endpoint yaml specification missing resources field

[amahab]

Doc here states k8s isntance_type can be specified via resources for endpoint - https://github.com/Azure/AML-Kubernetes/blob/master/docs/simple-flow.md

Adding this field causes error in endpoint creation.

Resources field is not in the spec - https://learn.microsoft.com/en-us/azure/machine-learning/reference-yaml-endpoint-online

[Bozhong68]

Hi @amahab thanks for feedback, the github documentation/examples now are outdated, the GitHub was meant for public preview purpose, since the product now is GAed, please use official documentation https://aka.ms/amlarc/doc and official AzureML example repo: https://github.com/Azure/azureml-examples.

Thank you so much.

[amahab]

Hi @Bozhong68

Where can I find the compute instance_type spec for endpoints for on prem Azure Arc k8s deployments. The link you point to has examples for Azure cloud.

[Bozhong68]

It is applicable to both cases - Instance type official documentation here: https://learn.microsoft.com/en-us/azure/machine-learning/how-to-manage-kubernetes-instance-types

[Bozhong68]

@amahab YAML schema for resource section here: https://learn.microsoft.com/en-us/azure/machine-learning/reference-yaml-deployment-kubernetes-online

[amahab]

@Bozhong68 The instance official doc helped. Thanks!

Regarding AML-K8s examples - This is the only site where I find on prem k8s examples. Any other links?

Also how can i specify k8s instance type via Az ML python SDK instead of CLI? I have filed issue #262 on this.

[jiaochenlu]

@amahab Have replied how to specify instance type using SDK in issue #262

More examples you can find in our official AzureML example repo: https://github.com/Azure/azureml-examples.

To use k8s compute for training job, you just need to specify your k8s compute name in "compute" parameter, and specify your instanceType in "instance_type" parameter.

To use k8s compute for real-time inference, you can follow this SDK example, or this CLI example