Databricks-ML-Assoc Model Deployment — 20 Questions

Question 1

Which TWO actions are necessary when migrating a model from the MLflow Model Registry to Unity Catalog?

Accepted Answer

Use the registry URI format 'models:/catalog.schema.model_name'. Migrating to Unity Catalog requires using the correct namespace and ensuring models are registered under a catalog and schema. This transition is essential for centralized governance, lineage tracking, and simplified access control across workspaces. By using the Unity Catalog-compatible model registry, organizations achieve unified management of their machine learning assets, allowing for more robust security policies and easier cross-workspace model sharing.

Answer

Re-train the model on a GPU cluster

Answer

Manually copy the MLflow artifacts to an external S3 bucket

Answer

Convert the model to an ONNX format

Question 2

Refer to the exhibit. A user attempts to deploy this endpoint configuration, but it fails. What is the most likely cause?

Accepted Answer

The model version does not exist in the registry. The exhibit shows a configuration that includes 'scale_to_zero_enabled' set to true, which is valid, but the deployment failure often stems from a missing 'compute_config' or invalid 'workload_type'. In Databricks Model Serving, every model definition must explicitly declare the compute type. If the infrastructure does not support the requested configuration or if the model version does not exist, the API will reject the request.

Answer

The model name is too short

Answer

Scale-to-zero is not supported for production endpoints

Answer

The configuration lacks an environment variable block

Question 3

When evaluating a model for production deployment, which TWO metrics or artifacts are essential to ensure the model is 'ready' for the business?

Accepted Answer

The full set of training data snapshots. Ensuring a model is production-ready involves more than just a high accuracy score. Documentation of the training data lineage ensures that the model is built on reliable, compliant sources. Furthermore, verifying that the model produces consistent outputs across various edge cases is critical. Together, these steps minimize risk, ensure compliance, and build confidence that the model will behave as expected in a dynamic production environment.

Answer

A list of the developer's favorite colors

Answer

The number of hours spent training the model

Answer

A manual email to the stakeholder confirming completion

Question 4

A data scientist needs to deploy a MLflow model to a production environment. The model requires specific external libraries not present in the default Databricks Runtime. Which deployment method ensures the exact environment is replicated?

Accepted Answer

Include a conda.yaml file with the model artifact and use MLflow Model Servicing.. Using MLflow Model Servicing with a custom container or specifying dependencies via conda.yaml ensures the runtime environment matches the training environment exactly. This is critical for preventing 'works on my machine' issues where library version mismatches lead to inference errors. By containerizing the environment, the model gains portability and reliability, which are foundational requirements for production-grade machine learning pipelines within the Databricks ecosystem and managed model serving infrastructures.

Answer

Install libraries manually on the driver node using pip install.

Answer

Create a custom Docker image and specify it in the model signature.

Answer

Deploy the model as a library using the Databricks Workspace UI.

Question 5

Which TWO actions should be taken to ensure an MLflow model is ready for production deployment via Model Registry?

Accepted Answer

Transition the model version stage to 'Production'.. Transitioning a model to production requires validating both its performance metrics and its operational readiness. By registering the model and transitioning its stage, you establish a clear lineage of what is considered production-ready. This process is vital for governance, as it prevents unverified code from reaching end-users while ensuring that all stakeholders have access to the specific version history and quality metrics associated with the current production model.

Answer

Add a descriptive tag to the model version.

Answer

Delete all previous versions of the model.

Answer

Export the model as a CSV file.

Question 6

Refer to the exhibit. A team is configuring a Databricks Serving Endpoint. What is the implication of setting 'min_instances' to 2?

Accepted Answer

The endpoint is guaranteed to have two active pods running at all times.. Setting 'min_instances' to 2 ensures high availability by maintaining two active instances at all times. This is essential for production workloads where zero downtime is required during scale events or node failures. By guaranteeing a minimum capacity, the system avoids the 'cold start' latency that occurs when an endpoint must scale up from zero, ensuring consistent response times for critical applications requiring immediate inference results.

Answer

The endpoint will automatically scale down to zero during off-peak hours.

Answer

The endpoint will limit concurrency to two requests per instance.

Answer

The endpoint will only support two concurrent users at any given time.

Question 7

Which THREE features are provided by Databricks Model Serving for production inference?

Accepted Answer

Automatic scaling based on request volume.. Databricks Model Serving provides a comprehensive set of features designed to simplify the transition from experimentation to production. These features collectively ensure that models remain performant, available, and observable. By leveraging native platform tools, data scientists can focus on model quality while the platform handles the complexities of infrastructure management, scaling, and integration with existing enterprise security and monitoring frameworks, which is essential for scalable MLOps.

Answer

Automatic conversion of Python code to C++.

Answer

Direct write access to the production database.

Question 8

Which artifact is essential for enabling MLflow's 'Model Signature' feature in a deployed model?

Accepted Answer

An MLmodel file.. The model signature defines the expected schema for the model's input and output. By defining this, you allow Databricks to enforce strict validation on incoming requests. This is essential for debugging and preventing malformed inputs from causing runtime errors. Having a well-defined signature is a best practice that ensures the model is self-documenting and compatible with automated deployment tools, simplifying the integration process for consuming applications.

