@aws/ml-container-creator 0.7.1 → 0.9.0

package/servers/lib/catalogs/model-servers.json

@@ -1,5 +1,85 @@
 {
     "vllm": [
+        {
+            "image": "vllm/vllm-openai:v0.20.2",
+            "tag": "v0.20.2",
+            "architecture": "amd64",
+            "created": "2026-05-10T00:00:00Z",
+            "labels": {
+                "cuda_version": "12.9",
+                "python_version": "3.12",
+                "framework_version": "0.20.2"
+            },
+            "registry": "dockerhub",
+            "repository": "vllm/vllm-openai",
+            "defaults": {
+                "envVars": {
+                    "VLLM_TENSOR_PARALLEL_SIZE": "1",
+                    "VLLM_GPU_MEMORY_UTILIZATION": "0.9",
+                    "VLLM_MAX_NUM_SEQS": "256",
+                    "VLLM_MAX_MODEL_LEN": "4096",
+                    "VLLM_ENABLE_PREFIX_CACHING": "true"
+                },
+                "inferenceAmiVersion": "al2-ami-sagemaker-inference-gpu-3-1"
+            },
+            "accelerator": {
+                "type": "cuda",
+                "version": "12.9",
+                "versionRange": {
+                    "min": "12.4",
+                    "max": "12.9"
+                }
+            },
+            "validationLevel": "community-validated",
+            "profiles": {
+                "low-latency": {
+                    "displayName": "Low Latency",
+                    "description": "Optimized for single-request latency with prefix caching",
+                    "envVars": {
+                        "VLLM_MAX_NUM_SEQS": "32",
+                        "VLLM_GPU_MEMORY_UTILIZATION": "0.85",
+                        "VLLM_ENABLE_PREFIX_CACHING": "true"
+                    },
+                    "notes": "Prefix caching improves latency for repeated prompts"
+                },
+                "high-throughput": {
+                    "displayName": "High Throughput",
+                    "description": "Optimized for batch processing with continuous batching",
+                    "envVars": {
+                        "VLLM_MAX_NUM_SEQS": "512",
+                        "VLLM_GPU_MEMORY_UTILIZATION": "0.95",
+                        "VLLM_MAX_MODEL_LEN": "2048",
+                        "VLLM_ENABLE_PREFIX_CACHING": "false"
+                    },
+                    "notes": "Continuous batching maximizes GPU utilization"
+                },
+                "multi-gpu": {
+                    "displayName": "Multi-GPU",
+                    "description": "Tensor parallel across multiple GPUs for large models",
+                    "envVars": {
+                        "VLLM_TENSOR_PARALLEL_SIZE": "4",
+                        "VLLM_GPU_MEMORY_UTILIZATION": "0.9",
+                        "VLLM_MAX_NUM_SEQS": "256"
+                    },
+                    "notes": "Requires instance with 4+ GPUs. Set TENSOR_PARALLEL_SIZE to match GPU count"
+                }
+            },
+            "notes": "vLLM 0.20.2 adds Gemma 4 support, CUDA 12.9, improved multi-GPU. Requires CUDA compat on drivers < 570.",
+            "supportedModelTypes": [
+                "gemma",
+                "gemma2",
+                "gemma3",
+                "llama",
+                "mistral",
+                "mixtral",
+                "qwen2",
+                "qwen3",
+                "qwen3_moe",
+                "deepseek_v3",
+                "phi3",
+                "command-r"
+            ]
+        },
         {
             "image": "vllm/vllm-openai:v0.10.1",
             "tag": "v0.10.1",

