Google Professional Machine Learning Engineer Practice Test 73
Question 720
While preparing a production training workflow on Gemini Enterprise Agent Platform, the following condition occurs: A custom training image expects a local CSV under the developer's home directory but managed training supplies data through configured Cloud Storage paths. Engineers need an auditable, resource-aware correction to the issue. Which TWO selections form an appropriate before-and-after verification pair? Select TWO.
- The training manifest contains traceable permitted sources and safety tests detect known prohibited outputs.
- A clean managed training run reads the intended versioned dataset without notebook-local path assumptions.
- The chosen threshold detects clinically relevant positives while the false-alarm rate is documented for deployment review.
- Inspect gradient ranges, loss-scaling logs, overflow events and the affected operations with test batches.
- Inspect task input configuration, container logs, expected file patterns and staged object paths.
Explanation
Correct Answers
B:A clean managed training run reads the intended versioned dataset without notebook-local path assumptions. This is specific acceptance evidence for custom training input schema mismatch. Notebook filesystem locations are not portable contracts for managed container training jobs. It confirms the controlled correction: Adapt the training program to the managed data input contract or stage files to an explicit configurable path.
E:Inspect task input configuration, container logs, expected file patterns and staged object paths. This inspects evidence relevant to custom training input schema mismatch. Notebook filesystem locations are not portable contracts for managed container training jobs. After identifying the cause, the matching correction is: Adapt the training program to the managed data input contract or stage files to an explicit configurable path.
Incorrect Answers
A:The training manifest contains traceable permitted sources and safety tests detect known prohibited outputs. This relates to fine-tuning image generator data rights but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect task input configuration, container logs, expected file patterns and staged object paths. The specific acceptance result is: A clean managed training run reads the intended versioned dataset without notebook-local path assumptions.
C:The chosen threshold detects clinically relevant positives while the false-alarm rate is documented for deployment review. This relates to cost-sensitive threshold for rare positives but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect task input configuration, container logs, expected file patterns and staged object paths. The specific acceptance result is: A clean managed training run reads the intended versioned dataset without notebook-local path assumptions.
D:Inspect gradient ranges, loss-scaling logs, overflow events and the affected operations with test batches. This relates to mixed precision overflow diagnosis but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect task input configuration, container logs, expected file patterns and staged object paths. The specific acceptance result is: A clean managed training run reads the intended versioned dataset without notebook-local path assumptions.
Question 721
As the team tests a new governed AI release against adversarial and normal data, the following condition occurs: An assistant summarizes third-party documents and malicious attachments can contain hidden exfiltration instructions, with no content screening before model invocation. The fix must retain legitimate model usefulness and appropriate controls. What acceptance result verifies the intended behavior after remediation?
- A controlled prompt-injection fixture is blocked or neutralized without disrupting ordinary allowed documents.
- An evaluation fixture correctly associates each matured label with exactly one prior prediction.
- Review documentation correctly distinguishes model influence from proven real-world causal effects.
- Test prompts preserve support utility without exposing real account identifiers to unauthorized model requests.
Explanation
Correct Answer
A:A controlled prompt-injection fixture is blocked or neutralized without disrupting ordinary allowed documents. This is specific acceptance evidence for model armor untrusted content screening. Screening helps detect malicious instructions or sensitive content crossing the model trust boundary. It confirms the controlled correction: Add appropriate Model Armor screening to untrusted inputs and outputs and reject or quarantine policy-violating content. This matters here because as the team tests a new governed AI release against adversarial and normal data, the following condition occurs: An….
Incorrect Answers
B:An evaluation fixture correctly associates each matured label with exactly one prior prediction. This result could demonstrate success for ground truth association for quality monitoring but not for the stated model armor untrusted content screening issue. The direct test is: A controlled prompt-injection fixture is blocked or neutralized without disrupting ordinary allowed documents.
C:Review documentation correctly distinguishes model influence from proven real-world causal effects. This result could demonstrate success for explainability versus causal claim but not for the stated model armor untrusted content screening issue. The direct test is: A controlled prompt-injection fixture is blocked or neutralized without disrupting ordinary allowed documents.
