Dear SAP Community,

In this edition of our quarterly What’s New blog post, we are happy to introduce the latest features that were released for SAP AI Core & SAP AI Launchpad in Q1 2022.

SAP AI Core is a service in the SAP Business Technology Platform which is designed to handle the execution and operations of your AI scenarios. It provides seamless integration with your SAP solutions and supports full lifecycle management of AI scenarios.

SAP AI Launchpad is a software as a service (SaaS) application in SAP Business Technology Platform. Customers and partners can use SAP AI Launchpad to manage AI scenarios across multiple instances of AI runtimes, such as SAP AI Core.

Contents

• • SAP AI Core
    • Metaflow Python Library for SAP AI Core
    • Meta API for AI API Runtime Implementations
    • Register Generic Secrets
  • SAP AI Launchpad
    • Download of Executions and Deployment Logs
    • Update Deployment with a Chosen Configuration
    • Expanded Filter and Search

SAP AI Core

Metaflow Python Library for SAP AI Core

python trainflow.py --with=kubernetes:secrets=default-object-store-secret argo create --image-pull-secret=<AI Core docker secret> --label={"scenarios.ai.sap.com/id":"<Ml-Scenario>","ai.sap.com/version":"1.0.0"} --annotation={"scenarios.ai.sap.com/name":"<ML-Scenario-Name>","executables.ai.sap.com/name":"trainflow"} --only-json > trainflow.json

Metaflow is an open-source tool, developed by Netflix, that helps data scientists and developers to create scalable machine learning services and bring them to production faster. The Metaflow Python Library for SAP AI Core is a command-line tool that extends Metaflow’s capabilities to run ML pipelines as Argo Workflows which can be executed on SAP AI Core. This enables developers to transition more easily from the experimentation phase to production by automating the conversion from Metaflow pipelines to Argo Workflows. Generating an Argo Workflow like the one below from Metaflow pipelines with our new library is as easy as executing the command above.

{
    "apiVersion": "argoproj.io/v1alpha1",
    "kind": "WorkflowTemplate",
    "metadata": {
        "name": "trainflow",
        "labels": {
            "metaflow/runtime": "dev",
            "metaflow/date": "2022-05-02",
            "metaflow/metaflow_version": "2.5.0",
            "metaflow/python_version": "3.8.9",
            "scenarios.ai.sap.com/id": "metaflow-demo",
            "ai.sap.com/version": "1.0.0",
            "app": "metaflow",
            "metaflow/workflow_template": "trainflow",
        },
        "annotations": {
            "scenarios.ai.sap.com/name": "trainflow",
            "executables.ai.sap.com/name": "train demo",
            "metaflow/flow_name": "TrainFlow"
        }
    },
    "spec": {
        "entrypoint": "entry",
        "workflowMetadata": {
            "labels": {
                "app": "metaflow",
                "metaflow/workflow_template": "trainflow"
            },
            "annotations": {
                "metaflow/flow_name": "TrainFlow"
            }
        }
 ...
}

For more details, check out our new blog post that explains how to use the Metaflow Python Library for SAP AI Core.

Help: Metaflow Python Library for SAP AI Core

Meta API for AI API Runtime Implementations

{
 …
 "aiApi": {
 	"version": "2.12.0",
 	"capabilities": {
 		"multitenant": true,
 		"staticDeployments": false,
 		"userDeployments": false,
 		"createExecutions": true,
 		"logs": {
 			"executions": true,
 			"deployments": false
 		}
 	},
 …
}

The new Meta API endpoint in the AI API specification lets clients query the capabilities that are supported by a runtime that implements AI API. Based on the response, the client can trigger an appropriate behavior, such as enabling or disabling certain features based on the capabilities of the particular runtime. This way you can create a clean user experience by simply hiding the features that are not supported.

Help: AI API Runtime Implementations

Register Generic Secrets

{
 "name": "my-generic-secret",
 "data": {
 	"some-credential": "bXktc2VjcmV0LWNyZWRlbnRpYWw=",
 	"other-credentials": "bXktc2VjcmV0LW90aGVyLWNyZWRlbnRpYWw="
 }
}

You can now store generic secrets at main-tenant level as well as on resource-group level. Generic secrets are different to system secrets (such as secrets for object stores, Docker registry, and so on) and can be used to store sensitive information, either for the main tenant or for each resource group via an API.

spec:
	containers:
	- name: my-kfserving-container
		image: centaur
		envFrom:
		- secretRef:
			name: my-generic-secret					

You can then attach the secrets to containers in executions or deployments as environment variables or volume mounts as shown in the screenshot above.

Help: Register Generic Secrets

SAP AI Launchpad

Download of Execution and Deployment Logs

You can now download the logs of training executions and deployments to your device, offering you the freedom of processing the logs in any tool of your choice. This allows you to run more sophisticated analyses and to find the root cause of any problems faster.
Trigger the download of the logs from the ML Operations application in your SAP AI Launchpad and optionally specify a time range if you want to review only a portion of the logs. The logs are downloaded in JSON format.

Help: View Execution Logs | Help: View Deployment Logs

Update Deployment with a Chosen Configuration

Continuously improve your models without disrupting ongoing business! You can now update deployments with your choice of configuration while retaining the API endpoint of the deployment. You can choose to apply a new configuration, or re-use and older or previously used configuration. For example, you could replace the currently used model with a new one or update any parameters of the deployment. The API endpoint of your deployment does not change so that applications consuming your model don’t require any changes. This enables you to make continuous improvements.

Help: Update a Deployment

Expanded Filter and Search

Filter and search have been enabled for more pages in the Function Explorer application in your SAP AI Launchpad. You can now search for models, datasets, or configurations using their ID or description as well as filter for jobs by specifying attribute values or conditions.

At the same time, we added additional columns to the lists in the Function Explorer and ML Operations applications, showing additional scenario details and allowing you to intuitively find what you are looking for.

This blog post only provides a high-level overview of the latest features and enhancements of SAP AI Core & SAP AI Launchpad. You can find more details on these features in the SAP Help Portal and the respective pages for SAP AI Core & SAP AI Launchpad.

If you have any questions, you can comment below or post a question with the SAP AI Core tag or SAP AI Launchpad tag. Also check out our SAP Road Map Explorer to see more upcoming features.