package/servers/lib/catalogs/models.json

@@ -555,98 +555,6 @@
             "text-generation"
         ]
     },
-    "stabilityai/stable-diffusion-3.5-medium": {
-        "family": "stable-diffusion-3",
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "stable-diffusion"
-        ],
-        "architecture": "StableDiffusion3Pipeline",
-        "profiles": {
-            "default": {
-                "displayName": "SD3.5 Medium",
-                "envVars": {}
-            }
-        },
-        "notes": "Stable Diffusion 3.5 medium model. Supported natively by vLLM-Omni StableDiffusion3Pipeline.",
-        "chatTemplate": null,
-        "frameworkCompatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validationLevel": "experimental",
-        "modelType": "diffusor",
-        "tasks": [
-            "text-to-image"
-        ]
-    },
-    "black-forest-labs/FLUX.1-dev": {
-        "family": "flux",
-        "gated": true,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "flux"
-        ],
-        "architecture": "FluxPipeline",
-        "profiles": {
-            "default": {
-                "displayName": "FLUX.1 Dev",
-                "envVars": {}
-            }
-        },
-        "notes": "FLUX.1-dev high-quality generation model. Uses dual text encoders (CLIP + T5) and FlowMatchEuler scheduler. Requires significant VRAM.",
-        "chatTemplate": null,
-        "frameworkCompatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validationLevel": "experimental",
-        "modelType": "diffusor",
-        "tasks": [
-            "text-to-image"
-        ]
-    },
-    "black-forest-labs/FLUX.1-schnell": {
-        "family": "flux",
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "flux"
-        ],
-        "architecture": "FluxPipeline",
-        "notes": "FLUX.1-schnell fast generation model. Fewer denoising steps for faster inference at slightly lower quality",
-        "chatTemplate": null,
-        "frameworkCompatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validationLevel": "experimental",
-        "modelType": "diffusor",
-        "tasks": [
-            "text-to-image"
-        ]
-    },
-    "Wan-AI/Wan2.1-T2V-14B-Diffusers": {
-        "family": "wan",
-        "gated": false,
-        "tags": [
-            "video-generation",
-            "diffusion",
-            "wan"
-        ],
-        "architecture": "WanPipeline",
-        "notes": "Wan2.1 text-to-video 14B model (diffusers format). Requires multi-GPU instance (ml.g5.12xlarge or larger). Must use the -Diffusers variant — the base Wan2.1-T2V-14B repo lacks model_index.json required by vLLM-Omni",
-        "chatTemplate": null,
-        "frameworkCompatibility": {
-            "vllm-omni": ">=0.16.0"
-        },
-        "validationLevel": "experimental",
-        "modelType": "diffusor",
-        "tasks": [
-            "text-to-video"
-        ]
-    },
     "meta-llama/Llama-3*": {
         "family": "llama-3",
         "gated": true,
@@ -731,45 +639,5 @@
         "tasks": [
             "text-generation"
         ]
-    },
-    "stabilityai/stable-diffusion-*": {
-        "family": "stable-diffusion",
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "stable-diffusion"
-        ],
-        "architecture": null,
-        "notes": "Fallback for Stable Diffusion variants not explicitly listed",
-        "chatTemplate": null,
-        "frameworkCompatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validationLevel": "experimental",
-        "modelType": "diffusor",
-        "tasks": [
-            "text-to-image"
-        ]
-    },
-    "black-forest-labs/FLUX*": {
-        "family": "flux",
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "flux"
-        ],
-        "architecture": null,
-        "notes": "Fallback for FLUX model variants not explicitly listed",
-        "chatTemplate": null,
-        "frameworkCompatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validationLevel": "experimental",
-        "modelType": "diffusor",
-        "tasks": [
-            "text-to-image"
-        ]
     }
 }