D:Test prompts preserve support utility without exposing real account identifiers to unauthorized model requests. This result could demonstrate success for sensitive identifiers leaked to llm prompts but not for the stated model armor untrusted content screening issue. The direct test is: A controlled prompt-injection fixture is blocked or neutralized without disrupting ordinary allowed documents. This matters here because as the team tests a new governed AI release against adversarial and normal data, the following condition occurs: An….
Question 722
During a staged restoration test at a secondary serving deployment, the following condition occurs: A new model revision passes deployment health checks but produces many application-level 5xx errors after traffic increases; rollback was never configured. The team must prove an actual user-visible improvement without weakening access controls. Which technical explanation best accounts for the reported behavior?
- Batch workloads need throughput and scheduling guarantees, whereas real-time endpoints serve interactive low-latency calls.
- Container readiness checks alone do not guarantee successful predictions under real traffic and data distributions.
- Unlimited retries amplify overload; backpressure and controlled concurrency are needed along with scaling.
- Embedding matrices interpret token IDs according to one vocabulary; mismatched preprocessing changes input meaning.
Explanation
Correct Answer
B:Container readiness checks alone do not guarantee successful predictions under real traffic and data distributions. This explains the mechanism behind model traffic rollback on high error rate and why the described failure is possible. A discriminating investigation is: Inspect endpoint revision metrics, release alarms, traffic fractions and availability of the prior known-good model. The corresponding repair is: Add guarded traffic shifts and automatic rollback to the retained healthy revision when threshold alarms trip.
Incorrect Answers
A:Batch workloads need throughput and scheduling guarantees, whereas real-time endpoints serve interactive low-latency calls. This principle describes batch versus online inference schedule, which is not the cause of the observed model traffic rollback on high error rate failure. The controlling explanation is: Container readiness checks alone do not guarantee successful predictions under real traffic and data distributions.
C:Unlimited retries amplify overload; backpressure and controlled concurrency are needed along with scaling. This principle describes predictive serving rate limit and backpressure, which is not the cause of the observed model traffic rollback on high error rate failure. The controlling explanation is: Container readiness checks alone do not guarantee successful predictions under real traffic and data distributions.
D:Embedding matrices interpret token IDs according to one vocabulary; mismatched preprocessing changes input meaning. This principle describes inference preprocessing tokenization mismatch, which is not the cause of the observed model traffic rollback on high error rate failure. The controlling explanation is: Container readiness checks alone do not guarantee successful predictions under real traffic and data distributions.
Question 723
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: An analyst runs a long-lived custom Apache Beam streaming service to compute a simple daily SQL join and aggregate over tables already in BigQuery. Collaborators must avoid expanding data access or breaking experiment reproducibility. Which underlying engineering principle explains why the original approach fails?
- Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment.
- Arrival time and event time are different; event-time windows with lateness rules preserve business-time meaning.
- Notebook execution can retain stale objects; a clean deterministic run detects hidden state dependencies.
- Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing.
Explanation
Correct Answer
D:Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing. This explains the mechanism behind bigquery versus dataflow preparation selection and why the described failure is possible. A discriminating investigation is: Examine source locations, batch cadence, SQL transform complexity, streaming latency requirements and operational burden. The corresponding repair is: Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
Incorrect Answers
A:Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment. This principle describes feature store online and offline entity alignment, which is not the cause of the observed bigquery versus dataflow preparation selection failure. The controlling explanation is: Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing.
B:Arrival time and event time are different; event-time windows with lateness rules preserve business-time meaning. This principle describes dataflow event-time window correctness, which is not the cause of the observed bigquery versus dataflow preparation selection failure. The controlling explanation is: Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing.
C:Notebook execution can retain stale objects; a clean deterministic run detects hidden state dependencies. This principle describes notebook transient state and execution order, which is not the cause of the observed bigquery versus dataflow preparation selection failure. The controlling explanation is: Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing.
Question 724
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: An analyst runs a long-lived custom Apache Beam streaming service to compute a simple daily SQL join and aggregate over tables already in BigQuery. Collaborators must avoid expanding data access or breaking experiment reproducibility. What acceptance result verifies the intended behavior after remediation?
- The scheduled query produces equivalent daily features with less service orchestration and within the run window.
- A held-out clean sample preserves meaningful ranges while invalid spikes trigger visible data-quality exceptions.
- Each candidate's reported metrics resolve to its actual artifact and replayable configuration without manual lookups.
- The equivalent daily query scans only the eligible partitions and returns the same number of matched entities.
