Databricks SDK for JavaScript
    Preparing search index...

    Interface GenAiComputeTask

    DEPRECATED — use AiRuntimeTask for all new BYOT multi-node GPU workloads (see ai_runtime_task.proto). AiRuntimeTask is the only supported BYOT task type for new workloads; this proto is retained only for AIR CLI (fka SGCLI) pywheel backwards compatibility and will be removed once the pywheel → databricks-cli migration completes (post- PuPr).

    interface GenAiComputeTask {
        command?: string;
        compute?: ComputeConfig;
        dlRuntimeImage?: string;
        mlflowExperimentName?: string;
        source?: Source;
        trainingScriptPath?: string;
        yamlParameters?: string;
        yamlParametersFilePath?: string;
    }
    Index

    Properties

    command? compute? dlRuntimeImage? mlflowExperimentName? source? trainingScriptPath? yamlParameters? yamlParametersFilePath?

    Properties

    command?: string

    Command launcher to run the actual script, e.g. bash, python etc.

    compute?: ComputeConfig
    dlRuntimeImage?: string

    Runtime image

    mlflowExperimentName?: string

    Optional string containing the name of the MLflow experiment to log the run to. If name is not found, backend will create the mlflow experiment using the name.

    source?: Source

    Optional location type of the training script. When set to WORKSPACE, the script will be retrieved from the local workspace. When set to GIT, the script will be retrieved from a Git repository defined in git_source. If the value is empty, the task will use GIT if git_source is defined and WORKSPACE otherwise.

    • WORKSPACE: Script is located in workspace.
    • GIT: Script is located in cloud Git provider.
    trainingScriptPath?: string

    The training script file path to be executed. Cloud file URIs (such as dbfs:/, s3:/, adls:/, gcs:/) and workspace paths are supported. For python files stored in the workspace, the path must be absolute and begin with /. For files stored in a remote repository, the path must be relative. This field is required.

    yamlParameters?: string

    Optional string containing model parameters passed to the training script in yaml format. If present, then the content in yaml_parameters_file_path will be ignored.

    yamlParametersFilePath?: string

    Optional path to a YAML file containing model parameters passed to the training script.

    Settings

    Member Visibility

    On This Page

    Properties