@aws/ml-container-creator 1.0.3 → 1.1.0

package/README.md

@@ -74,6 +74,14 @@ ml-container-creator my-model \
 ./do/test         # Test the endpoint
 ```
 
+### Get help from the advisor
+
+```bash
+ml-container-creator hey   # Conversational AI advisor (powered by Bedrock)
+```
+
+Ask questions about your project, get optimization recommendations, troubleshoot issues, and plan workflows. See [Agent docs](https://awslabs.github.io/ml-container-creator/agent/) for details.
+
 ## Documentation
 
 Full documentation is available at [awslabs.github.io/ml-container-creator](https://awslabs.github.io/ml-container-creator/).
@@ -83,6 +91,7 @@ Full documentation is available at [awslabs.github.io/ml-container-creator](http
 - [Deployment Guide](https://awslabs.github.io/ml-container-creator/deployments/) — All deployment targets and lifecycle scripts
 - [CI Integration](https://awslabs.github.io/ml-container-creator/ci-integration/) — Automated lifecycle testing for all deployment configurations
 - [Examples](https://awslabs.github.io/ml-container-creator/EXAMPLES/) — Framework-specific walkthroughs
+- [Advisory Agent](https://awslabs.github.io/ml-container-creator/agent/) — Conversational AI advisor (`ml-container-creator hey`)
 - [Troubleshooting](https://awslabs.github.io/ml-container-creator/TROUBLESHOOTING/) — Common issues and solutions
 
 ## Prerequisites
@@ -97,7 +106,7 @@ Full documentation is available at [awslabs.github.io/ml-container-creator](http
 
 ### Python dependencies
 
-The `do/` lifecycle scripts (`do/tune`, `do/stage`, `do/adapter`) require Python packages. Install them in your Python environment before first use:
+The `do/` lifecycle scripts (`do/tune`, `do/train`, `do/stage`, `do/adapter`) require Python packages. Install them in your Python environment before first use:
 
 ```bash
 # Recommended (fast):

package/bin/cli.js

@@ -4,10 +4,15 @@
 
 import { createRequire } from 'module';
 import path from 'path';
+import { fileURLToPath } from 'url';
+import { spawn, execSync } from 'child_process';
 import { program, Option, Help } from 'commander';
 import { run } from '../src/app.js';
 import { cliOptions, helpGroups } from '../src/lib/generated/cli-options.js';
 
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
 const require = createRequire(import.meta.url);
 const { version } = require('../package.json');
 
@@ -314,4 +319,56 @@ program
         }
     });
 
+program
+    .command('hey')
+    .description('Chat with the ml-container-creator advisor')
+    .option('--project-dir <dir>', 'Project directory to analyze', process.cwd())
+    .option('-o, --offline', 'Static reference mode (no Bedrock calls)')
+    .action(async (options) => {
+        // 1. Check python3 is available
+        try {
+            execSync('python3 --version', { stdio: 'ignore' });
+        } catch {
+            console.error('❌ python3 not found. Install Python 3.10+ to use the advisor.');
+            console.error('   macOS: brew install python3');
+            console.error('   Ubuntu: sudo apt install python3');
+            process.exit(1);
+        }
+
+        // 2. If not offline, check strands-agents is installed
+        if (!options.offline) {
+            try {
+                execSync('python3 -c "import strands"', { stdio: 'ignore' });
+            } catch {
+                console.error('❌ strands-agents not installed. Run:');
+                console.error('   pip install -r src/agent/requirements-agent.txt');
+                process.exit(1);
+            }
+        }
+
+        // 3. Resolve agent script path
+        const agentScript = path.join(__dirname, '..', 'src', 'agent', 'agent.py');
+
+        // 4. Build args and spawn
+        const args = [agentScript, '--project-dir', options.projectDir];
+        if (options.offline) {
+            args.push('--offline');
+        }
+
+        const child = spawn('python3', args, {
+            stdio: 'inherit',
+            env: { ...process.env, PYTHONUNBUFFERED: '1' }
+        });
+
+        // 5. Forward exit code
+        child.on('close', (code) => {
+            process.exit(code ?? 0);
+        });
+
+        child.on('error', (err) => {
+            console.error(`❌ Failed to start agent: ${err.message}`);
+            process.exit(1);
+        });
+    });
+
 program.parse();

package/config/agent.json

@@ -0,0 +1,16 @@
+{
+  "modelId": "global.anthropic.claude-sonnet-4-5-20250929-v1:0",
+  "mcpServers": [
+    "instance-sizer",
+    "base-image-picker",
+    "model-picker",
+    "workload-picker",
+    "e2e-status",
+    "agent-knowledge"
+  ],
+  "inputCostPer1k": 0.003,
+  "outputCostPer1k": 0.015,
+  "exitCommands": ["exit", "quit", "bye", "q"],
+  "reloadCommands": ["reload"],
+  "mcpServerTimeout": 30
+}

package/infra/ci-harness/lib/ci-harness-stack.ts

@@ -1057,6 +1057,49 @@ export class MlccCiHarnessStack extends cdk.Stack {
         glueTable.addDependency(glueDatabase);
         glueTable.cfnOptions.condition = benchmarkInfraCondition;
 
