Question 760A tabular model runs cheaply on CPU while a vision transformer needs an accelerator for throughput targets. Which technical mechanism is most relevant to this outcome? A custom container controls dependencies and request handling but inherits security, compatibility and patching duties. Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. Generative inference memory depends on model parameters, KV-cache and concurrent sequence lengths, not only average CPU use. Edge inference trades centralized elasticity for local latency and offline availability with hardware constraints. ExplanationCorrect Answer B:Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. This supports the specific implementation: Benchmark compatible CPU and accelerator inference configurations on representative traffic and select the least-cost compliant option. A suitable independent check is: Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations. Other Answers A:A custom container controls dependencies and request handling but inherits security, compatibility and patching duties. The cause in the stated scenario is different: Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. C:Generative inference memory depends on model parameters, KV-cache and concurrent sequence lengths, not only average CPU use. The cause in the stated scenario is different: Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. D:