Databricks-ML-Pro Model Deployment Practice Test 3 — 15 Questions

Question 1

An ML engineer is deploying a model to Databricks Model Serving that requires a custom Python package. The package is not available on PyPI but is stored as a wheel file in Unity Catalog volumes. What is the correct way to include this package in the model's environment?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Reference the wheel file in the conda.yaml using an absolute path to the Unity Catalog volume, such as /Volumes/catalog/schema/volume/package.whl.. Databricks Model Serving builds the environment based on the model's conda.yaml. To install a custom wheel stored in Unity Catalog, you must provide the absolute path to the wheel file in the pip section of conda.yaml. This allows the serving infrastructure to access and install the package from the volume.

Answer

Upload the wheel file to DBFS and reference it in the model's conda.yaml using a dbfs:// path.

Answer

Convert the wheel file to a notebook and install it using %pip in the model's code.

Answer

Include the wheel file in the model's artifacts and add a relative path to it in the conda.yaml pip section.

Question 2

A fraud detection team is deploying a new model to Databricks Model Serving. They want to route a small percentage of live traffic to the new model while keeping the existing model as the primary, and they need to monitor both models' performance. Which TWO configurations are required to achieve this? (Choose two.)

Accepted Answer

Configure the serving endpoint with multiple served entities, each referencing a different model version, and assign traffic percentages.. To route a percentage of traffic to a new model and monitor both, you must configure the endpoint with multiple served entities and assign traffic percentages. Additionally, enabling inference logging via `auto_capture_config` captures request and response data for analysis. These two configurations together allow controlled rollout and performance comparison. The other options describe unsupported workarounds or irrelevant settings.

Answer

Use MLflow's `transition_model_version_stage` to set the new model to 'Staging' and the old model to 'Production'.

Answer

Set the endpoint's workload size to 'Large' to handle the increased load from two models.

Answer

Create a separate endpoint for the new model and use a load balancer to distribute requests.

Question 3

An ML engineer is deploying a model to Databricks Model Serving that requires a custom Python package. The package is not available on PyPI but is stored as a wheel file in Unity Catalog volumes. What is the correct way to include this package in the model's environment?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Use the Databricks Model Serving API to specify the wheel file as an additional library in the served entity configuration.. The correct approach is to specify the wheel file as an additional library in the served entity configuration via the Databricks Model Serving API. This explicitly installs the custom library in the serving environment. Other methods like conda YAML or PYTHONPATH do not reliably install the library for serving.

Answer

Include the wheel file in the MLflow model artifact and add it to the model's requirements.txt.

Answer

Add the wheel file path to the model's conda environment YAML file under the pip section.

Answer

Set the environment variable PYTHONPATH to the Unity Catalog Volume path in the endpoint configuration.

Question 4

A machine learning engineer is deploying a model to Databricks Model Serving that requires a custom Python library. The library is not available in the default Databricks Runtime. The engineer needs to ensure the library is available at inference time. Which TWO methods can be used to include the custom library? (Choose two.)

Accepted Answer

Package the library as a wheel file and include it in the model artifacts when logging the model.. To include a custom Python library for a model served on Databricks Model Serving, you must ensure the library is part of the model's environment. This can be done by adding it to the conda.yaml or requirements.txt when logging the model, or by packaging it as a wheel file and including it in the model artifacts. Both methods ensure the library is installed when the serving environment is created.

Answer

Use the `%pip install` magic command in a notebook before deploying the model.

Answer

Configure the serving endpoint to use a custom Docker image that already has the library installed.

Answer

Install the library on the driver node of the serving cluster using an init script.

Question 5

A data scientist has registered a model in Unity Catalog and wants to deploy it to Databricks Model Serving. The model version is currently in 'Staging' stage. What must the data scientist do before the model can be served?

Accepted Answer

Grant the serving endpoint's service principal USE CATALOG, USE SCHEMA, and EXECUTE on the model.. When deploying a model from Unity Catalog to Databricks Model Serving, the endpoint's service principal must have the appropriate Unity Catalog privileges: USE CATALOG and USE SCHEMA on the containing catalog and schema, and EXECUTE on the model itself. These permissions allow the serving infrastructure to read the model artifacts. Without them, the deployment will fail with a permission error. Stages are not used in Unity Catalog, so aliases or version numbers are used instead.

Answer

Set the model version's stage to 'Production' in the MLflow UI.

Answer

Transition the model version to 'Production' stage.

Answer

Ensure the model version is the latest version in the registry.

Question 6

A team is deploying a model to Databricks Model Serving and wants to implement A/B testing between two model versions. They need to route a percentage of traffic to each version and collect metrics. Which TWO configurations are required to achieve this? (Choose two.)

