100% Latest & Updated Google Professional Machine Learning Engineer Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 550
An ML engineering team must resolve this production problem: A complex teacher classifier meets quality goals but misses cost and latency budgets for a high-traffic mobile endpoint. Which action best addresses the actual technical requirement?
- Configure supported autoscaling on tested serving capacity metrics and safe min-max replicas.
- Keep minimum warm replicas or compatible provisioned capacity for the required SLO.
- Train a smaller student model using an approved knowledge-distillation objective.
- Use supported streamed token output and safe client rendering.
Explanation
Correct Answer
C:Train a smaller student model using an approved knowledge-distillation objective. Compare student metrics, tail latency and fairness with teacher and baseline. Teacher errors can be transferred to the student through distillation.
Incorrect Answers
A:That measure applies when a fraud-scoring endpoint becomes overloaded during weekday peaks and overprovisioned overnight. The required intervention here is different: Train a smaller student model using an approved knowledge-distillation objective. The need is a complex teacher classifier meets quality goals but misses cost and latency budgets for a high-traffic mobile endpoint.
B:That measure applies when an API is latency-sensitive but its inference container takes 45 seconds to load model weights when scaled from zero. The required intervention here is different: Train a smaller student model using an approved knowledge-distillation objective. The need is a complex teacher classifier meets quality goals but misses cost and latency budgets for a high-traffic mobile endpoint.
D:That measure applies when a conversational app needs users to see partial text quickly but its large model takes several seconds for complete output. The required intervention here is different: Train a smaller student model using an approved knowledge-distillation objective. The need is a complex teacher classifier meets quality goals but misses cost and latency budgets for a high-traffic mobile endpoint.
Question 551
An ML platform team is resolving 2 independent workstreams. Workstream A: Several departments share one foundation model but each has a different small fine-tuned adapter and cannot afford full copies. Workstream B: A recommendation app searches millions of embedding vectors and exact linear scan is too slow for online requests. Select exactly TWO actions, one directly addressing each separate workstream.
- Use a supported multi-adapter serving design if the platform explicitly supports it.
- Use a supported approximate nearest-neighbor vector index and tune recall-latency tradeoff.
- Chunk or summarize source text and choose a model/context strategy meeting the requirement.
- Apply bounded client concurrency, backoff and approved quota planning.
- Train a smaller student model using an approved knowledge-distillation objective.
Explanation
Correct Answers
A:Use a supported multi-adapter serving design if the platform explicitly supports it. Test adapter identity isolation and per-tenant output validity. Unsupported dynamic adapter loading can cause version or memory faults.
B:Use a supported approximate nearest-neighbor vector index and tune recall-latency tradeoff. Benchmark recall@k and p95 latency against an exact reference subset. An approximate index may omit some true nearest neighbors.
Incorrect Answers
C:The approach “Chunk or summarize source text and choose a model/context strategy meeting the requirement” is relevant when a summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping; it does not establish the result required by this workstream.
D:Applying bounded client concurrency, backoff and approved quota planning is relevant when a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads; it does not establish the result required by this workstream.
E:Training a smaller student model using an approved knowledge-distillation objective is relevant when a complex teacher classifier meets quality goals but misses cost and latency budgets for a high-traffic mobile endpoint; it does not establish the result required by this workstream.
Question 552
A Google Cloud change request describes the following issue: A support assistant repeatedly fetches identical public policy passages and spends most latency on retrieval calls. Which action best addresses the actual technical requirement?
- Keep minimum warm replicas or compatible provisioned capacity for the required SLO.
- Use an authorized low-latency online Feature Store representation for serving features.
- Cache approved retrieval results with tenant-scoped keys and explicit invalidation.
- Use a lower-cost right-sized serving instance that passes the p95 target.
Explanation
Correct Answer
C:Cache approved retrieval results with tenant-scoped keys and explicit invalidation. Test freshness after policy revision and ensure tenant permission isolation. Stale cached passages or cross-tenant leakage can cause harmful wrong answers.
Incorrect Answers
A:That measure applies when an API is latency-sensitive but its inference container takes 45 seconds to load model weights when scaled from zero. The required intervention here is different: Cache approved retrieval results with tenant-scoped keys and explicit invalidation. The need is a support assistant repeatedly fetches identical public policy passages and spends most latency on retrieval calls.
