如何使用 az ml cli 获取在 Azure 机器学习服务 Model 注册表中注册的最新版本的 Model ID？

Question

Using Azure Machine Learning CLI extension, how do we get the Model ID for the latest version of a Model (with known model name)?

To get the entire list of Model Details with a given name the command is

az ml model list --model-name [Model_Name] --resource-group [RGP_NAME] --subscription-id [SUB_ID] --workspace-name [WS_NAME]

Running this will give a list of all the models:

[
  {
    "createdTime": "2021-03-19T07:02:03.814172+00:00",
    "framework": "Custom",
    "frameworkVersion": null,
    "id": "model:2"
    "name": "model",
    "version": 3
  },
  {
    "createdTime": "2021-03-19T06:46:34.301054+00:00",
    "framework": "Custom",
    "frameworkVersion": null,
    "id": "model:2",
    "name": "model",
    "version": 2
  },
  {
    "createdTime": "2021-03-19T06:38:56.558385+00:00",
    "framework": "Custom",
    "frameworkVersion": null,
    "id": "model:1",
    "name": "model",
    "version": 1
  }
]

The Microsoft Documentation mentions, we can use a -l parameter to get the latest version details:

az ml model list --model-name [Model_Name] --resource-group [RGP_NAME] --subscription-id [SUB_ID] --workspace-name [WS_NAME] -l

However, running this gives the following error:

ERROR: UnrecognizedArgumentError: unrecognized arguments: -l

What is the syntax to use this -l flag?

Answer 1

I've just spent a fun few hours wrestling with this. Running something like

az ml model list -w your_workspace_name -g your_resource_group_name -l -n name_of_your_registered_model

will get you back some JSON that looks like this:

[
  {
    "createdTime": "2022-03-04T16:05:47.103407+00:00",
    "framework": "Custom",
    "frameworkVersion": null,
    "id": "name_of_your_registered_model:3",
    "name": "name_of_your_registered_model",
    "version": 3
  }
]

This is well and good for a human, but not much use to a machine. The azure cli supports something called JMESPath , which allows you to write a query against the result of a CLI command. Running

az ml model list -w your_workspace_name -g your_resource_group_name -l -n name_of_your_registered_model --query "[0].{id:id}" -o tsv

should get you back

"name_of_your_registered_model:3"

Which you can then use in an env var or whatever other use case you have.

Answer 2

If we wish to obtain the model-id for the latest model, instead of using az ml model list with -l flag, using az model show will return the details for the latest model. The syntax to get a string for model-id will be:

az ml model show --model-id $(TRN_MODEL_ID) --resource-group $(AML_TRN_RG) --subscription-id $(AML_TRN_SUB_ID) --workspace-name $(AML_TRN_WS) --query name -o tsv