Explanation
Correct Answer
A:The scheduled query produces equivalent daily features with less service orchestration and within the run window. This is specific acceptance evidence for bigquery versus dataflow preparation selection. Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing. It confirms the controlled correction: Implement the batch aggregate in scheduled BigQuery SQL when the required transformation is expressible and fits its SLA.
Incorrect Answers
B:A held-out clean sample preserves meaningful ranges while invalid spikes trigger visible data-quality exceptions. This result could demonstrate success for outlier handling before feature standardization but not for the stated bigquery versus dataflow preparation selection issue. The direct test is: The scheduled query produces equivalent daily features with less service orchestration and within the run window.
C:Each candidate's reported metrics resolve to its actual artifact and replayable configuration without manual lookups. This result could demonstrate success for pipeline experiment versus notebook-only tracking but not for the stated bigquery versus dataflow preparation selection issue. The direct test is: The scheduled query produces equivalent daily features with less service orchestration and within the run window.
D:The equivalent daily query scans only the eligible partitions and returns the same number of matched entities. This result could demonstrate success for bigquery partition filter cost optimization but not for the stated bigquery versus dataflow preparation selection issue. The direct test is: The scheduled query produces equivalent daily features with less service orchestration and within the run window.
Question 725
Within a jointly maintained Google Cloud data science workspace, the following condition occurs: A daily feature query scans two years of sales even though it uses only yesterday's records because the filter wraps the partition date in an incompatible expression. Collaborators must avoid expanding data access or breaking experiment reproducibility. Which statement identifies the primary reason for this failure?
- Collaborative notebooks execute with real identities; editor convenience does not justify unrestricted production-data access.
- Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme.
- LLM-based scoring can be sensitive to phrasing, order and model preferences; evaluation needs explicit calibration.
- A scaler learned from corrupted or held-out values can distort typical feature magnitudes and leak evaluation information.
Explanation
Correct Answer
B:Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme. This explains the mechanism behind bigquery partition filter cost optimization and why the described failure is possible. A discriminating investigation is: Examine the partitioning column, query plan, bytes scanned and predicate form against the table's partition definition. The corresponding repair is: Rewrite the filter to permit effective partition pruning while preserving its requested date boundaries.
Incorrect Answers
A:Collaborative notebooks execute with real identities; editor convenience does not justify unrestricted production-data access. This principle describes colab enterprise notebook data access scope, which is not the cause of the observed bigquery partition filter cost optimization failure. The controlling explanation is: Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme.
C:LLM-based scoring can be sensitive to phrasing, order and model preferences; evaluation needs explicit calibration. This principle describes llm-as-a-judge rubric reliability, which is not the cause of the observed bigquery partition filter cost optimization failure. The controlling explanation is: Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme.
D:A scaler learned from corrupted or held-out values can distort typical feature magnitudes and leak evaluation information. This principle describes outlier handling before feature standardization, which is not the cause of the observed bigquery partition filter cost optimization failure. The controlling explanation is: Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme.
Question 726
At a Google Cloud analytics team preparing a governed pilot, the following condition occurs: A policy assistant's answers become stale every week when procedures change, and engineers repeatedly fine-tune the base model to incorporate each revision. The team must satisfy a reproducible low-code AI assessment. Which outcome most directly demonstrates a successful, measurable fix?
- A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling.
- Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall.
- The scheduled report processes the full authorized partition and stores traceable predictions without a local export.
- A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
Explanation
Correct Answer
D:A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version. This is specific acceptance evidence for foundation model tuning versus retrieval update. Frequently changing external knowledge is better supplied through current context than repeatedly embedded into model weights. It confirms the controlled correction: Index versioned policy sources and use retrieval-grounded Gemini responses for changing facts, reserving tuning for justified behavior adaptation. Here, the relevant situation is: the following condition occurs: A policy assistant's answers become stale every week when procedures change, and….
Incorrect Answers
A:A held-out invoice batch produces correct structured fields with measurable per-field precision and reject handling. This result could demonstrate success for document ai structured invoice extraction but not for the stated foundation model tuning versus retrieval update issue. The direct test is: A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
B:Classification F1 remains above the acceptance threshold while end-to-end latency and cost per thousand requests fall. This result could demonstrate success for gemini model choice for latency versus complexity but not for the stated foundation model tuning versus retrieval update issue. The direct test is: A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
C:The scheduled report processes the full authorized partition and stores traceable predictions without a local export. This result could demonstrate success for bigquery ml sql batch prediction integration but not for the stated foundation model tuning versus retrieval update issue. The direct test is: A revised policy becomes answerable after document indexing without retraining, and retrieved citations match the active version.