package/servers/lib/catalogs/popular-diffusors.json

@@ -1,110 +1 @@
-{
-    "stabilityai/stable-diffusion-3.5-medium": {
-        "family": "stable-diffusion-3",
-        "chat_template": null,
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "stable-diffusion"
-        ],
-        "architecture": "StableDiffusion3Pipeline",
-        "framework_compatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validation_level": "experimental",
-        "profiles": {
-            "default": {
-                "displayName": "SD3.5 Medium",
-                "envVars": {}
-            }
-        },
-        "notes": "Stable Diffusion 3.5 medium model. Supported natively by vLLM-Omni StableDiffusion3Pipeline."
-    },
-    "black-forest-labs/FLUX.1-dev": {
-        "family": "flux",
-        "chat_template": null,
-        "gated": true,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "flux"
-        ],
-        "architecture": "FluxPipeline",
-        "framework_compatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validation_level": "experimental",
-        "profiles": {
-            "default": {
-                "displayName": "FLUX.1 Dev",
-                "envVars": {}
-            }
-        },
-        "notes": "FLUX.1-dev high-quality generation model. Uses dual text encoders (CLIP + T5) and FlowMatchEuler scheduler. Requires significant VRAM."
-    },
-    "black-forest-labs/FLUX.1-schnell": {
-        "family": "flux",
-        "chat_template": null,
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "flux"
-        ],
-        "architecture": "FluxPipeline",
-        "framework_compatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validation_level": "experimental",
-        "notes": "FLUX.1-schnell fast generation model. Fewer denoising steps for faster inference at slightly lower quality"
-    },
-    "Wan-AI/Wan2.1-T2V-14B-Diffusers": {
-        "family": "wan",
-        "chat_template": null,
-        "gated": false,
-        "tags": [
-            "video-generation",
-            "diffusion",
-            "wan"
-        ],
-        "architecture": "WanPipeline",
-        "framework_compatibility": {
-            "vllm-omni": ">=0.16.0"
-        },
-        "validation_level": "experimental",
-        "notes": "Wan2.1 text-to-video 14B model (diffusers format). Requires multi-GPU instance (ml.g5.12xlarge or larger). Must use the -Diffusers variant — the base Wan2.1-T2V-14B repo lacks model_index.json required by vLLM-Omni"
-    },
-    "stabilityai/stable-diffusion-*": {
-        "family": "stable-diffusion",
-        "chat_template": null,
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "stable-diffusion"
-        ],
-        "architecture": null,
-        "framework_compatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validation_level": "experimental",
-        "notes": "Fallback for Stable Diffusion variants not explicitly listed"
-    },
-    "black-forest-labs/FLUX*": {
-        "family": "flux",
-        "chat_template": null,
-        "gated": false,
-        "tags": [
-            "image-generation",
-            "diffusion",
-            "flux"
-        ],
-        "architecture": null,
-        "framework_compatibility": {
-            "vllm-omni": ">=0.14.0"
-        },
-        "validation_level": "experimental",
-        "notes": "Fallback for FLUX model variants not explicitly listed"
-    }
-}
+{}

package/servers/model-picker/index.js

@@ -1531,18 +1531,25 @@ async function resolveModel({ model_id, fields, mode = 'discover', context }) {
     let values = {}
     let message = null
 
+    // Reject deprecated JumpStart prefixes
+    if (model_id.startsWith('jumpstart://') || model_id.startsWith('jumpstart-hub://')) {
+        const bareId = model_id.replace(/^jumpstart(-hub)?:\/\//, '')
+        message = `JumpStart is no longer supported. Use the HuggingFace model ID directly: ${bareId}`
+        return {
+            content: [{
+                type: 'text',
+                text: JSON.stringify({ values: {}, choices: {}, message })
+            }]
+        }
+    }
+
     if (mode === 'static') {
         // Static mode: use StaticCatalogResolver only
-        // For jumpstart:// prefixed IDs, resolve from JumpStart static catalog
         const metadata = await staticResolver.fetchModelMetadata(model_id, { fields })
         if (metadata) {
             values = { ...metadata }
         } else {
-            if (model_id.startsWith('jumpstart://')) {
-                message = `Model not found in JumpStart static catalog: ${model_id}`
-            } else {
-                message = `Model not found in static catalog: ${model_id}`
-            }
+            message = `Model not found in static catalog: ${model_id}`
         }
     } else {
         // Discover mode: use ResolverRegistry for live data, merge with static
@@ -1564,11 +1571,7 @@ async function resolveModel({ model_id, fields, mode = 'discover', context }) {
             values = { ...merged }
             // If the resolver failed but we got data from static catalog, note the fallback
             if (resolverFailed && !liveData && staticData) {
-                if (model_id.startsWith('jumpstart://')) {
-                    message = '[jumpstart] SageMaker API unreachable. Using static catalog fallback.'
-                } else if (model_id.startsWith('jumpstart-hub://')) {
-                    message = '[jumpstart-hub] SageMaker API unreachable. Using static catalog fallback.'
-                } else if (model_id.startsWith('registry://')) {
+                if (model_id.startsWith('registry://')) {
                     message = '[registry] SageMaker API unreachable. Using static catalog fallback.'
                 } else if (model_id.startsWith('s3://')) {
                     message = '[s3] S3 API unreachable. Using static catalog fallback.'
@@ -1577,11 +1580,7 @@ async function resolveModel({ model_id, fields, mode = 'discover', context }) {
         } else {
             // No data from either source
             if (resolverFailed) {
-                if (model_id.startsWith('jumpstart://')) {
-                    message = `[jumpstart] Resolver could not fetch data for: ${model_id}`
-                } else if (model_id.startsWith('jumpstart-hub://')) {
-                    message = `[jumpstart-hub] Resolver could not fetch data for: ${model_id}`
-                } else if (model_id.startsWith('registry://')) {
+                if (model_id.startsWith('registry://')) {
                     message = `[registry] Resolver could not fetch data for: ${model_id}`
                 } else if (model_id.startsWith('s3://')) {
                     message = `[s3] Resolver could not fetch data for: ${model_id}`
@@ -1613,6 +1612,18 @@ async function resolveModel({ model_id, fields, mode = 'discover', context }) {
         values = filtered
     }
 
