER_TEST_DURATION=0
+          else
+            (
+              set -e
+              cd /tmp/ci-project
+              for conf in do/adapters/*.conf; do
+                [ -f "$conf" ] || continue
+                [[ "$(basename "$conf")" == ".gitkeep" ]] && continue
+                ADAPTER_NAME=$(basename "$conf" .conf)
+                echo "Testing adapter: ${ADAPTER_NAME}"
+                # Source to get weights URI
+                source "$conf"
+                ./do/adapter add "${ADAPTER_NAME}" --weights "${ADAPTER_WEIGHTS_URI}"
+                ./do/test --ic "${ADAPTER_NAME}"
+                ./do/adapter remove "${ADAPTER_NAME}"
+              done
+            ) 2>"$STAGE_STDERR_FILE"; STAGE_EXIT=$?
+            STAGE_END=$(date +%s)
+            ADAPTER_TEST_DURATION=$((STAGE_END - STAGE_START))
+            if [ "$STAGE_EXIT" -eq 0 ]; then
+              ADAPTER_TEST_STATUS="pass"
+              echo "Adapter_Test stage passed in ${ADAPTER_TEST_DURATION}s"
+            else
+              ADAPTER_TEST_STATUS="fail"
+              ADAPTER_TEST_ERROR_SUMMARY=$(tail -c 500 "$STAGE_STDERR_FILE" | tr -d '\000' | tr '"' "'" | tr '\n' ' ')
+              FIRST_FAILURE="adapter_test"
+              echo "Adapter_Test stage FAILED (exit code $STAGE_EXIT) in ${ADAPTER_TEST_DURATION}s"
+            fi
+          fi
+        fi
+      - rm -f "$STAGE_STDERR_FILE"
+
       # --- Stage: Register (placeholder) ---
       - echo "=== Stage: Register ==="
       - STAGE_START=$(date +%s)
@@ -260,6 +312,7 @@ phases:
             validate)    FINAL_ERROR_MESSAGE="$VALIDATE_ERROR_SUMMARY" ;;
             build)       FINAL_ERROR_MESSAGE="$BUILD_ERROR_SUMMARY" ;;
             deploy_test) FINAL_ERROR_MESSAGE="$DEPLOY_TEST_ERROR_SUMMARY" ;;
+            adapter_test) FINAL_ERROR_MESSAGE="$ADAPTER_TEST_ERROR_SUMMARY" ;;
             register)    FINAL_ERROR_MESSAGE="$REGISTER_ERROR_SUMMARY" ;;
             *)           FINAL_ERROR_MESSAGE="Unknown failure stage" ;;
           esac
@@ -272,6 +325,7 @@ phases:
         ESCAPED_VALIDATE_ERROR=$(printf '%s' "$VALIDATE_ERROR_SUMMARY" | sed 's/\\/\\\\/g; s/"/\\"/g')
         ESCAPED_BUILD_ERROR=$(printf '%s' "$BUILD_ERROR_SUMMARY" | sed 's/\\/\\\\/g; s/"/\\"/g')
         ESCAPED_DEPLOY_TEST_ERROR=$(printf '%s' "$DEPLOY_TEST_ERROR_SUMMARY" | sed 's/\\/\\\\/g; s/"/\\"/g')
+        ESCAPED_ADAPTER_TEST_ERROR=$(printf '%s' "$ADAPTER_TEST_ERROR_SUMMARY" | sed 's/\\/\\\\/g; s/"/\\"/g')
         ESCAPED_REGISTER_ERROR=$(printf '%s' "$REGISTER_ERROR_SUMMARY" | sed 's/\\/\\\\/g; s/"/\\"/g')
         ESCAPED_TEARDOWN_ERROR=$(printf '%s' "$TEARDOWN_ERROR_SUMMARY" | sed 's/\\/\\\\/g; s/"/\\"/g')
         ESCAPED_FINAL_ERROR=$(printf '%s' "$FINAL_ERROR_MESSAGE" | sed 's/\\/\\\\/g; s/"/\\"/g')
@@ -314,6 +368,12 @@ phases:
                   \"logPointer\": {\"S\": \"$DEPLOY_TEST_LOG_POINTER\"},
                   \"errorSummary\": {\"S\": \"$ESCAPED_DEPLOY_TEST_ERROR\"}
                 }},
+                \"adapter_test\": {\"M\": {
+                  \"status\": {\"S\": \"$ADAPTER_TEST_STATUS\"},