Answer

A Dockerfile.

Answer

A requirements.txt file.

Answer

A Jupyter notebook checkpoint.

Question 9

A data science team is preparing to deploy an MLflow model to a Databricks Model Serving endpoint and needs to ensure low-latency inference with automatic scaling. Which TWO configurations or practices should they implement? (Choose 2)

Accepted Answer

Configure provisioned concurrency to eliminate cold start latency for traffic spikes.. Configuring provisioned concurrency prevents cold starts by maintaining warm container instances ready to serve incoming requests immediately. Additionally, specifying explicit input examples and output signatures in MLflow allows the serving endpoint to validate payloads, optimize payload serialization, and catch runtime errors early before requests hit the model scoring function.

Answer

Attach an interactive all-purpose cluster to the endpoint to process batch inference jobs concurrently.

Answer

Store model training datasets directly inside the endpoint configuration JSON payload.

Answer

Rely entirely on default CPU node types without specifying workload size parameters.

Question 10

A machine learning engineer has registered a scikit-learn model in Unity Catalog as `prod.ml.risk_model` and wants a REST endpoint that automatically serves the newest model version whenever a new version is registered, without editing the endpoint config each time. Which configuration should they use when creating the Databricks Model Serving endpoint?

Accepted Answer

A served entity whose entity_name is `prod.ml.risk_model` and entity_version is left unset so the endpoint tracks the latest version.. A Databricks Model Serving endpoint can track a registered model without pinning a version. Leaving entity_version unset makes the endpoint resolve the newest registered version at load time, so newly registered versions become serveable without manual endpoint edits. Pinning an integer version freezes the endpoint, and artifact URIs or the literal `latest` are not valid entity identifiers.

Answer

A served entity whose entity_name is the full artifact URI of the model run and whose entity_version is the run ID.

Answer

A served entity whose entity_name is `prod.ml.risk_model` and entity_version is set to the string `latest`.

Answer

A served entity whose entity_name is `prod.ml.risk_model` and entity_version is set to the integer value of the currently newest version.

Question 11

A data scientist has trained a model and registered it in the Databricks Model Registry. They now want to deploy it to a Databricks Model Serving endpoint for real-time inference. Which of the following is a prerequisite for creating a serving endpoint for a registered model?

Accepted Answer

The model must be registered in the MLflow Model Registry.. Creating a serving endpoint requires the model to be registered in the MLflow Model Registry. This allows the endpoint to reference a specific model version and manage its lifecycle. While other factors like signature and stage are beneficial, they are not mandatory. The Model Registry provides the necessary metadata and version control for deployment.

Answer

The model must be logged with a specific flavor, such as scikit-learn.

Answer

The model must have a signature defined.

Answer

The model must be in the 'Production' stage.

Question 12

A retail data science team has trained a scikit-learn model and logged it with MLflow. They now want Databricks Model Serving to automatically pick up new model versions as they are registered in Unity Catalog, without editing the endpoint each time. Which serving endpoint configuration achieves this?

Accepted Answer

Serve the model using its Unity Catalog name and a version alias such as 'Champion'.. Serving a Unity Catalog registered model through an alias lets the endpoint decouple the deployed URL from a specific version number. Reassigning the alias to a newer version in the registry is enough for subsequent scoring requests to use the new artifact. This supports promotion workflows where a candidate model is validated and then swapped into the serving alias without downtime or endpoint reconfiguration.

Answer

Attach the model as a workspace file dependency and load it inside a custom pyfunc wrapper.

Answer

Enable automatic model refresh by setting the endpoint's scale-to-zero policy to disabled.

Answer

Deploy the endpoint using the model's run ID instead of the model name.

Question 13

A machine learning engineer has registered a model in Unity Catalog as `prod.ml.risk_model`. The model was trained with a signature that expects a single DOUBLE column named `features`. The engineer wants to serve this model with Databricks Model Serving and needs to know the exact JSON payload format to send to the endpoint's REST API. Which payload correctly matches the model signature?

Accepted Answer

{"dataframe_records": [{"features": 0.5}]}. Databricks Model Serving accepts two JSON payload shapes for custom MLflow models: dataframe_records, a list of objects keyed by the signature's input column names, and dataframe_split, a columns/data pair. Because the model signature names a single input column features, the payload must supply that key. The dataframe_records form with {"features": 0.5} satisfies the signature, passes schema enforcement, and returns predictions in the matching response format.

Answer

{"dataframe_records": [[0.5]]}

Answer

{"dataframe_split": {"columns": ["features"], "data": [[0.5]]}}

Answer

{"inputs": [{"features": 0.5}], "parameters": {}}

Question 14

A data science team is deploying an MLflow model to a Databricks Model Serving endpoint that must handle bursty production traffic with low latency. They want to avoid cold-start delays while controlling cost during idle periods. Which two configuration choices should they make? (Choose two.)