Question 727
During a controlled replay of the data and model handoff to a second engineering team, the following condition occurs: A credit service looks up a feature by account number online but the training table keys records by customer identifier, causing inconsistent histories. The handoff must preserve privacy, lineage and testable behavior. Which statement identifies the primary reason for this failure?
- MLOps provenance requires a recorded chain from code and data to model and deployment, not merely a friendly name.
- Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment.
- Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing.
- Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme.
Explanation
Correct Answer
B:Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment. This explains the mechanism behind feature store online and offline entity alignment and why the described failure is possible. A discriminating investigation is: Compare entity IDs, key types, online feature-view definitions, offline joins and records for the same verified customer. The corresponding repair is: Standardize stable entity identifiers and validated feature definitions across training and online serving pathways.
Incorrect Answers
A:MLOps provenance requires a recorded chain from code and data to model and deployment, not merely a friendly name. This principle describes ml metadata lineage for production model, which is not the cause of the observed feature store online and offline entity alignment failure. The controlling explanation is: Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment.
C:Choosing a managed SQL engine for warehouse-resident batch joins avoids unnecessary always-on stream processing. This principle describes bigquery versus dataflow preparation selection, which is not the cause of the observed feature store online and offline entity alignment failure. The controlling explanation is: Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment.
D:Partition pruning avoids irrelevant I/O when predicates match the storage partitioning scheme. This principle describes bigquery partition filter cost optimization, which is not the cause of the observed feature store online and offline entity alignment failure. The controlling explanation is: Feature stores rely on consistent entity keys; incompatible identifiers lead to silent training-serving misalignment.
Question 728
During a controlled replay of the data and model handoff to a second engineering team, the following condition occurs: A Colab Enterprise notebook crashes when an engineer loads a multi-terabyte BigQuery table into one pandas DataFrame to compute simple counts. The handoff must preserve privacy, lineage and testable behavior. Which TWO complementary choices jointly establish the evidence and appropriate correction? Select TWO.
- Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training.
- Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table.
- Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL.
- Inspect Python runtime versions, pip dependency trees, package hashes and notebook execution metadata for both runs.
- Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment.
Explanation
Correct Answers
B:Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table. This is the targeted corrective response for notebook memory exhaustion on full data extraction. Single-machine notebook memory is limited; warehouse operations should execute near large warehouse-resident data. Verify the diagnosis using: Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL. Confirm success using: The exploration completes within notebook memory while its aggregate statistics match trusted BigQuery queries. This matters here because during a controlled replay of the data and model handoff to a second engineering team, the following condition….
C:Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL. This inspects evidence relevant to notebook memory exhaustion on full data extraction. Single-machine notebook memory is limited; warehouse operations should execute near large warehouse-resident data. After identifying the cause, the matching correction is: Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table. This matters here because during a controlled replay of the data and model handoff to a second engineering team, the following condition….
Incorrect Answers
A:Choose a compatible Model Garden model/runtime or perform a validated checkpoint conversion before training. This is an action or diagnostic relevant to model garden jax accelerator compatibility, not one of the required two steps. In this scenario verify: Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL. Then correct: Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table.
D:Inspect Python runtime versions, pip dependency trees, package hashes and notebook execution metadata for both runs. This is an action or diagnostic relevant to notebook dependency lock reproducibility, not one of the required two steps. In this scenario verify: Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL. Then correct: Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table.
E:Rebuild offline training data with point-in-time eligible joins and remove features available only after each prediction moment. This is an action or diagnostic relevant to point-in-time feature leakage prevention, not one of the required two steps. In this scenario verify: Inspect query result sizes, runtime memory, pandas allocations and which aggregations can run in BigQuery SQL. Then correct: Push aggregations into BigQuery or stream bounded samples to the notebook rather than materializing the whole table.
Question 729
During a regulated MLOps pipeline deployment, the following condition occurs: A production workflow starts model registration before validation and evaluation steps have finished, occasionally promoting untested weights. The workflow must be repeatable and protect the last approved production model. Which TWO complementary choices jointly establish the evidence and appropriate correction? Select TWO.