+    // Exclude jumpstart:// prefixed results from output
+    const resolvedModelId = values.modelId || model_id
+    if (resolvedModelId.startsWith('jumpstart://') || resolvedModelId.startsWith('jumpstart-hub://')) {
+        const bareId = resolvedModelId.replace(/^jumpstart(-hub)?:\/\//, '')
+        return {
+            content: [{
+                type: 'text',
+                text: JSON.stringify({ values: {}, choices: {}, message: `JumpStart is no longer supported. Use the HuggingFace model ID directly: ${bareId}` })
+            }]
+        }
+    }
+
     // Build choices with provider prefix labels
     const choices = {}
     if (Object.keys(values).length > 0) {

package/src/app.js

@@ -142,30 +142,27 @@ export async function run(projectName, options) {
         // Infer modelSource from model name prefix if not set
         const modelName = answers.modelName;
         if (!answers.modelSource && modelName) {
+            // Reject deprecated JumpStart prefixes with migration message
+            if (modelName.startsWith('jumpstart://') || modelName.startsWith('jumpstart-hub://')) {
+                const bareId = modelName.replace(/^jumpstart(-hub)?:\/\//, '');
+                console.error(`\n   ⚠️  JumpStart is no longer supported. Use the HuggingFace model ID directly: ${bareId}`);
+                console.error('   JumpStart model sources have been removed. Use one of:');
+                console.error('     • HuggingFace model ID (e.g., meta-llama/Llama-2-7b-hf)');
+                console.error('     • s3://bucket/path/model.tar.gz');
+                console.error('     • registry://model-package-name');
+                console.error('     • marketplace://arn:aws:sagemaker:...\n');
+                process.exit(1);
+            }
             if (modelName.startsWith('s3://')) {
                 answers.modelSource = 's3';
                 if (!answers.artifactUri) {
                     answers.artifactUri = modelName;
                 }
-            } else if (modelName.startsWith('jumpstart://')) {
-                answers.modelSource = 'jumpstart';
-            } else if (modelName.startsWith('jumpstart-hub://')) {
-                answers.modelSource = 'jumpstart-hub';
             } else if (modelName.startsWith('registry://')) {
                 answers.modelSource = 'registry';
             }
         }
 
-        // Warn about unsupported model sources
-        if (answers.modelSource === 'jumpstart-hub') {
-            console.log('\n   ⚠️  JumpStart Private Hub models are not yet fully supported.');
-            console.log('   The generated project will not be able to download model artifacts at runtime.');
-            console.log('   This feature is tracked for a future release.');
-            console.log('   Falling back to HuggingFace source.\n');
-            answers.modelSource = 'huggingface';
-            delete answers.artifactUri;
-        }
-
         // Note about registry model requirements
         if (answers.modelSource === 'registry') {
             console.log('\n   ℹ️  Registry model: the container will resolve the artifact URI at startup');
@@ -352,20 +349,82 @@ export async function writeProject(templateDir, destDir, answers, registryConfig
     }
 
