Databricks SDK for JavaScript
    Preparing search index...

    Interface AiRuntimeTask

    AiRuntimeTask: multi-node GPU compute task definition for Databricks AI Runtime workloads.

    Jobs-framework-level concepts (retries, per-task timeout, idempotency token, usage/budget policy, permissions) live on the surrounding TaskSettings / run-submit request and are intentionally NOT duplicated here. Users compose ai_runtime_task with the standard Jobs/DABs task wrapper to get those.

    interface AiRuntimeTask {
        codeSourcePath?: string;
        deployments?: DeploymentSpec[];
        dockerImageUrl?: string;
        experiment?: string;
        mlflowArtifactLocation?: string;
        mlflowExperimentDirectory?: string;
        mlflowRun?: string;
    }
    Index

    Properties

    codeSourcePath? deployments? dockerImageUrl? experiment? mlflowArtifactLocation? mlflowExperimentDirectory? mlflowRun?

    Properties

    codeSourcePath?: string

    Workspace or UC volume path of the code-source archive, unpacked on each node and exposed through $CODE_SOURCE. Set by first-party tooling; not for direct callers.

    deployments?: DeploymentSpec[]

    Deployment specs for this task. Exactly one deployment is currently supported (a single entry where every node runs the same command); this is a current-Preview constraint. Role-split workloads (driver + worker, parameter server, separate eval node, etc.) with multiple entries are the eventual intent but not yet supported.

    dockerImageUrl?: string

    Optional Docker image URL for a custom container image. When set, the task runs on the specified container image instead of the default client image. Format: {organization}/{repository}:{tag}

    experiment?: string

    MLflow experiment name for this run. If an experiment with this name already exists under the calling user, the run is appended to it; otherwise a new experiment is created. To target a specific MLflow storage location (for example, when running as a service principal), set mlflow_experiment_directory.

    mlflowArtifactLocation?: string

    Optional root location for MLflow artifacts logged by the run. If this field isn't specified the default artifact location will be in dbfs i.e. dbfs:/databricks/mlflow-tracking/<experiment_id>/... If dbfs access is restricted or UC is preferred this can be a custom location in UC: dbfs:/Volumes/<catalog>/<schema>/<volume>/... The location should be unique for each experiment.

    mlflowExperimentDirectory?: string

    Optional workspace directory under which the MLflow experiment named in experiment is created. Must start with /Workspace. Set this when running as a service principal that has no default user directory; for regular users the experiment defaults to the user's home directory.

    mlflowRun?: string

    Optional display name for the MLflow run created under experiment. If omitted, MLflow generates a default name.

    Settings

    Member Visibility

    On This Page

    Properties