B:That measure applies when a deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. The required intervention here is different: Cache approved retrieval results with tenant-scoped keys and explicit invalidation. The need is a support assistant repeatedly fetches identical public policy passages and spends most latency on retrieval calls.
D:That measure applies when a sentence classifier gets 25 requests per second and its GPU endpoint is idle while a CPU instance nearly meets latency targets. The required intervention here is different: Cache approved retrieval results with tenant-scoped keys and explicit invalidation. The need is a support assistant repeatedly fetches identical public policy passages and spends most latency on retrieval calls.
Question 553
An ML platform team is resolving 2 independent workstreams. Workstream A: A conversational app needs users to see partial text quickly but its large model takes several seconds for complete output. Workstream B: A canary variant appears healthy in aggregate but has poor p99 latency because requests are mixed with fast legacy traffic. Select exactly TWO actions, one directly addressing each separate workstream.
- Enable bounded dynamic request batching with latency-aware maximum wait.
- Train a smaller student model using an approved knowledge-distillation objective.
- Create per-variant serving metrics and scaling controls for the candidate.
- Use supported streamed token output and safe client rendering.
- Apply bounded client concurrency, backoff and approved quota planning.
Explanation
Correct Answers
C:Create per-variant serving metrics and scaling controls for the candidate. Trigger a canary-specific latency alarm and verify rollback behavior. Monitoring only aggregate latency can mask small failing canary cohorts.
D:Use supported streamed token output and safe client rendering. Measure time-to-first-token and correct handling of stream interruption. Streaming does not reduce total computation or guarantee safety of partial output.
Incorrect Answers
A:The approach “Enable bounded dynamic request batching with latency-aware maximum wait” is relevant when a transformer endpoint receives many tiny independent requests and GPU execution spends most time on launch overhead; it does not establish the result required by this workstream.
B:Training a smaller student model using an approved knowledge-distillation objective is relevant when a complex teacher classifier meets quality goals but misses cost and latency budgets for a high-traffic mobile endpoint; it does not establish the result required by this workstream.
E:Applying bounded client concurrency, backoff and approved quota planning is relevant when a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads; it does not establish the result required by this workstream.
Question 554
An engineer is choosing an appropriate design for this workload: A summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping. Which action best addresses the actual technical requirement?
- Chunk or summarize source text and choose a model/context strategy meeting the requirement.
- Apply bounded client concurrency, backoff and approved quota planning.
- Cache approved retrieval results with tenant-scoped keys and explicit invalidation.
- Enable bounded dynamic request batching with latency-aware maximum wait.
Explanation
Correct Answer
A:Chunk or summarize source text and choose a model/context strategy meeting the requirement. Check that all required document sections have traceable coverage in outputs. Arbitrary truncation can remove important evidence and bias the summary.
Incorrect Answers
B:That measure applies when a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads. The required intervention here is different: Chunk or summarize source text and choose a model/context strategy meeting the requirement. The need is a summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping.
C:That measure applies when a support assistant repeatedly fetches identical public policy passages and spends most latency on retrieval calls. The required intervention here is different: Chunk or summarize source text and choose a model/context strategy meeting the requirement. The need is a summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping.
D:That measure applies when a transformer endpoint receives many tiny independent requests and GPU execution spends most time on launch overhead. The required intervention here is different: Chunk or summarize source text and choose a model/context strategy meeting the requirement. The need is a summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping.
Question 555
An ML platform team is resolving 2 independent workstreams. Workstream A: A recommendation app searches millions of embedding vectors and exact linear scan is too slow for online requests. Workstream B: A deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. Select exactly TWO actions, one directly addressing each separate workstream.
- Apply bounded client concurrency, backoff and approved quota planning.
- Use a supported approximate nearest-neighbor vector index and tune recall-latency tradeoff.
- Enable bounded dynamic request batching with latency-aware maximum wait.
- Use an authorized low-latency online Feature Store representation for serving features.
- Apply model compression and export an eligible optimized edge format.
Explanation
Correct Answers
B:Use a supported approximate nearest-neighbor vector index and tune recall-latency tradeoff. Benchmark recall@k and p95 latency against an exact reference subset. An approximate index may omit some true nearest neighbors.
D:Use an authorized low-latency online Feature Store representation for serving features. Check p95 lookup latency and match online values to controlled batch updates. An online latest-feature value must not be used retrospectively in historical training.
Incorrect Answers
A:Applying bounded client concurrency, backoff and approved quota planning is relevant when a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads; it does not establish the result required by this workstream.