+        // Glue Table: mlcc_evaluations — model quality evaluation results
+        // Written by do/evaluate via .eval_helper.py eval-write subcommand.
+        // Partitioned by model + adapter for efficient comparison queries.
+        const evalGlueTable = new glue.CfnTable(this, 'EvaluationResultsTable', {
+            catalogId: this.account,
+            databaseName: 'mlcc_ci',
+            tableInput: {
+                name: 'mlcc_evaluations',
+                tableType: 'EXTERNAL_TABLE',
+                parameters: {
+                    'classification': 'json',
+                },
+                storageDescriptor: {
+                    columns: [
+                        { name: 'project_name', type: 'string', comment: 'MCC project name' },
+                        { name: 'model_name', type: 'string', comment: 'HuggingFace model ID' },
+                        { name: 'adapter_name', type: 'string', comment: 'Adapter name or IC name' },
+                        { name: 'technique', type: 'string', comment: 'Training technique (sft, dpo)' },
+                        { name: 'eval_dataset', type: 'string', comment: 'Evaluation dataset URI or name' },
+                        { name: 'samples_evaluated', type: 'int', comment: 'Number of samples evaluated' },
+                        { name: 'metrics', type: 'string', comment: 'JSON blob of all computed metrics' },
+                        { name: 'timestamp', type: 'string', comment: 'ISO 8601 UTC timestamp' },
+                        { name: 'region', type: 'string', comment: 'AWS region' },
+                    ],
+                    location: `s3://mlcc-benchmark-results-${this.account}-${this.region}/evaluations/`,
+                    inputFormat: 'org.apache.hadoop.mapred.TextInputFormat',
+                    outputFormat: 'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat',
+                    serdeInfo: {
+                        serializationLibrary: 'org.openx.data.jsonserde.JsonSerDe',
+                        parameters: {
+                            'serialization.format': '1',
+                        },
+                    },
+                },
+                partitionKeys: [
+                    { name: 'model', type: 'string', comment: 'Model name (partition key)' },
+                    { name: 'adapter', type: 'string', comment: 'Adapter name (partition key)' },
+                ],
+            },
+        });
+        evalGlueTable.addDependency(glueDatabase);
+        evalGlueTable.cfnOptions.condition = benchmarkInfraCondition;
+
         // Configurable lifecycle parameters for the benchmark results bucket
         const benchmarkIaTransitionDays = new cdk.CfnParameter(this, 'BenchmarkIaTransitionDays', {
             type: 'Number',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aws/ml-container-creator",
-  "version": "1.0.3",
+  "version": "1.1.0",
   "description": "Build and deploy custom ML containers on AWS SageMaker with minimal configuration.",
   "main": "src/index.js",
   "bin": {
@@ -51,6 +51,8 @@
     "servers/workload-picker/index.js",
     "servers/workload-picker/manifest.json",
     "servers/workload-picker/package.json",
+    "servers/agent-knowledge/index.js",
+    "servers/agent-knowledge/package.json",
     "servers/lib/bedrock-client.js",
     "servers/lib/custom-validators.js",
     "servers/lib/dynamic-resolver.js",
@@ -61,6 +63,7 @@
     "config/bootstrap-stack.json",
     "config/bootstrap-e2e-stack.json",
     "config/parameter-schema-v2.json",
+    "config/agent.json",
     "config/tune-catalog.json",
     "config/presets/",
     "infra/ci-harness/bin/",
@@ -88,7 +91,7 @@
   },
   "scripts": {
     "test": "mocha 'test/**/*.test.js' --ignore 'test/property/**' --recursive --timeout 30000 --parallel",
-    "test:property": "mocha 'test/property/**/*.test.js' --recursive --timeout 60000 --parallel",
+    "test:property": "NODE_OPTIONS='--max-old-space-size=8192' mocha 'test/property/**/*.test.js' --recursive --timeout 60000 --parallel --jobs 4",
     "test:all": "npm run test && npm run test:property",
     "test:fast": "mocha 'test/**/*.test.js' --recursive --timeout 15000 --parallel",
     "test:unit": "mocha 'test/unit/**/*.test.js' --recursive --timeout 15000",

package/pyproject.toml

@@ -15,6 +15,9 @@ dependencies = [
     "pyyaml>=6.0",
 ]
 
+[tool.pytest.ini_options]
+addopts = "--import-mode=importlib"
+
 [dependency-groups]
 dev = [
     "pytest>=8.0",