Accepted Answer

Create a single endpoint with two served models, each referencing a different model version, and set traffic percentages in the served model configuration.. To implement A/B testing with Databricks Model Serving, you create an endpoint with two served models, each pointing to a different model version, and assign traffic percentages. You also enable inference tables to capture request and response data for analysis. These two configurations provide the necessary traffic control and data collection for evaluating model performance.

Answer

Configure a separate endpoint for each model version and use an external load balancer to split traffic.

Answer

Use MLflow's 'transition_model_version_stage' to assign one version to 'Staging' and the other to 'Production'.

Answer

Set the 'scale_to_zero_enabled' flag to true for both served models to ensure cost efficiency during the A/B test.

Question 7

A machine learning engineer is deploying a model to Databricks Model Serving and wants to ensure that the endpoint automatically scales down to zero when there is no traffic, to minimize costs. Which configuration setting should they enable?

Accepted Answer

Set `scale_to_zero_enabled` to `true` in the endpoint's configuration.. Scale-to-zero is enabled by setting `scale_to_zero_enabled` to `true` in the endpoint configuration. This allows the endpoint to shut down completely when idle, reducing costs. Other options reference non-existent parameters or confuse workload size with scaling behavior. It is important to note that scale-to-zero may introduce cold-start latency when traffic resumes.

Answer

Configure the endpoint with a `min_replica` count of 0 and a `max_replica` count of 1.

Answer

Use a `Small` workload size, which automatically scales to zero when idle.

Answer

Enable `auto_scaling` and set the `idle_timeout` to 0 seconds.

Question 8

An ML engineer is deploying a model to Databricks Model Serving that requires a custom Python package. The package is not available on PyPI but is stored as a wheel file in Unity Catalog volumes. What is the correct way to include this package in the model's environment?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Optimize the model's inference function to reduce memory usage, such as by using batch inference or reducing the model's precision.. The most cost-effective solution is to optimize the model's inference function to reduce its memory footprint. This could involve techniques such as quantization, using a smaller model, or processing data in smaller chunks. By reducing memory usage, the model can run within the existing 'Large' workload type, avoiding the need to scale up to a more expensive workload type.

Answer

Configure the endpoint to use a GPU-accelerated workload type, as GPUs have more memory.

Answer

Enable autoscaling on the endpoint to distribute the memory load across multiple instances.

Answer

Increase the endpoint's workload type to 'Extra Large' to get more memory.

Question 9

A team is using Databricks Model Serving to serve a model that requires a custom Python library. They have packaged the library as a wheel file and uploaded it to Unity Catalog volumes. How can they ensure the library is installed when the model is served?

Accepted Answer

Include the wheel file path in the `pip_requirements` field of the MLflow model's conda environment, and ensure the path is accessible from the serving environment.. Custom Python libraries should be included as dependencies in the MLflow model's conda environment, typically via a pip requirements list that references the wheel file. Databricks Model Serving then installs these dependencies when loading the model. Other options incorrectly assume that notebook installs or environment variables affect the serving environment, or reference non-existent configuration options.

Answer

Upload the wheel file to the Databricks workspace and use the `%pip install` magic command in a notebook before deploying the model.

Answer

Set the environment variable `PYTHONPATH` to include the volume path in the endpoint configuration.

Answer

Use the `--extra-index-url` option in the endpoint's pip configuration to point to the volume path.

Question 10

An ML engineer needs a serving endpoint that routes traffic between two model versions, sending 90 percent of requests to version 3 and 10 percent to version 4, while guaranteeing that any single client's consecutive requests during a session land on the same version. Which Databricks Model Serving capability should be used?

Accepted Answer

Configure the endpoint with a traffic split across the two served model versions and enable sticky sessions so that a client is consistently routed to the same version.. A single Model Serving endpoint can host multiple served model versions with an explicit percentage traffic configuration, and the platform's session-based routing keeps a given client on one version across consecutive calls. Together these give the controlled rollout with per-session consistency, without any external load balancer.

Answer

Create two separate endpoints and place a client-side load balancer in front of them that hashes the client IP address to pick an endpoint.

Answer

Register both versions under the same Unity Catalog alias and let the alias resolution randomly select a version on each request.

Answer

Use a single served entity with the 'scale to zero' option enabled, which naturally distributes requests evenly and preserves session affinity.

Question 11

A team deploys a model to a Databricks Model Serving endpoint and enables Inference Tables. They later notice that requests are being served successfully but no rows are appearing in the inference table. Which explanation is most likely?

