@aws/ml-container-creator 1.0.0 → 1.0.3

package/bin/cli.js

@@ -162,7 +162,7 @@ program
     .command('bootstrap')
     .description('Set up AWS infrastructure (IAM role, ECR repo, S3 buckets)')
     .passThroughOptions()
-    .argument('[action]', 'Bootstrap action (status, use, list, remove, scan, prune, update, sync-schemas)')
+    .argument('[action]', 'Bootstrap action (status, use, list, remove, scan, prune, update, migrate, sync-schemas, sync-model-families)')
     .argument('[args...]', 'Additional arguments')
     .option('--profile <profile>', 'AWS profile name')
     .option('--region <region>', 'AWS region')

package/config/tune-catalog.json

@@ -1,6 +1,6 @@
 {
   "version": "2026-05-27",
-  "lastSynced": "2026-05-28T09:48:25.209Z",
+  "lastSynced": "2026-06-26T19:01:02.821Z",
   "source": "https://docs.aws.amazon.com/sagemaker/latest/dg/model-customize-open-weight.html",
   "models": {
     "huggingface-llm-qwen2-5-7b-instruct": {
@@ -1614,6 +1614,24 @@
               "prompt": "array"
             }
           }
+        },
+        "dpo": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-dpo",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "chosen",
+              "rejected"
+            ],
+            "types": {
+              "prompt": "string",
+              "chosen": "string",
+              "rejected": "string"
+            }
+          }
         }
       },
       "goldenPath": false
@@ -1667,6 +1685,24 @@
               "prompt": "array"
             }
           }
+        },
+        "dpo": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-dpo",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "chosen",
+              "rejected"
+            ],
+            "types": {
+              "prompt": "string",
+              "chosen": "string",
+              "rejected": "string"
+            }
+          }
         }
       },
       "goldenPath": false
@@ -1773,6 +1809,272 @@
               "prompt": "array"
             }
           }
+        },
+        "dpo": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-dpo",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "chosen",
+              "rejected"
+            ],
+            "types": {
+              "prompt": "string",
+              "chosen": "string",
+              "rejected": "string"
+            }
+          }
+        }
+      },
+      "goldenPath": false
+    },
+    "huggingface-llm-nvidia-nemotron-3-super-120b-a12b-bf16": {
+      "family": "huggingface-llm-nvidia-nemotron",
+      "provider": "unknown",
+      "displayName": "NVIDIA-Nemotron-3-Super-120B-A12B-BF16",
+      "huggingFaceId": "",
+      "techniques": {
+        "sft": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-sft",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "completion"
+            ],
+            "types": {
+              "prompt": "string",
+              "completion": "string"
+            }
+          }
+        },
+        "rlvr": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlvr",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        },
+        "rlaif": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlaif",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        }
+      },
+      "goldenPath": false
+    },
+    "huggingface-reasoning-nvidia-nemotron-3-nano-30b-a3b-bf16": {
+      "family": "huggingface-reasoning-nvidia-nemotron",
+      "provider": "unknown",
+      "displayName": "NVIDIA-Nemotron-3-Nano-30B-A3B-BF16",
+      "huggingFaceId": "",
+      "techniques": {
+        "sft": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-sft",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "completion"
+            ],
+            "types": {
+              "prompt": "string",
+              "completion": "string"
+            }
+          }
+        },
+        "rlaif": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlaif",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        },
+        "rlvr": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlvr",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        }
+      },
+      "goldenPath": false
+    },
+    "huggingface-vlm-gemma-4-e4b-it": {
+      "family": "huggingface-vlm",
+      "provider": "unknown",
+      "displayName": "gemma-4-e4b-it",
+      "huggingFaceId": "",
+      "techniques": {
+        "dpo": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-dpo",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "chosen",
+              "rejected"
+            ],
+            "types": {
+              "prompt": "string",
+              "chosen": "string",
+              "rejected": "string"
+            }
+          }
+        },
+        "sft": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-sft",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "completion"
+            ],
+            "types": {
+              "prompt": "string",
+              "completion": "string"
+            }
+          }
+        },
+        "rlvr": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlvr",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        },
+        "rlaif": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlaif",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        }
+      },
+      "goldenPath": false
+    },
+    "huggingface-vlm-gemma-4-31b-it": {
+      "family": "huggingface-vlm",
+      "provider": "unknown",
+      "displayName": "gemma-4-31b-it",
+      "huggingFaceId": "",
+      "techniques": {
+        "dpo": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-dpo",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "chosen",
+              "rejected"
+            ],
+            "types": {
+              "prompt": "string",
+              "chosen": "string",
+              "rejected": "string"
+            }
+          }
+        },
+        "sft": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-sft",
+          "datasetSchema": {
+            "required": [
+              "prompt",
+              "completion"
+            ],
+            "types": {
+              "prompt": "string",
+              "completion": "string"
+            }
+          }
+        },
+        "rlaif": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlaif",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
+        },
+        "rlvr": {
+          "trainingTypes": [
+            "lora"
+          ],
+          "datasetFormat": "default-rlvr",
+          "datasetSchema": {
+            "required": [
+              "prompt"
+            ],
+            "types": {
+              "prompt": "array"
+            }
+          }
         }
       },
       "goldenPath": false