     // Exclude tune files when framework is NOT transformers OR deploymentTarget is batch-transform
-    if (architecture !== 'transformers' || answers.deploymentTarget === 'batch-transform') {
+    const tuneIncluded = architecture === 'transformers' && answers.deploymentTarget !== 'batch-transform';
+    if (!tuneIncluded) {
         ignorePatterns.push('**/do/tune');
         ignorePatterns.push('**/do/.tune_helper.py');
     }
 
+    // Exclude train files when deploymentTarget is batch-transform
+    const trainIncluded = answers.deploymentTarget !== 'batch-transform';
+    if (!trainIncluded) {
+        ignorePatterns.push('**/do/train');
+        ignorePatterns.push('**/do/.train_build_request.py');
+        ignorePatterns.push('**/do/.train_status_parser.py');
+        ignorePatterns.push('**/do/.train_poll_parser.py');
+        ignorePatterns.push('**/do/training/**');
+    }
+
+    // Exclude feedback.sh when neither tune nor train is included
+    if (!tuneIncluded && !trainIncluded) {
+        ignorePatterns.push('**/do/lib/feedback.sh');
+    }
+
     // Exclude do/test when hosted-model-endpoint is not selected
     const testTypes = answers.testTypes || [];
     if (!testTypes.includes('hosted-model-endpoint')) {
         ignorePatterns.push('**/do/test');
     }
 
-    // Always exclude triton and diffusors source directories
-    ignorePatterns.push('**/triton/**');
-    ignorePatterns.push('**/diffusors/**');
+    // Marketplace projects: exclude everything container-related
+    if (architecture === 'marketplace') {
+        ignorePatterns.push('**/Dockerfile');
+        ignorePatterns.push('**/code/**');
+        ignorePatterns.push('**/do/build');
+        ignorePatterns.push('**/do/push');
+        ignorePatterns.push('**/do/submit');
+        ignorePatterns.push('**/do/adapter');
+        ignorePatterns.push('**/do/adapters/**');
+        ignorePatterns.push('**/do/tune');
+        ignorePatterns.push('**/do/.tune_helper.py');
+        ignorePatterns.push('**/do/train');
+        ignorePatterns.push('**/do/.train_build_request.py');
+        ignorePatterns.push('**/do/.train_status_parser.py');
+        ignorePatterns.push('**/do/.train_poll_parser.py');
+        ignorePatterns.push('**/do/training/**');
+        ignorePatterns.push('**/do/add-ic');
+        ignorePatterns.push('**/do/run');
+        ignorePatterns.push('**/sample_model/**');
+        ignorePatterns.push('**/requirements.txt');
+        ignorePatterns.push('**/nginx-*.conf');
+        ignorePatterns.push('**/triton/**');
+        ignorePatterns.push('**/diffusors/**');
+        ignorePatterns.push('**/hyperpod/**');
+        ignorePatterns.push('**/MIGRATION.md');
+        ignorePatterns.push('**/TEMPLATE_SYSTEM.md');
+        ignorePatterns.push('**/IAM_PERMISSIONS.md');
+        ignorePatterns.push('**/PROJECT_README.md');
+        ignorePatterns.push('**/deploy_notebook_generator.py');
+        ignorePatterns.push('**/buildspec.yml');
+        ignorePatterns.push('**/test/**');
+        // Exclude templates that reference container-specific variables (framework, modelServer)
+        // Marketplace overlays its own config, deploy, and test templates
+        ignorePatterns.push('**/do/config');
+        ignorePatterns.push('**/do/deploy');
+        ignorePatterns.push('**/do/test');
+        ignorePatterns.push('**/do/README.md');
+        ignorePatterns.push('**/do/export');
+        ignorePatterns.push('**/do/validate');
+        ignorePatterns.push('**/do/ic/**');
+    }
+
+    // Always exclude architecture-specific source directories from main copy
+    // (they are overlaid separately for their respective architectures)
+    ignorePatterns.push('**/marketplace/**');
+    if (architecture !== 'marketplace') {
+        ignorePatterns.push('**/triton/**');
+        ignorePatterns.push('**/diffusors/**');
+    }
 