C:The approach “Enable bounded dynamic request batching with latency-aware maximum wait” is relevant when a transformer endpoint receives many tiny independent requests and GPU execution spends most time on launch overhead; it does not establish the result required by this workstream.
E:Applying model compression and export an eligible optimized edge format is relevant when a local retail device needs image classification under a strict memory limit and cannot store the original 4 GB model; it does not establish the result required by this workstream.
Question 556
An ML engineering team must resolve this production problem: A local retail device needs image classification under a strict memory limit and cannot store the original 4 GB model. Which action best addresses the actual technical requirement?
- Configure supported autoscaling on tested serving capacity metrics and safe min-max replicas.
- Use event-time-aware historical feature retrieval for training and separate current online features for serving.
- Apply model compression and export an eligible optimized edge format.
- Use an authorized low-latency online Feature Store representation for serving features.
Explanation
Correct Answer
C:Apply model compression and export an eligible optimized edge format. Compare cold-start memory and holdout image accuracy on the physical device. Aggressive pruning can reduce recall on difficult image classes.
Incorrect Answers
A:That measure applies when a fraud-scoring endpoint becomes overloaded during weekday peaks and overprovisioned overnight. The required intervention here is different: Apply model compression and export an eligible optimized edge format. The need is a local retail device needs image classification under a strict memory limit and cannot store the original 4 GB model.
B:That measure applies when a churn model trains against live online values and reports unrealistic accuracy on examples from last year. The required intervention here is different: Apply model compression and export an eligible optimized edge format. The need is a local retail device needs image classification under a strict memory limit and cannot store the original 4 GB model.
D:That measure applies when a deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. The required intervention here is different: Apply model compression and export an eligible optimized edge format. The need is a local retail device needs image classification under a strict memory limit and cannot store the original 4 GB model.
Question 557
An ML platform team is resolving 2 independent workstreams. Workstream A: A canary variant appears healthy in aggregate but has poor p99 latency because requests are mixed with fast legacy traffic. Workstream B: A customer application requires authenticated internet access to a low-latency model but the endpoint currently permits anonymous calls. Select exactly TWO actions, one directly addressing each separate workstream.
- Apply model compression and export an eligible optimized edge format.
- Create per-variant serving metrics and scaling controls for the candidate.
- Chunk or summarize source text and choose a model/context strategy meeting the requirement.
- Enable an appropriate supported public access path with authenticated invokers and rate limiting.
- Apply bounded client concurrency, backoff and approved quota planning.
Explanation
Correct Answers
B:Create per-variant serving metrics and scaling controls for the candidate. Trigger a canary-specific latency alarm and verify rollback behavior. Monitoring only aggregate latency can mask small failing canary cohorts.
D:Enable an appropriate supported public access path with authenticated invokers and rate limiting. Verify invalid credentials are rejected and authorized clients meet latency SLO. Network reachability does not imply that requests are authorized or safe.
Incorrect Answers
A:Applying model compression and export an eligible optimized edge format is relevant when a local retail device needs image classification under a strict memory limit and cannot store the original 4 GB model; it does not establish the result required by this workstream.
C:The approach “Chunk or summarize source text and choose a model/context strategy meeting the requirement” is relevant when a summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping; it does not establish the result required by this workstream.
E:Applying bounded client concurrency, backoff and approved quota planning is relevant when a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads; it does not establish the result required by this workstream.
Question 558
A Google Cloud change request describes the following issue: A model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads. Which action best addresses the actual technical requirement?
- Apply bounded client concurrency, backoff and approved quota planning.
- Keep minimum warm replicas or compatible provisioned capacity for the required SLO.
- Chunk or summarize source text and choose a model/context strategy meeting the requirement.
- Use a lower-cost right-sized serving instance that passes the p95 target.
Explanation
Correct Answer
A:Apply bounded client concurrency, backoff and approved quota planning. Load test quota-limited error handling and resulting customer response time. Blind immediate retries can worsen throttling and duplicate side effects.
Incorrect Answers
B:That measure applies when an API is latency-sensitive but its inference container takes 45 seconds to load model weights when scaled from zero. The required intervention here is different: Apply bounded client concurrency, backoff and approved quota planning. The need is a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads.
C:That measure applies when a summarization service receives more tokens than the deployed model can accept, causing errors or silent input clipping. The required intervention here is different: Apply bounded client concurrency, backoff and approved quota planning. The need is a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads.