Accepted Answer

The endpoint was created without the required Unity Catalog permissions on the inference table's parent schema, so payload logging silently fails.. Inference Tables are Delta tables in Unity Catalog, so the endpoint needs create and modify permissions on the target schema and catalog. When those grants are missing, payload writes fail quietly while the model still returns predictions, producing exactly the observed symptom of successful responses with an empty log.

Answer

The model was logged with an MLflow signature, which prevents Inference Tables from recording the request payload for privacy reasons.

Answer

Inference Tables only capture rows for endpoints configured with a Large workload size, so a smaller endpoint will not log anything.

Answer

Inference Tables require the endpoint to be restarted after every batch of requests before rows are flushed to Delta.

Question 12

A machine learning engineer needs to deploy a custom scikit-learn model to a Databricks Model Serving endpoint with a strict response time SLA of under 50 milliseconds. The model includes an extensive text-cleaning pipeline that utilizes heavy regex matching. How should the engineer package the model to ensure maximum inference efficiency and meet the low-latency requirement?

Accepted Answer

Implement a custom MLflow PyFunc model where both the text preprocessing and the scikit-learn predictor are encapsulated inside the predict method.. Integrating the preprocessing directly into the MLflow PyFunc wrapper ensures that the input transformation runs within the optimized inference container memory space, avoiding out-of-band network calls and reducing serialization overhead. This architectural pattern prevents latency bottlenecks often introduced by separate preprocessing microservices, satisfying strict production SLAs.

Answer

Store the preprocessing logic in a Delta table and have the serving endpoint query the table asynchronously during the scoring request.

Answer

Develop a separate Azure or AWS Lambda function to handle the text cleaning before forwarding the request payload to the model endpoint.

Answer

Deploy the scikit-learn model natively without custom wrapper code and require client applications to execute the regex cleaning logic locally.

Question 13

A machine learning engineer needs to deploy a custom PyTorch model to a Databricks Model Serving endpoint. The model requires a custom pre-processing step that is not part of the standard MLflow transformers or pyfunc flavor. Which deployment approach ensures the custom logic executes reliably within the serverless serving container?

Accepted Answer

Implement a custom MLflow pyfunc PythonModel subclass that encapsulates both the PyTorch model and the custom pre-processing transformations, then log it with MLflow.. Packaging the custom pre-processing logic directly into the MLflow pyfunc model artifact by overriding the predict context ensures that all required transformations travel with the model weights. This guarantees identical execution behavior between local testing and production serverless endpoints without relying on external pipeline code.

Answer

Register the vanilla PyTorch state dict and apply pre-processing transformations inside the client application before sending HTTP payloads.

Answer

Define a separate Spark UDF inside the serving endpoint configuration file to intercept incoming JSON batches.

Answer

Store the pre-processing code in a separate volume and reference its absolute file path in the model serving endpoint environment variables.

Question 14

A data science team is preparing to deploy a high-throughput recommendation model using Databricks Model Serving. Which TWO factors must be considered to optimize endpoint latency and resource utilization? (Choose two)

Accepted Answer

Configuring the minimum and maximum number of concurrent scaling units based on expected query traffic spikes.. Selecting appropriate instance types with GPU acceleration and configuring autoscaling bounds based on traffic patterns directly dictate inference latency and operational cost. These considerations are critical in production because underprovisioned endpoints cause timeout errors while overprovisioned endpoints waste cloud computing resources unnecessarily.

Answer

Converting all feature lookup tables from Delta Lake format directly into local CSV files inside the serving container.

Answer

Enabling Apache Spark adaptive query execution on the serving endpoint cluster to optimize joins during scoring.

Answer

Ensuring the MLflow model uses pickle serialization exclusively because it outperforms all other serialization formats.

Question 15

A machine learning engineer needs to deploy a custom PyTorch model to a Databricks Model Serving endpoint. The model requires custom post-processing logic and loading auxiliary tokenizer files alongside the serialized weights. Which approach provides the correct mechanism to package and serve this custom artifact?

Accepted Answer

Subclass mlflow.pyfunc.PythonModel, implement the load_context and predict methods, bundle the tokenizer files in artifacts, and log via mlflow.pyfunc.log_model.. Custom PyTorch models requiring custom code and auxiliary files must be logged using MLflow's pyfunc flavor with custom artifact dependencies. This allows packaging artifacts and arbitrary Python code safely so that the serving infrastructure can instantiate the pyfunc wrapper, execute custom tokenization, and perform required post-processing cleanly during real-time inference.

Answer

Log the model as a standard torch.jit artifact without metadata and point the serving endpoint directly to the raw .pt file in Unity Catalog.

Answer

Create an external FastAPI application outside Databricks, wrap the model, and route requests via a custom reverse proxy.

Answer

Store the tokenizer files in a Delta table and configure the serving endpoint to query the table on every incoming inference request.