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aws/ml-container-creator",
-  "version": "1.0.0",
+  "version": "1.0.3",
   "description": "Build and deploy custom ML containers on AWS SageMaker with minimal configuration.",
   "main": "src/index.js",
   "bin": {
@@ -107,6 +107,7 @@
     "prepare": "husky || true"
   },
   "dependencies": {
+    "@aws/ml-container-creator": "^1.0.2",
     "@inquirer/prompts": "^8.4.2",
     "@modelcontextprotocol/sdk": "^1.27.1",
     "ajv": "^8.12.0",

package/servers/endpoint-picker/index.js

@@ -200,22 +200,37 @@ async function fetchEndpoints(client, { limit = 10, showFull = false } = {}) {
 
             const variantName = primaryVariant.VariantName || 'AllTraffic';
             let instanceType = primaryVariant.InstanceType || null;
+            let instancePools = primaryVariant.InstancePools || null;
 
             // For IC-based endpoints, InstanceType may not be in the variant runtime response.
-            // Fall back to DescribeEndpointConfig which always has it.
-            if (!instanceType && detail.EndpointConfigName) {
+            // Fall back to DescribeEndpointConfig which has either InstanceType or InstancePools.
+            if (!instanceType && !instancePools && detail.EndpointConfigName) {
                 try {
                     const ecCmd = new _DescribeEndpointConfigCommand({ EndpointConfigName: detail.EndpointConfigName });
                     const ecDetail = await client.send(ecCmd);
                     const ecVariant = (ecDetail.ProductionVariants || [])[0];
                     if (ecVariant?.InstanceType) {
                         instanceType = ecVariant.InstanceType;
+                    } else if (ecVariant?.InstancePools && ecVariant.InstancePools.length > 0) {
+                        instancePools = ecVariant.InstancePools;
                     }
                 } catch (ecErr) {
                     log(`Warning: could not describe endpoint config for "${endpointName}": ${ecErr.message}`);
                 }
             }
-            instanceType = instanceType || 'unknown';
+
+            // Resolve instanceType display string from pools if needed
+            if (!instanceType && instancePools && instancePools.length > 0) {
+                // Sort by priority, use highest-priority (lowest number) for GPU lookup
+                const sorted = [...instancePools].sort((a, b) => (a.Priority || 99) - (b.Priority || 99));
+                instanceType = sorted[0].InstanceType || 'unknown';
+                // Build display string showing the pool: "ml.g5.12xl (pool: 3 types)"
+                if (sorted.length > 1) {
+                    instanceType = `${instanceType} (pool: ${sorted.length} types)`;
+                }
+            } else {
+                instanceType = instanceType || 'unknown';
+            }
 
             const instanceCount = primaryVariant.CurrentInstanceCount ?? primaryVariant.DesiredInstanceCount ?? 1;
             const hasInstancePools = !!(primaryVariant.InstancePools && primaryVariant.InstancePools.length > 0);
@@ -244,7 +259,12 @@ async function fetchEndpoints(client, { limit = 10, showFull = false } = {}) {
             } while (icNextToken);
 
             // Capacity estimation
-            const gpusPerInstance = getGpusForInstance(instanceType);
+            // For pool endpoints, instanceType may be "ml.g5.12xlarge (pool: 3 types)"
+            // Extract the raw type for catalog lookup
+            const instanceTypeForLookup = instanceType.includes(' (pool:')
+                ? instanceType.split(' (pool:')[0]
+                : instanceType;
+            const gpusPerInstance = getGpusForInstance(instanceTypeForLookup);
             let availableGpus;
             if (gpusPerInstance === null) {
                 availableGpus = '?';