fix: batch-transform deploy/test/clean UX improvements

Author: frgud

generators/app/index.js

@@ -204,6 +204,47 @@ export default class extends Generator {
             type: Number,
             description: 'Max concurrent invocations per instance for async inference (default: 1)'
         });
+
+        // Batch transform options
+        this.option('batch-input-path', {
+            type: String,
+            description: 'S3 input path for batch transform data'
+        });
+
+        this.option('batch-output-path', {
+            type: String,
+            description: 'S3 output path for batch transform results'
+        });
+
+        this.option('batch-instance-count', {
+            type: Number,
+            description: 'Number of instances for batch transform job (default: 1)'
+        });
+
+        this.option('batch-split-type', {
+            type: String,
+            description: 'Input data split type: Line, RecordIO, None (default: Line)'
+        });
+
+        this.option('batch-strategy', {
+            type: String,
+            description: 'Batch strategy: MultiRecord, SingleRecord (default: MultiRecord)'
+        });
+
+        this.option('batch-join-source', {
+            type: String,
+            description: 'Join source: Input, None (default: None)'
+        });
+
+        this.option('batch-max-concurrent', {
+            type: Number,
+            description: 'Max concurrent transforms per instance (default: 1)'
+        });
+
+        this.option('batch-max-payload', {
+            type: Number,
+            description: 'Max payload size in MB, 0-100 (default: 6)'
+        });
     }
 
     /**
@@ -447,6 +488,12 @@ export default class extends Generator {
             architecture = this.answers.framework === 'transformers' ? 'transformers' : 'http';
         }
 
+        // Exclude sample_model directory when not needed
+        // Transformers and diffusors don't use sample models (they load from HuggingFace Hub)
+        if (!this.answers.includeSampleModel || architecture === 'transformers' || architecture === 'diffusors') {
+            ignorePatterns.push('**/sample_model/**');
+        }
+
         // Always exclude triton and diffusors source directories from initial copy (they are sources, not output)
         ignorePatterns.push('**/triton/**');
         ignorePatterns.push('**/diffusors/**');

generators/app/lib/config-manager.js

@@ -600,6 +600,94 @@ export default class ConfigManager {
                 required: false,
                 default: 1,
                 valueSpace: 'bounded'
+            },
+            batchInputPath: {
+                cliOption: 'batch-input-path',
+                envVar: 'ML_BATCH_INPUT_PATH',
+                configFile: true,
+                packageJson: false,
+                mcp: true,
+                promptable: true,
+                required: false,
+                default: null,
+                valueSpace: 'unbounded'
+            },
+            batchOutputPath: {
+                cliOption: 'batch-output-path',
+                envVar: 'ML_BATCH_OUTPUT_PATH',
+                configFile: true,
+                packageJson: false,
+                mcp: true,
+                promptable: true,
+                required: false,
+                default: null,
+                valueSpace: 'unbounded'
+            },
+            batchInstanceCount: {
+                cliOption: 'batch-instance-count',
+                envVar: null,
+                configFile: true,
+                packageJson: false,
+                mcp: false,
+                promptable: true,
+                required: false,
+                default: 1,
+                valueSpace: 'bounded'
+            },
+            batchSplitType: {
+                cliOption: 'batch-split-type',
+                envVar: null,
+                configFile: true,
+                packageJson: false,
+                mcp: false,
+                promptable: true,
+                required: false,
+                default: 'Line',
+                valueSpace: 'bounded'
+            },
+            batchStrategy: {
+                cliOption: 'batch-strategy',
+                envVar: null,
+                configFile: true,
+                packageJson: false,
+                mcp: false,
+                promptable: true,
+                required: false,
+                default: 'MultiRecord',
+                valueSpace: 'bounded'
+            },
+            batchJoinSource: {
+                cliOption: 'batch-join-source',
+                envVar: null,
+                configFile: true,
+                packageJson: false,
+                mcp: false,
+                promptable: true,
+                required: false,
+                default: 'None',
+                valueSpace: 'bounded'
+            },
+            batchMaxConcurrentTransforms: {
+                cliOption: 'batch-max-concurrent',
+                envVar: null,
+                configFile: true,
+                packageJson: false,
+                mcp: false,
+                promptable: true,
+                required: false,
+                default: 1,
+                valueSpace: 'bounded'
+            },
+            batchMaxPayloadInMB: {
+                cliOption: 'batch-max-payload',
+                envVar: null,
+                configFile: true,
+                packageJson: false,
+                mcp: false,
+                promptable: true,
+                required: false,
+                default: 6,
+                valueSpace: 'bounded'
             }
         };
     }

generators/app/lib/prompt-runner.js

@@ -22,6 +22,7 @@ import {
     infraRegionAndTargetPrompts,
     infraInstancePrompts,
     infraAsyncPrompts,
+    infraBatchTransformPrompts,
     infraHyperPodPrompts,
     infraBuildPrompts,
     projectPrompts,
@@ -69,10 +70,11 @@ export default class PromptRunner {
         await this._queryMcpForRegion({}, explicitConfig);
         const regionAndTargetAnswers = await this._runPhase(infraRegionAndTargetPrompts, {}, explicitConfig, existingConfig);
 