+                  \"durationSeconds\": {\"N\": \"$ADAPTER_TEST_DURATION\"},
+                  \"logPointer\": {\"S\": \"$ADAPTER_TEST_LOG_POINTER\"},
+                  \"errorSummary\": {\"S\": \"$ESCAPED_ADAPTER_TEST_ERROR\"}
+                }},
                 \"register\": {\"M\": {
                   \"status\": {\"S\": \"$REGISTER_STATUS\"},
                   \"durationSeconds\": {\"N\": \"$REGISTER_DURATION\"},

package/infra/ci-harness/package-lock.json

@@ -48,6 +48,7 @@
         "semver"
       ],
       "license": "Apache-2.0",
+      "peer": true,
       "dependencies": {
         "jsonschema": "~1.4.1",
         "semver": "^7.7.4"
@@ -2150,6 +2151,7 @@
       "integrity": "sha512-wGdMcf+vPYM6jikpS/qhg6WiqSV/OhG+jeeHT/KlVqxYfD40iYJf9/AE1uQxVWFvU7MipKRkRv8NSHiCGgPr8Q==",
       "dev": true,
       "license": "MIT",
+      "peer": true,
       "dependencies": {
         "undici-types": "~6.21.0"
       }
@@ -2789,7 +2791,8 @@
       "version": "10.6.0",
       "resolved": "https://registry.npmjs.org/constructs/-/constructs-10.6.0.tgz",
       "integrity": "sha512-TxHOnBO5zMo/G76ykzGF/wMpEHu257TbWiIxP9K0Yv/+t70UzgBQiTqjkAsWOPC6jW91DzJI0+ehQV6xDRNBuQ==",
-      "license": "Apache-2.0"
+      "license": "Apache-2.0",
+      "peer": true
     },
     "node_modules/create-require": {
       "version": "1.1.1",
@@ -3694,6 +3697,7 @@
       "integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
       "dev": true,
       "license": "Apache-2.0",
+      "peer": true,
       "bin": {
         "tsc": "bin/tsc",
         "tsserver": "bin/tsserver"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aws/ml-container-creator",
-  "version": "0.4.0",
+  "version": "0.6.0",
   "description": "Generator for SageMaker AI BYOC paradigm for predictive inference use-cases.",
   "type": "module",
   "main": "src/app.js",

package/servers/README.md

@@ -15,7 +15,12 @@ servers/
 │   ├── test.js                 # Standalone tests (node test.js)
 │   ├── package.json
 │   └── LICENSE
-└── region-picker/              # AWS region suggestion server
+├── region-picker/              # AWS region suggestion server
+│   ├── index.js                # MCP server entry point
+│   ├── test.js                 # Standalone tests (node test.js)
+│   ├── package.json
+│   └── LICENSE
+└── endpoint-picker/            # SageMaker endpoint discovery server
     ├── index.js                # MCP server entry point
     ├── test.js                 # Standalone tests (node test.js)
     ├── package.json
@@ -74,6 +79,39 @@ Suggests AWS regions for SageMaker deployments based on a search term. Filters t
 }
 ```
 
+### endpoint-picker
+
+Discovers InService SageMaker real-time endpoints with available GPU capacity for attaching new inference components. Uses `ListEndpoints`, `DescribeEndpoint`, and `ListInferenceComponents` to calculate available capacity.
+
+**Discover mode:** Queries the SageMaker API using a 3-strategy credential fallback (explicit profile → default chain → detect profiles). No static mode — always requires AWS credentials.
