Skip to content

On-premise users: click in-app to access the full platform documentation for your version of DataRobot.

Manage key values for registered model versions

After you register a model, you can add key values and edit existing (user-created) key values from the Registry. For more information on key values, see the documentation.

Runtime parameter key values

You cannot create or edit runtime parameter key values in the Registry, those are defined in the Model workshop.

Add key values

To add a new key value to a registered model version:

  1. On the Model directory page, in the table of registered models, click the registered model containing the version you want to manage key values for, opening the list of versions.

  2. In the list of versions, click the version you want to edit, opening the registered model version panel:

  3. Click the Overview tab and locate the Key values section.

  4. In the group box for a key values category, click + Add (or, if one or more key values exist for that category, click + Add tag, + Add metric, etc.).

  5. In the Add key value(s) dialog box, select one of the following:

    To add a new key value to a registered model version, configure the following fields:

    Setting Description
    Category Defaults to the category of the group box where you clicked + Add. Select one of the following categories for the new key value to organize your key values by purpose:
    • Training parameter
    • Metric
    • Tag
    • Artifact
    Value type Select one of the following value types for the new key value:
    • String
    • Numeric
    • Boolean
    • URL
    • JSON
    • YAML
    Name Enter a descriptive name for the key in the key-value pair.
    Value If you selected one of the following value types, enter the appropriate data:
    • String: Enter any string up to 4 KB.
    • Numeric: Enter an integer or floating-point number.
    • Boolean: Select True or False.
    • URL: A URL in the format scheme://location; for example, https://example.com. DataRobot does not fetch the URL or provide a link to this URL in the user interface; however, in a downloaded compliance document, the URL may appear as a link.
    • JSON: Enter or upload JSON as a string. This JSON must parse correctly; otherwise, DataRobot won't accept it.
    • YAML: Enter or upload YAML as a string. DataRobot does not validate this YAML.
    Description (Optional) Enter a description of the key value's purpose.

    To copy key values from a previous registered model version, select All categories or a single category, and then click Add to copy the key values:

    If a key value with the same name exists in the newer version and it is not read-only, the value from the older version will overwrite it. Otherwise, a new key value with that name is created in the newer version. If you edit either key value to use a different file, the other key value is unaffected, and the file is no longer shared. System key values are not included in bulk copy; for example, model.version is not overwritten in a newer version with the old version's value.

    To import key values through the API, click Copy to clipboard to copy the Python code snippet for importing key values:

    This is a Python template for adding key values to a registered model version. Replace the implementation of fetch_values() with code that obtains the metrics, tags, parameters, and artifacts to import as key values for the DataRobot model version. You must provide your API key as an environment variable: export MLOPS_API_TOKEN=<value of API token from Developer Tools page>.

  6. Click Add to save the key value. The new key appears in the list for the selected Category.

Add key values for moderation and evaluation guard models

Availability information

Evaluation and moderation guardrails are a premium feature. Contact your DataRobot representative or administrator for information on enabling this feature.

Feature flag: Enable Moderation Guardrails (Premium), Enable Global Models in the Model Registry (Premium), Enable Additional Custom Model Output in Prediction Responses

If a model is intended for use as a custom deployment guard model for text-generation (LLM) model evaluation and moderation, you can define the input and output column names on the registered model version so that each user doesn't need to specify this information when they select the model during evaluation and moderation setup. To do this, define the following key values for custom model versions with a naming convention. Any user with write permissions for the custom model version can define the evaluation and moderation input and output columns. To define the input and output column names, in the Registry > Model directory, open the registered model version you want to deploy (or have deployed) as a guard model. In the Key Values section, expand the Tags panel, and do the following:

  • Click + Add Tag to create a new key value in the Tag category. Set the Value type to String and the Name to moderations.input_column_name. For Value, enter the name of the dataframe column your model reads as input; for example: promptText.

  • Click + Add Tag to create a new key value in the Tag category. Set the Value type to String and the Name to moderations.output_column_name. For Value, enter the name of the dataframe column your model reads as output; for example: sincerity_sincere_PREDICTION.

With this information configured, you can deploy the registered model version, and when the deployment is selected as a Custom Deployment evaluation (in the Playground or Model workshop), the Input column and Output column fields are configured automatically.

Edit or delete key values

To edit or delete added, copied, or imported key values:

  1. On the Model directory page, in the table of registered models, click the registered model containing the version you want to manage key values for, opening the list of versions.

  2. In the list of versions, click the version you want to edit, opening the registered model version panel:

  3. Click the Overview tab, locate the Key values section, and do either of the following:

    • Click Search to locate the key value you want to edit or delete.

    • Browse the Tags, Metrics, Training parameters, Runtime parameters, and Artifacts sections to locate editable key values.

  4. For key values you've created, you can click the edit () and delete () icons:


Updated October 24, 2024