mirror of
https://github.com/openshift/openshift-docs.git
synced 2026-02-05 12:46:18 +01:00
51 lines
2.8 KiB
Plaintext
51 lines
2.8 KiB
Plaintext
// Module included in the following assemblies:
|
|
//
|
|
// * microshift_ai/microshift-rhoai.adoc
|
|
|
|
:_mod-docs-content-type: CONCEPT
|
|
[id="microshift-rhoai-workflow_{context}"]
|
|
= Workflow for using {rhoai} with {microshift-short}
|
|
|
|
Using {rhoai} with {microshift-short} requires the following general workflow:
|
|
|
|
Getting your AI model ready::
|
|
|
|
* Choose the artificial intelligence (AI) model that best aligns with your edge application and the decisions that need to be made at {microshift-short} deployment sites.
|
|
* In the cloud or data center, develop, train, and test your model.
|
|
* Plan for the system requirements and additional resources your AI model requires to run.
|
|
|
|
Setting up the deployment environment::
|
|
|
|
* Configure your {op-system-bundle} for the specific hardware your deployment runs on, including driver and device plugins.
|
|
|
|
* To enable GPU or other hardware accelerators for {microshift-short}, follow the guidance specific for your edge device about what you need to install. For example, to use an NVIDIA GPU accelerator, begin by reading the following NVIDIA documentation: link:https://docs.nvidia.com/datacenter/cloud-native/edge/latest/nvidia-gpu-with-device-edge.html#running-a-gpu-accelerated-workload-on-red-hat-device-edge[Running a GPU-Accelerated Workload on Red Hat Device Edge] (NVIDIA documentation).
|
|
|
|
* For troubleshooting, consult the device documentation or product support.
|
|
+
|
|
[TIP]
|
|
====
|
|
Using only a driver and device plugin instead of an Operator might be more resource efficient.
|
|
====
|
|
|
|
Installing the {microshift-short} {rhoai} RPM::
|
|
|
|
* Install the `microshift-ai-model-serving` RPM package.
|
|
|
|
* Restart {microshift-short} if you are adding the RPM while {microshift-short} is running.
|
|
|
|
Getting ready to deploy::
|
|
|
|
* Package your AI model into an OCI image, otherwise known as the ModelCar format. If you already have S3-compatible storage or a persistent volume claim set up, you can skip this step, but only the ModelCar format is tested and supported for {microshift-short}.
|
|
|
|
* Select a model-serving runtime, which acts as your model server. Configure the runtime with the serving runtime and inference service.
|
|
|
|
** Copy the `ServingRuntime` custom resource (CR) from the default `redhat-ods-applications` namespace to your own namespace.
|
|
|
|
** Create the `InferenceService` CR.
|
|
|
|
* Optional: Create a `Route` object so that your model can connect outside the node.
|
|
|
|
Using your model::
|
|
|
|
* Make requests against the model server. For example, another pod running in your {microshift-short} deployment that is attached to a camera can stream an image back to the model-serving runtime. The model-serving runtime prepares that image as data for model inferencing. If the model was trained in the binary identification of a bee, the AI model outputs the likelihood that the image data is a bee.
|