DocsICP FilingConsole
官方文档
首页

Register and manage models

更新时间：
复制 MD 格式
产品详情
我的收藏

Model Management lets you register trained models in PAI, manage them across versions, and deploy them to Elastic Algorithm Service (EAS) for inference.

Prerequisites

Model training is complete and the model files are uploaded to OSS. Train and export models in Designer or DSW. To upload the model files to OSS, see Upload objects.

Register a model

Register a new model or add a version to an existing model:

  1. Go to the Model Management page.

    1. Log in to the PAI console.

    2. In the left navigation pane, click Workspaces, then click the target workspace name to enter the workspace.

    3. In the left navigation pane, choose AI Computing Asset Management > Models.

  2. On the Models page, click Register Model.

  3. In the Register Model panel, configure the parameters and click OK.

    Two options are available:

    • Create Model: Register a trained model as a new model entry.

    • Create Version: Add a new version to an already-registered model.

    Key parameters:

    Parameter

    Description

    Model Name

    Required only when Create Model is selected.

    Select Model

    Required only when Create Version is selected.

    Select an already-registered model.

    Version Number

    The version number of the model being registered. Editable only when Create Version is selected.

    Model Address

    Specify the model location in one of the following ways:

    • Enter the OSS path where the model files are stored.

    • Click image to browse and select the OSS path.

      In the Select OSS folder or file dialog box, select an existing model file or upload a local model file:

      1. (Optional) To upload model files to a new directory, click Create Directory to create a directory in the current OSS bucket. Double-click the new directory to navigate into it.

      2. Click Upload File.

      3. Click Browse Local Files or Drag and Drop File to Upload and follow the prompts to upload your local model file.

    Model Service Deployment

    Two options are available:

    • Automatic Configurations: After you set Model Format, the system generates the deployment configuration automatically.

    • Custom Configuration: Enter the deployment configuration directly in the text field. See the examples below and refer to JSON deployment for parameter descriptions.

      • For models deployed with a custom image:

        Custom image deployment example

        {
    "metadata": {
        "name": "",
        "instance": 1,
        "workspace_id": "210382",
        "disk": "30Gi",
        "rpc": {
            "keepalive": 60000
        },
        "accessibility": "public"
    },
    "cloud": {
        "computing": {
            "instances": [
                {
                    "type": "ecs.gn6e-c12g1.12xlarge"
                }
            ]
        }
    },
    "containers": [
        {
            "image": "eas-registry-vpc.cn-hangzhou.cr.aliyuncs.com/pai-eas/python-inference:py3.9-ubuntu2004",
            "script": "python app.py",
            "port": 8000
        }
    ],
    "options": {},
    "storage": [
        {
            "oss": {
                "path": "oss://*******/MODEL/",
                "readOnly": false,
                "role_arn": ""
            },
            "mount_path": "/mnt/data/"
        }
    ]
}

      • For models that use a built-in Processor:

        Built-in Processor deployment example

        {
    "metadata": {
        "name": "",
        "instance": 1,
        "workspace_id": "210382",
        "disk": "30Gi",
        "rpc": {
            "keepalive": 60000
        },
        "accessibility": "public"
    },
    "cloud": {
        "computing": {
            "instances": [
                {
                    "type": "ecs.gn6i-c16g1.4xlarge"
                }
            ]
        }
    },
    "options": {},
    "processor": "tensorflow_gpu_2.4",
    "model_config": {},
    "model_path": "oss://******/MODEL/"
}

    Advanced Configurations

    Select Advanced Configurations to configure the following parameters:

    • Version Approval Status

      Two approval statuses are available:

      • Pending (default)

      • Approved

    • Model Metrics: Set model metrics to compare performance across versions. For details, see Model metrics.

Manage models

The Model Management page lists all models you have permission to access. From this page, view model details, delete models, or deploy models to EAS.

  • Click View to open the model details page and view information such as the model ID. Click Delete to delete a model that is no longer needed.

  • Click the settings icon in the upper-right corner to customize which columns appear in the model list.

Deploy a service

Deploy a registered model to EAS. The Inference Specification set during model registration is pre-filled in the EAS deployment page.

  1. In the left navigation pane, choose AI Asset Management > Models. On the model list page, click Deploy in EAS in the Actions column. The page redirects to the EAS deployment page.

    Note: For models from Model Gallery, click Deploy to deploy the model.

  2. The deployment page is pre-filled with key parameters from the Inference Specification. Configure the remaining required parameters and deploy. For details, see Custom deployment.

Previous：Manage modelsNext：Model metrics