- Retain the last healthy model and require canary validation before switching all prediction traffic.
- Add explicit upstream validation and evaluation dependencies before registration and deployment components.
- Package the fitted preprocessing component with the model and apply the same validated transform in both paths.
- Inspect component definitions, pipeline package versions, parameters and source revisions used in both environments.
- Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs.
Explanation
Correct Answers
B:Add explicit upstream validation and evaluation dependencies before registration and deployment components. This is the targeted corrective response for pipeline component dependency order. Orchestration needs explicit data and control dependencies; notebook execution order is not a production DAG. Verify the diagnosis using: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Confirm success using: Every production model version traces to successfully completed validation and evaluation steps in that run.
E:Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. This inspects evidence relevant to pipeline component dependency order. Orchestration needs explicit data and control dependencies; notebook execution order is not a production DAG. After identifying the cause, the matching correction is: Add explicit upstream validation and evaluation dependencies before registration and deployment components.
Incorrect Answers
A:Retain the last healthy model and require canary validation before switching all prediction traffic. This is an action or diagnostic relevant to continuous training rollback protection, not one of the required two steps. In this scenario verify: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Then correct: Add explicit upstream validation and evaluation dependencies before registration and deployment components.
C:Package the fitted preprocessing component with the model and apply the same validated transform in both paths. This is an action or diagnostic relevant to training serving preprocessing artifact sharing, not one of the required two steps. In this scenario verify: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Then correct: Add explicit upstream validation and evaluation dependencies before registration and deployment components.
D:Inspect component definitions, pipeline package versions, parameters and source revisions used in both environments. This is an action or diagnostic relevant to pipeline environment parameter promotion, not one of the required two steps. In this scenario verify: Inspect the pipeline DAG, dependency declarations, artifact references and execution ordering from recent runs. Then correct: Add explicit upstream validation and evaluation dependencies before registration and deployment components.
Question 730
While reproducing a training orchestration incident in isolated staging, the following condition occurs: An orchestrated data pipeline inserts training rows twice after transient failure because a retried task appends without a stable batch identifier. The team requires an idempotent pipeline behavior and traceable evidence. Which TWO selections form an appropriate before-and-after verification pair? Select TWO.
- Inspect Airflow retries, write mode, batch IDs, target partitions and commit logs.
- Inspect task lineage, retries, durable checkpoints, object-store pressure and failed worker logs.
- A deliberately incompatible payload prevents the release and a compatible fixture completes an end-to-end prediction.
- Running the same task twice produces one logical batch and preserves prior validated partitions.
- The task completes within approved resources with predictable cost and produces the same number of embeddings.
Explanation
Correct Answers
A:Inspect Airflow retries, write mode, batch IDs, target partitions and commit logs. This inspects evidence relevant to airflow task idempotency on retry. Orchestrators often retry tasks; duplicate side effects are prevented by idempotent data operations. After identifying the cause, the matching correction is: Make the load idempotent with keyed merge or overwrite-per-version semantics and safe retry handling.
D:Running the same task twice produces one logical batch and preserves prior validated partitions. This is specific acceptance evidence for airflow task idempotency on retry. Orchestrators often retry tasks; duplicate side effects are prevented by idempotent data operations. It confirms the controlled correction: Make the load idempotent with keyed merge or overwrite-per-version semantics and safe retry handling.
Incorrect Answers
B:Inspect task lineage, retries, durable checkpoints, object-store pressure and failed worker logs. This relates to ray distributed task failure isolation but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect Airflow retries, write mode, batch IDs, target partitions and commit logs. The specific acceptance result is: Running the same task twice produces one logical batch and preserves prior validated partitions.
C:A deliberately incompatible payload prevents the release and a compatible fixture completes an end-to-end prediction. This relates to continuous delivery integration contract tests but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect Airflow retries, write mode, batch IDs, target partitions and commit logs. The specific acceptance result is: Running the same task twice produces one logical batch and preserves prior validated partitions.
E:The task completes within approved resources with predictable cost and produces the same number of embeddings. This relates to custom kfp component resource limits but does not establish the requested before-and-after evidence. The actual diagnostic is: Inspect Airflow retries, write mode, batch IDs, target partitions and commit logs. The specific acceptance result is: Running the same task twice produces one logical batch and preserves prior validated partitions.