Model Deployment: Databricks Machine Learning Associate Practice Questions

A batch scoring script constructs JSON requests for a custom model endpoint using pandas rows. The team wants the recommended DataFrame request format that preserves column ordering. Which payload style should they prefer?

1. Ainputs, because every pandas-based model endpoint expects tensors even when the model signature is tabular
2. Binstances, because row-oriented tensor payloads are the default for all MLflow pyfunc table models
3. Cdataframe_split, because Databricks recommends the split-oriented pandas format and it preserves column order
4. Ddataframe_records, because records orientation guarantees column order more strongly than split orientation
5. Epredictions, because endpoint requests should mirror the response wrapper used by Databricks serving APIs

A serving endpoint for a custom model is already deployed. The engineer wants to call it from an application for real-time inference. Which statement is correct?

1. AThe application should send a scoring request to the serving endpoint invocation URL using a supported request format and valid authentication
2. BThe application must query the Unity Catalog model registry directly, because registry versions expose prediction methods over HTTP
3. CThe application must always call the endpoint through Databricks SQL, because REST access is supported only for foundation models
4. DThe application should poll the training run artifacts in MLflow, because deployed endpoints do not accept external prediction requests
5. EThe application must upload a notebook to the endpoint with every request so the serving system can reconstruct the model context

A team wants to roll out a new model gradually while most live traffic still goes to the current stable model. What is the purpose of traffic splitting on serving endpoints?

1. ATo divide a training dataset into folds before cross-validation so each candidate model sees a different validation sample
2. BTo route percentages of live inference requests to different served models or endpoint configurations during rollout
3. CTo partition feature tables across storage locations so online and offline features are read from separate catalogs
4. DTo alternate request payload formats between dataframe_split and tensor inputs for automatic schema discovery
5. ETo divide one large prediction vector into smaller chunks so a single model can score them on different GPUs in parallel

Why might a machine learning engineer deliberately send only a small percentage of production requests to a challenger model while keeping the rest on the champion model?

1. ATo reduce the number of endpoint logs, because Databricks records only the majority route in inference tables
2. BTo compare the challenger under real traffic with limited blast radius before increasing its share or promoting it further
3. CTo avoid registering the challenger in Unity Catalog, because low-traffic models can be served without version control
4. DTo force batch inference behavior from a real-time endpoint, because small traffic shares automatically queue requests offline
5. ETo bypass request authentication for the challenger, because minor traffic percentages are treated as internal evaluation traffic

An operations team receives a steady stream of sensor events and wants model inference embedded inside a managed pipeline that processes arriving records continuously. According to the exam's Databricks deployment pattern, how is streaming inference performed with Delta Live Tables?

1. ABy exporting each micro-batch to a CSV file and scoring it later with a nightly batch notebook
2. BBy applying the model as a Spark UDF inside the Delta Live Tables pipeline so inference runs as the data flows through the pipeline
3. CBy querying the MLflow UI during each trigger interval and copying the latest model outputs into the target table
4. DBy replacing the data pipeline with a SQL warehouse dashboard that refreshes whenever events arrive from upstream systems
5. EBy writing raw events directly to the model registry, because registries can execute inference logic during table ingestion

A team compares streaming inference with nightly batch scoring. What is the main advantage of the streaming design for event-driven use cases?

1. AIt processes new data as it arrives, reducing prediction delay compared with waiting for the next scheduled batch window
2. BIt always costs less than batch inference, because continuous pipelines never maintain checkpoints or state between runs
3. CIt eliminates the need for model packaging, because streaming systems can execute raw notebook cells directly in production
4. DIt guarantees lower latency than a real-time serving endpoint for every per-event request regardless of workload shape
5. EIt makes feature engineering unnecessary, because micro-batches automatically infer model-ready columns from event streams

A custom model deploys successfully in training, but fails during endpoint creation because a helper package is missing from the serving container. What is the most direct fix?

1. AAdd the dependency to the MLflow model environment, such as pip requirements or other declared package dependencies, before redeploying
2. BRename the serving endpoint, because dependency resolution is tied to endpoint names and old names can block installs
3. CDisable signatures on the model, because missing packages occur only when the endpoint validates model schemas during startup
4. DTurn on scale-to-zero, because cold-start rebuilds automatically install packages that were absent from the original model
5. EQuery the endpoint with dataframe_records instead of dataframe_split, because package failures are caused by request shape mismatches

A data scientist wants to score a stored MLflow model against a pandas DataFrame inside Python code, without calling a remote serving endpoint. Which workflow is the standard MLflow approach?

1. ALoad the model with mlflow.pyfunc.load_model and call predict on the pandas DataFrame
2. BConvert the DataFrame to SQL text and send it through a REST invocation URL even for local batch scoring
3. CUse only spark.readStream because pandas objects are not supported inputs for MLflow pyfunc prediction
4. DRegister the DataFrame in Unity Catalog first, because pyfunc models can only score named tables and not in-memory data
5. ECall mlflow.start_run before every prediction row, because pyfunc predict requires an active tracking run for inference