+
+**Tool:** `get_inference_endpoints`
+
+| Input Field | Type | Description |
+|-------------|------|-------------|
+| `parameters` | `string[]` | Must include `"endpointName"` to get results |
+| `limit` | `number` | Max endpoints to return (default: 10) |
+| `context` | `object` | `awsRegion`, `awsProfile`, `deploymentTarget` (must be `realtime-inference`) |
+
+**Example response:**
+
+```json
+{
+  "values": { "endpointName": "my-endpoint-1234567890" },
+  "choices": { "endpointName": ["my-endpoint-1234567890", "prod-llm-endpoint"] },
+  "metadata": {
+    "my-endpoint-1234567890": {
+      "variantName": "AllTraffic",
+      "instanceType": "ml.g6e.48xlarge",
+      "instanceCount": 1,
+      "icCount": 2,
+      "availableGpus": 4,
+      "hasInstancePools": false
+    }
+  }
+}
+```
+
 ## Usage
 
 ### Adding a Bundled Server
@@ -297,6 +335,7 @@ The Bedrock API didn't respond within 10 seconds. This usually means network con
 ```bash
 node servers/region-picker/test.js
 node servers/instance-recommender/test.js
+node servers/endpoint-picker/test.js
 ```
 
 ### Smart Mode Not Activating
@@ -313,6 +352,7 @@ Each server has standalone tests that run without AWS credentials or network acc
 # Run individual server tests
 node servers/region-picker/test.js
 node servers/instance-recommender/test.js
+node servers/endpoint-picker/test.js
 
 # Run all server tests from the project root
 npm run test:servers

package/servers/instance-sizer/index.js

@@ -51,7 +51,7 @@ try {
 
 // ── Mode configuration ───────────────────────────────────────────────────────
 
-const DISCOVER_MODE = process.argv.includes('--discover') || process.env.DISCOVER_MODE === 'true'
+const DISCOVER_MODE = process.env.DISCOVER_MODE !== 'false' && !process.argv.includes('--no-discover')
 const SMART_MODE = process.env.BEDROCK_SMART === 'true'
 const BEDROCK_MODEL = process.env.BEDROCK_MODEL || 'global.anthropic.claude-sonnet-4-20250514-v1:0'
 const BEDROCK_REGION = process.env.BEDROCK_REGION || process.env.AWS_REGION || 'us-east-1'
@@ -383,6 +383,7 @@ async function handleGetInstanceRecommendation(params) {
     // Step 3a: Quota & availability filtering (discover mode only)
     let preQuotaFilterCount = 0
     let allFilteredByQuota = false
+    let preQuotaRecommendations = []
     if (DISCOVER_MODE && recommendations.length > 0) {
         try {
             const region = process.env.AWS_REGION || process.env.AWS_DEFAULT_REGION || BEDROCK_REGION
@@ -396,6 +397,7 @@ async function handleGetInstanceRecommendation(params) {
             ])
 
             preQuotaFilterCount = recommendations.length
+            preQuotaRecommendations = [...recommendations]
             recommendations = applyAvailabilityRanking(
                 recommendations,
                 quotas.status === 'fulfilled' ? quotas.value : null,
@@ -404,6 +406,10 @@ async function handleGetInstanceRecommendation(params) {
             )
             if (recommendations.length === 0 && preQuotaFilterCount > 0) {
                 allFilteredByQuota = true
+                // Restore pre-filter recommendations so user can see compatible instances
+                // and request quota increases for the ones they want
+                recommendations = preQuotaRecommendations
+                log(`All ${preQuotaFilterCount} instances filtered by zero-quota — restoring unfiltered list`)
             }
         } catch (err) {
             // Graceful degradation: if credentials are missing or any unexpected
@@ -587,10 +593,10 @@ const isMain = process.argv[1] && resolve(process.argv[1]) === __filename
 if (isMain) {
     if (SMART_MODE) {
         log(`Smart mode enabled (model: ${BEDROCK_MODEL}, region: ${BEDROCK_REGION})`)
-    } else if (DISCOVER_MODE) {
-        log('Discover mode enabled (HuggingFace API lookups active)')
+    } else if (!DISCOVER_MODE) {
+        log('Static mode (catalog-only, no network calls) — use --no-discover to force this')
     } else {
-        log('Static mode (catalog-only, no network calls)')
+        log('Discover mode (HuggingFace API + quota lookups active)')
     }
 
     const transport = new StdioServerTransport()

package/servers/instance-sizer/lib/model-resolver.js

@@ -207,7 +207,7 @@ const isHuggingFacePattern = (modelName) => {
  * @returns {Promise<{ parameterCount: number, dtype: string, architecture: string, maxPositionEmbeddings: number, source: string } | null>}
  */
 const resolveModelMetadata = async (modelName, options = {}) => {
-    const { discover = false, catalogPath } = options
+    const { discover = true, catalogPath } = options
 
     // Tier 1: Catalog lookup
     const catalog = await loadCatalog(catalogPath)