Accepted Answer

Set `min_instances` to a value greater than zero on the served entity.. Keeping at least one replica warm via a nonzero `min_instances` removes cold-start latency for bursty traffic, and matching `workload_size` to the model's resource needs ensures each replica can serve requests quickly. Scale-to-zero and lifecycle stages do not address latency, and inference tables are for logging rather than performance.

Answer

Register the model version with a stage of `Production` in the workspace Model Registry.

Answer

Enable inference tables on the endpoint.

Answer

Set `scale_to_zero_enabled` to true on the served entity.

Question 15

A data scientist has registered a scikit-learn model in Unity Catalog as `ml_prod.churn.model_v3` and wants Databricks Model Serving to automatically pick up future versions of that model without editing the endpoint. The endpoint must serve the newest ready version at all times. Which model URI should be used as the served entity?

Accepted Answer

models:/ml_prod.churn.model_v3. Omitting the version and alias from a Unity Catalog model URI makes the reference resolve to the latest ready version of the registered model, which is exactly what an endpoint needs when it must always serve the newest approved artifact. Version-pinned and alias-based URIs both require a deliberate action to advance, and run-relative URIs never track registry state at all.

Answer

models:/ml_prod.churn.model_v3@champion

Answer

runs:/<run_id>/model

Answer

models:/ml_prod.churn.model_v3/3

Question 16

A machine learning engineer is preparing to deploy a custom MLflow model to a Databricks Model Serving endpoint. The model uses a third-party Python library that is not part of the Databricks Runtime ML base environment. The engineer wants to ensure the endpoint can load and run the model successfully. Which two actions are required? (Choose two.)

Accepted Answer

Log the model with the required library captured in the model's conda or pip requirements.. Serving a custom model with a non-standard library requires two things: the library must be captured in the model's conda or pip requirements at log time so the serving container installs it, and the model must be registered in Unity Catalog so Model Serving can reference the version and its environment metadata. Cluster-level installs, init scripts, and runtime selection do not propagate into the managed serving environment.

Answer

Install the third-party library on the driver node of the cluster used for training.

Answer

Configure the endpoint to use a Databricks Runtime version that includes the third-party library.

Answer

Add the third-party library to the cluster's init script so it loads at startup.

Question 17

A machine learning engineer has an MLflow model registered in Unity Catalog and wants to create a Databricks Model Serving endpoint for real-time inference. The endpoint must be able to scale down to zero when idle to minimize cost, while still handling bursts of traffic. Which configuration should the engineer choose?

Accepted Answer

Set `scale_to_zero_enabled` to true and configure a `workload_size` appropriate for the expected burst.. Databricks Model Serving endpoints can scale to zero when idle, eliminating compute cost between requests, and scale up when traffic arrives. Enabling `scale_to_zero_enabled` satisfies the cost goal, while selecting an appropriate `workload_size` ensures the endpoint has enough compute per instance to handle bursts. Together they deliver cost-efficient, responsive real-time inference.

Answer

Enable autoscaling on the endpoint by setting `max_instances` to a high value and leave `scale_to_zero_enabled` at its default of false.

Answer

Deploy the model to a single-node cluster and expose it via a REST API using a Databricks job schedule.

Answer

Set `min_instances` to 0 and rely on the default `workload_size` to absorb traffic bursts without further configuration.

Question 18

A data scientist needs to deploy a model to Databricks Model Serving. Which component is strictly required to be logged in MLflow to enable the 'Model Serving' feature?

Accepted Answer

The model signature defining input and output schema. Databricks Model Serving requires a model artifact to be logged with a signature. The signature defines the expected input schema, which allows the serving endpoint to validate incoming requests. Without a signature, the model cannot be correctly parsed by the serving infrastructure, preventing the deployment from starting. This ensures that the production inference endpoint operates with expected data formats, maintaining reliability in downstream applications.

Answer

The model's training accuracy metrics

Answer

A Unity Catalog registered function

Answer

A dedicated high-concurrency cluster

Question 19

When deploying a model using Model Serving, how does Databricks ensure that the environment remains consistent between the training workspace and the serving environment?

Accepted Answer

By capturing the model's dependencies during log_model. Databricks uses Conda or virtual environment dependencies captured during the MLflow logging process. When the model is logged, the environment details, including library versions, are saved. During deployment, the serving infrastructure creates a container that replicates this environment. This practice is crucial for avoiding 'dependency hell', where models fail in production due to subtle library version mismatches compared to the original training environment.

Answer

By requiring the user to provide a Dockerfile

Answer

By strictly enforcing the use of the latest stable libraries

Answer

By running the model on the same training cluster

Question 20

Which strategy is most effective for managing model drift in a production Databricks environment?

Accepted Answer

Log inference data to Delta tables for analysis. Managing model drift requires active monitoring of input data and output predictions. Setting up Databricks Model Serving with feature tables allows for logging of inference requests and responses to a Delta table. By comparing these production distributions against the training data using tools like Great Expectations or custom SQL queries, teams can detect shifts in data patterns that necessitate retraining or model adjustments to maintain performance.

Answer

Manually check the model accuracy once a year

Answer

Re-deploy the model with new data every hour

Answer

Use a static model that never changes