-        // 1b. Instance type — query MCP and prompt for managed-inference, async-inference, and hyperpod-eks
+        // 1b. Instance type — query MCP and prompt for managed-inference, async-inference, batch-transform, and hyperpod-eks
         let instanceAnswers = {};
         if (regionAndTargetAnswers.deploymentTarget === 'managed-inference' ||
             regionAndTargetAnswers.deploymentTarget === 'async-inference' ||
+            regionAndTargetAnswers.deploymentTarget === 'batch-transform' ||
             regionAndTargetAnswers.deploymentTarget === 'hyperpod-eks') {
             await this._queryMcpForInstance({}, explicitConfig);
             const mcpInstanceChoices = this.configManager?.mcpChoices?.instanceType;
@@ -89,6 +91,17 @@ export default class PromptRunner {
             asyncAnswers = await this._runPhase(infraAsyncPrompts, { ...regionAndTargetAnswers }, explicitConfig, existingConfig);
         }
 
+        // 1b-batch. Batch transform-specific prompts (only when deploymentTarget === 'batch-transform')
+        let batchTransformAnswers = {};
+        if (regionAndTargetAnswers.deploymentTarget === 'batch-transform') {
+            batchTransformAnswers = await this._runPhase(
+                infraBatchTransformPrompts,
+                { ...regionAndTargetAnswers },
+                explicitConfig,
+                existingConfig
+            );
+        }
+
         // 1c. HyperPod prompts — only query MCP and prompt when deployment target is hyperpod-eks
         let hyperPodAnswers = {};
         if (regionAndTargetAnswers.deploymentTarget === 'hyperpod-eks') {
@@ -106,6 +119,7 @@ export default class PromptRunner {
             ...regionAndTargetAnswers,
             ...instanceAnswers,
             ...asyncAnswers,
+            ...batchTransformAnswers,
             ...hyperPodAnswers,
             ...buildAnswers
         };

generators/app/lib/prompts.js

@@ -611,6 +611,7 @@ const infraRegionAndTargetPrompts = [
         choices: [
             { name: 'SageMaker Managed Inference - Real Time', value: 'managed-inference' },
             { name: 'SageMaker Managed Inference - Async', value: 'async-inference' },
+            { name: 'SageMaker Managed Inference - Batch', value: 'batch-transform' },
             { name: 'SageMaker HyperPod - EKS', value: 'hyperpod-eks' }
         ],
         default: 'managed-inference'
@@ -622,7 +623,7 @@ const infraInstancePrompts = [
     {
         type: 'list',
         name: 'instanceType',
-        when: answers => answers.deploymentTarget === 'managed-inference' || answers.deploymentTarget === 'async-inference' || answers.deploymentTarget === 'hyperpod-eks',
+        when: answers => answers.deploymentTarget === 'managed-inference' || answers.deploymentTarget === 'async-inference' || answers.deploymentTarget === 'batch-transform' || answers.deploymentTarget === 'hyperpod-eks',
         message: (answers) => {
             const framework = answers.framework || answers.deploymentConfig?.split('-')[0];
 
@@ -854,6 +855,80 @@ const infraAsyncPrompts = [
     }
 ];
 
+/**
+ * Sub-phase: Batch transform-specific prompts (only when deploymentTarget === 'batch-transform')
+ * Requirements: 2.1, 2.2, 2.4, 2.5, 2.6, 2.7, 2.8, 2.9
+ */
+const infraBatchTransformPrompts = [
+    {
+        type: 'input',
+        name: 'batchInputPath',
+        message: 'S3 input path for batch transform data (leave empty for default: s3://ml-container-creator-batch-{region}-{account-id}/{project-name}/input/):',
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'input',
+        name: 'batchOutputPath',
+        message: 'S3 output path for batch transform results (leave empty for default: s3://ml-container-creator-batch-{region}-{account-id}/{project-name}/output/):',
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'number',
+        name: 'batchInstanceCount',
+        message: 'How many instances should run the batch job in parallel?',
+        default: 1,
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'list',
+        name: 'batchSplitType',
+        message: 'Input file format — how should SageMaker read your input files?',
+        choices: [
+            { name: 'Line — one record per line (JSON lines, CSV)', value: 'Line' },
+            { name: 'RecordIO — Amazon RecordIO format', value: 'RecordIO' },
+            { name: 'None — send each file as a single request', value: 'None' }
+        ],
+        default: 'Line',
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'list',
+        name: 'batchStrategy',
+        message: 'How many records should be sent per inference request?',
+        choices: [
+            { name: 'MultiRecord — batch multiple records per request (higher throughput)', value: 'MultiRecord' },
+            { name: 'SingleRecord — one record per request (simpler, more predictable)', value: 'SingleRecord' }
+        ],
+        default: 'MultiRecord',
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'list',
+        name: 'batchJoinSource',
+        message: 'Include original input data alongside predictions in the output?',
+        choices: [
+            { name: 'No — output predictions only', value: 'None' },
+            { name: 'Yes — merge input with predictions (useful for traceability)', value: 'Input' }
+        ],
+        default: 'None',
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'number',
+        name: 'batchMaxConcurrentTransforms',
+        message: 'Max concurrent inference requests per instance?',
+        default: 1,
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    },
+    {
+        type: 'number',
+        name: 'batchMaxPayloadInMB',
+        message: 'Max request payload size in MB (0-100)?',
+        default: 6,
+        when: answers => answers.deploymentTarget === 'batch-transform'
+    }
+];
+
 // Combined view for tests and backward compatibility
 const infrastructurePrompts = [
     ...infraRegionAndTargetPrompts,
@@ -977,6 +1052,7 @@ export {
     infraRegionAndTargetPrompts,
     infraInstancePrompts,
     infraAsyncPrompts,
+    infraBatchTransformPrompts,
     infraHyperPodPrompts,
     infraBuildPrompts,
     projectPrompts,

