databricks-cli/bundle/tests/model_serving_endpoint/databricks.yml

resources:
  model_serving_endpoints:
    my_model_serving_endpoint:
      name: "my-endpoint"
      config:
        served_models:
          - model_name: "model-name"
            model_version: "1"
            workload_size: "Small"
            scale_to_zero_enabled: true
        traffic_config:
          routes:
            - served_model_name: "model-name-1"
              traffic_percentage: 100
      permissions:
        - level: CAN_QUERY
          group_name: users

targets:
  development:
    mode: development
    resources:
      model_serving_endpoints:
        my_model_serving_endpoint:
          name: "my-dev-endpoint"

  staging:
    resources:
      model_serving_endpoints:
        my_model_serving_endpoint:
          name: "my-staging-endpoint"

  production:
    mode: production
    resources:
      model_serving_endpoints:
        my_model_serving_endpoint:
          name: "my-prod-endpoint"