D:That measure applies when a sentence classifier gets 25 requests per second and its GPU endpoint is idle while a CPU instance nearly meets latency targets. The required intervention here is different: Apply bounded client concurrency, backoff and approved quota planning. The need is a model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads.
Question 559
Before changing the system, the team investigates the following behavior: A deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. Which observation most directly identifies the cause or requirements before a change?
- Inspect online feature store keys, freshness and lookup latency distributions.
- Compare throughput per replica, request queueing and HPA or managed autoscaler behavior.
- Benchmark supported lower-bit quantization against current accuracy and memory use.
- Review region-specific request and token quotas with observed rate spikes.
Explanation
Correct Answer
A:The evidence for this case shows situation: Inspect online feature store keys, freshness and lookup latency distributions. It directly examines the reported situation: A deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. The corresponding corrective action is: Use an authorized low-latency online Feature Store representation for serving features.
Incorrect Answers
B:This evidence concerns a separate issue: A fraud-scoring endpoint becomes overloaded during weekday peaks and overprovisioned overnight. It does not establish what is happening in the present case: A deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. The diagnostic evidence to examine here is Inspect online feature store keys, freshness and lookup latency distributions.
C:This evidence concerns a separate issue: A large language model endpoint has high serving memory cost and modest quality tolerance for a constrained classification task. It does not establish what is happening in the present case: A deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. The diagnostic evidence to examine here is Inspect online feature store keys, freshness and lookup latency distributions.
D:This evidence concerns a separate issue: A model-as-a-service app receives sudden event-driven bursts and begins returning quota errors despite sufficient client threads. It does not establish what is happening in the present case: A deployed recommendation endpoint needs current feature values from multiple sources with response latency below 30 milliseconds. The diagnostic evidence to examine here is Inspect online feature store keys, freshness and lookup latency distributions.
Question 560
An ML platform team is resolving 3 independent workstreams. Workstream A: A churn model trains against live online values and reports unrealistic accuracy on examples from last year. Workstream B: A sentence classifier gets 25 requests per second and its GPU endpoint is idle while a CPU instance nearly meets latency targets. Workstream C: An API is latency-sensitive but its inference container takes 45 seconds to load model weights when scaled from zero. Select exactly THREE diagnostic observations, one directly addressing each separate workstream.
- Inspect ingress configuration, IAM principals and token validation on the selected endpoint.
- Benchmark CPU and GPU with equivalent request sizes and concurrency.
- Compare example timestamps to feature versions available at that moment.
- Measure load time and first-response delay across revision cold starts.
- Inspect vector size, query volume, recall target and index support.
- Review service attachment, PSC/network connectivity and invoker IAM scope.
Explanation
Correct Answers
B:The evidence for this case shows situation: Benchmark CPU and GPU with equivalent request sizes and concurrency. It directly examines the reported situation: A sentence classifier gets 25 requests per second and its GPU endpoint is idle while a CPU instance nearly meets latency targets. The corresponding corrective action is: Use a lower-cost right-sized serving instance that passes the p95 target.
C:The evidence for this case shows situation: Compare example timestamps to feature versions available at that moment. It directly examines the reported situation: A churn model trains against live online values and reports unrealistic accuracy on examples from last year. The corresponding corrective action is: Use event-time-aware historical feature retrieval for training and separate current online features for serving.
D:The evidence for this case shows situation: Measure load time and first-response delay across revision cold starts. It directly examines the reported situation: An API is latency-sensitive but its inference container takes 45 seconds to load model weights when scaled from zero. The corresponding corrective action is: Keep minimum warm replicas or compatible provisioned capacity for the required SLO.
Incorrect Answers
A:Inspecting ingress configuration, IAM principals and token validation on the selected endpoint is relevant when a customer application requires authenticated internet access to a low-latency model but the endpoint currently permits anonymous calls; it does not establish the result required by this workstream.
E:Inspecting vector size, query volume, recall target and index support is relevant when a recommendation app searches millions of embedding vectors and exact linear scan is too slow for online requests; it does not establish the result required by this workstream.
F:Reviewing service attachment, PSC/network connectivity and invoker IAM scope is relevant when a bank's model must be callable only from internal VPC services even though the training dataset lives in a separate project; it does not establish the result required by this workstream.
Professional Machine Learning Engineer Practice Tests