Release 2026-03-30

docs/troubleshooting.md

@@ -0,0 +1,26 @@
+# Troubleshooting (FAQ)
+
+## Overview
+
+This document provides guidance for common problems that might be faced when using the Nova Forge SDK. 
+
+## Permissions-Based Issues
+### Unable to Deploy a Custom Model to Bedrock
+If you are unable to use the SDK's built-in `deploy()` function due to permissioning issues, you can manually call the Bedrock APIs to import and deploy your models. 
+This will still require some IAM permissions to be set up. 
+The steps are outlined below.
+
+#### Step 1: Locate Your Training Artifacts and extract your checkpoint_s3_path
+* First, find where your training job is saved (`output_s3_path`) in S3. 
+* For SMTJ jobs, follow the steps [here](https://docs.aws.amazon.com/nova/latest/nova2-userguide/nova-iterative-training.html#nova-iterative-how-it-works) to get the s3 escrow location where your model is saved. 
+* For SMHP, when you navigate to your `output_s3_path` S3 folder, open the `manifest.json` file which will only contain the `checkpoint_s3_path` value. 
+
+#### Step 2: Import your custom model from S3 Escrow
+* Follow the steps to [Create a Custom Model](https://docs.aws.amazon.com/bedrock/latest/userguide/create-custom-model-sdks.html) here. 
+* Provide the `checkpoint_s3_path` value from Step 1 for the `s3Uri` value under `modelSourceConfig`. 
+
+#### Step 3: Deploy your custom model in Bedrock
+* After your custom model is imported from escrow, you can follow the steps [here](https://docs.aws.amazon.com/bedrock/latest/userguide/deploy-custom-model-on-demand.html#deploy-custom-model) to deploy the model to Bedrock using the Console, AWS CLI, or Bedrock APIs. 
+
+#### Notes:
+* If you're running into permission issues with importing and deploying your custom Nova model, please review the AWS documentation: [Create a service role for importing pre-trained models](https://docs.aws.amazon.com/bedrock/latest/userguide/model-import-iam-role.html).

pyproject.toml

@@ -19,7 +19,7 @@ readme = "README.md"
 dependencies = [
     "boto3",
     "matplotlib",
-    "sagemaker==3.5.0",
+    "sagemaker>=3.5.0",
 ]
 
 [project.optional-dependencies]

samples/nova_quickstart.ipynb

@@ -652,6 +652,37 @@
     "monitor.plot_metrics(training_method=TrainingMethod.SFT_LORA)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### C) Enable Job Notifications (SMTJ, SMHP)\n",
+    "* The Forge SDK provides job notification support via email. \n",
+    "* For more information, refer to [`docs/job_notifications.md`](job_notifications.md)."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Enable job notifications for SMTJ (given a TrainingResult object)\n",
+    "# Note: This will attempt to set up a CloudFormation stack if the infrastructure isn't deployed yet.\n",
+    "training_result.enable_job_notifications(\n",
+    "    emails=[\"user@example.com\"],\n",
+    "    # kms_key_id=\"1234abcd-12ab-34cd-56ef-1234567890ab\", # Optional customer KMS key\n",
+    ")\n",
+    "\n",
+    "# Enable job notifications for SMHP (given a TrainingResult object)\n",
+    "training_result.enable_job_notifications(\n",
+    "    emails=[\"user@example.com\"],\n",
+    "    namespace=\"kubeflow\",  # REQUIRED: Kubernetes namespace where job runs\n",
+    "    kubectl_layer_arn=\"arn:aws:lambda:<region>:123456789012:layer:kubectl:1\",  # REQUIRED\n",
+    "    # kms_key_id=\"1234abcd-12ab-34cd-56ef-1234567890ab\"  # Optional customer KMS key\n",
+    ")"
+   ]
+  },
   {
    "cell_type": "markdown",
    "metadata": {},
@@ -1139,7 +1170,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.12.11"
+   "version": "3.12.4"
   }
  },
  "nbformat": 4,

samples/rft_multiturn_quickstart.ipynb

@@ -403,9 +403,7 @@
    "source": [
     "## Step 5: Evaluation Setup and Execution (Optional)\n",
     "\n",
-    "You can run evaluation before training to test your base model, or after training to test your fine-tuned model.\n",
-    "\n",
-    "**Important**: RFT multiturn evaluation uses the environment's built-in examples, NOT a separate dataset. The environment generates examples based on the `num_examples` parameter."
+    "You can run evaluation before training to test your base model, or after training to test your fine-tuned model.\n"
    ]
   },
   {
@@ -659,7 +657,6 @@
     ")\n",
     "\n",
     "print(f\"\\n✅ Evaluation started: {eval_result.job_id}\")\n",
-    "print(f\"   Environment will generate {50} examples\")\n",
     "eval_result.dump(file_name=\"eval_result.json\")"
    ]
   },
@@ -669,6 +666,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "# from amzn_nova_forge.model.result import EvaluationResult\n",
+    "\n",
     "# Load evaluation result from file (e.g., after notebook restart)\n",
     "# loaded_eval_result = EvaluationResult.load(\"eval_result.json\")\n",
     "# print(\"✅ Evaluation result loaded from file\")\n",
@@ -1018,6 +1017,8 @@
    "metadata": {},
    "outputs": [],
    "source": [
+    "# from amzn_nova_forge.model.result import TrainingResult\n",
+    "\n",
     "# Load training result from file (e.g., after notebook restart)\n",
     "# loaded_training_result = TrainingResult.load(\"training_result.json\")\n",
     "# print(\"✅ Training result loaded from file\")\n",