generators/app/lib/template-manager.js

@@ -64,7 +64,7 @@ export default class TemplateManager {
                 'diffusors-vllm-omni'
             ],
             buildTargets: ['codebuild'],
-            deploymentTargets: ['managed-inference', 'async-inference', 'hyperpod-eks'],
+            deploymentTargets: ['managed-inference', 'async-inference', 'batch-transform', 'hyperpod-eks'],
             testTypes: ['local-model-cli', 'local-model-server', 'hosted-model-endpoint'],
             awsRegions: [
                 'us-east-1', 'us-east-2', 'us-west-1', 'us-west-2',
@@ -131,6 +131,9 @@ export default class TemplateManager {
 
         // Validate async inference specific fields
         this._validateAsyncConfig()
+
+        // Validate batch transform specific fields
+        this._validateBatchTransformConfig()
 
         // Validate instance type format (ml.*.*) - only for managed-inference
         if (this.answers.instanceType && this.answers.instanceType !== 'custom') {
@@ -229,6 +232,71 @@ export default class TemplateManager {
         }
     }
 
+    /**
+     * Validates batch transform specific configuration
+     * @private
+     * @throws {Error} If batch transform configuration is invalid
+     */
+    _validateBatchTransformConfig() {
+        if (this.answers.deploymentTarget !== 'batch-transform') return
+
+        // Validate S3 input path format if provided
+        if (this.answers.batchInputPath && this.answers.batchInputPath.trim() !== '') {
+            if (!this.answers.batchInputPath.startsWith('s3://')) {
+                throw new Error('⚠️  batchInputPath must start with "s3://". Example: s3://my-bucket/input/')
+            }
+        }
+
+        // Validate S3 output path format if provided
+        if (this.answers.batchOutputPath && this.answers.batchOutputPath.trim() !== '') {
+            if (!this.answers.batchOutputPath.startsWith('s3://')) {
+                throw new Error('⚠️  batchOutputPath must start with "s3://". Example: s3://my-bucket/output/')
+            }
+        }
+
+        // Validate instance count
+        if (this.answers.batchInstanceCount !== undefined) {
+            const val = this.answers.batchInstanceCount
+            if (!Number.isInteger(val) || val < 1) {
+                throw new Error('⚠️  batchInstanceCount must be an integer >= 1')
+            }
+        }
+
+        // Validate split type
+        const validSplitTypes = ['Line', 'RecordIO', 'None']
+        if (this.answers.batchSplitType && !validSplitTypes.includes(this.answers.batchSplitType)) {
+            throw new Error(`⚠️  batchSplitType must be one of: ${validSplitTypes.join(', ')}`)
+        }
+
+        // Validate batch strategy
+        const validStrategies = ['MultiRecord', 'SingleRecord']
+        if (this.answers.batchStrategy && !validStrategies.includes(this.answers.batchStrategy)) {
+            throw new Error(`⚠️  batchStrategy must be one of: ${validStrategies.join(', ')}`)
+        }
+
+        // Validate join source
+        const validJoinSources = ['Input', 'None']
+        if (this.answers.batchJoinSource && !validJoinSources.includes(this.answers.batchJoinSource)) {
+            throw new Error(`⚠️  batchJoinSource must be one of: ${validJoinSources.join(', ')}`)
+        }
+
+        // Validate max concurrent transforms
+        if (this.answers.batchMaxConcurrentTransforms !== undefined) {
+            const val = this.answers.batchMaxConcurrentTransforms
+            if (!Number.isInteger(val) || val < 0) {
+                throw new Error('⚠️  batchMaxConcurrentTransforms must be an integer >= 0')
+            }
+        }
+
+        // Validate max payload in MB
+        if (this.answers.batchMaxPayloadInMB !== undefined) {
+            const val = this.answers.batchMaxPayloadInMB
+            if (!Number.isInteger(val) || val < 0 || val > 100) {
+                throw new Error('⚠️  batchMaxPayloadInMB must be an integer between 0 and 100')
+            }
+        }
+    }
+
     /**
      * Validates GPU instance type requirement for GPU-requiring backends.
      * Called when deploymentConfig is present.