     // For triton and diffusors, exclude the default Dockerfile
     if (architecture === 'triton' || architecture === 'diffusors') {
@@ -431,6 +490,14 @@ export async function writeProject(templateDir, destDir, answers, registryConfig
         _copyFile(path.join(templateDir, 'diffusors/patch_image_api.py'), path.join(destDir, 'code/patch_image_api.py'));
         break;
 
+    case 'marketplace':
+        // Marketplace projects: overlay marketplace-specific templates
+        // These replace the default do/config, do/deploy, and do/test with marketplace versions
+        _renderTemplate(path.join(templateDir, 'marketplace/config'), path.join(destDir, 'do/config'), templateVars);
+        _renderTemplate(path.join(templateDir, 'marketplace/deploy'), path.join(destDir, 'do/deploy'), templateVars);
+        _renderTemplate(path.join(templateDir, 'marketplace/test'), path.join(destDir, 'do/test'), templateVars);
+        break;
+
     default:
         // Fallback to HTTP behavior
         _unlinkIfExists(path.join(destDir, 'code/chat_template.jinja'));
@@ -450,7 +517,10 @@ export async function writeProject(templateDir, destDir, answers, registryConfig
     }
 
     // Copy PROJECT_README.md as README.md (overwriting the template README)
-    _renderTemplate(path.join(templateDir, 'PROJECT_README.md'), path.join(destDir, 'README.md'), templateVars);
+    // Marketplace projects don't use the standard README (no container/framework info)
+    if (architecture !== 'marketplace') {
+        _renderTemplate(path.join(templateDir, 'PROJECT_README.md'), path.join(destDir, 'README.md'), templateVars);
+    }
 
     // Copy do/lib/ Node.js modules (plain copy, no EJS)
     const doLibDir = path.join(destDir, 'do', 'lib');
@@ -491,7 +561,7 @@ export async function writeProject(templateDir, destDir, answers, registryConfig
  */
 export async function postGenerate(destDir, answers, tritonBackends = {}) {
     // Set executable permissions on shell scripts
-    _setExecutablePermissions(destDir);
+    _setExecutablePermissions(destDir, answers);
 
     // Run sample model training if requested
     const architecture = answers.architecture;
@@ -1092,8 +1162,25 @@ function _unlinkIfExists(filePath) {
  *
  * @param {string} destDir - Path to the generated project directory
  */
-function _setExecutablePermissions(destDir) {
-    const shellScripts = [
+function _setExecutablePermissions(destDir, answers = {}) {
+    const architecture = answers.architecture;
+
+    // Marketplace projects have a reduced set of scripts
+    const marketplaceScripts = [
+        'do/config',
+        'do/deploy',
+        'do/test',
+        'do/logs',
+        'do/clean',
+        'do/register',
+        'do/ci',
+        'do/manifest',
+        'do/benchmark',
+        'do/optimize',
+        'do/status'
+    ];
+
+    const defaultScripts = [
         'do/config',
         'do/build',
         'do/push',
@@ -1111,9 +1198,12 @@ function _setExecutablePermissions(destDir) {
         'do/status',
         'do/add-ic',
         'do/adapter',
-        'do/tune'
+        'do/tune',
+        'do/train'
     ];
 
+    const shellScripts = architecture === 'marketplace' ? marketplaceScripts : defaultScripts;
+
     shellScripts.forEach(script => {
         const scriptPath = path.join(destDir, script);
         try {

package/src/lib/cli-handler.js

@@ -178,7 +178,7 @@ CLI OPTIONS:
   --project-name=<name>       Project name
   --project-dir=<dir>         Output directory path
   --framework=<framework>     ML framework (sklearn|xgboost|tensorflow|transformers)
-  --model-name=<name>         HuggingFace model name (for transformers framework)
+  --model-name=<name>         Model identifier (<hf-org/model>, s3://..., registry://..., marketplace://...)
   --model-server=<server>     Model server (flask|fastapi|vllm|sglang|tensorrt-llm|lmi|djl)
   --model-format=<format>     Model format (depends on framework)
   --include-sample            Include sample model code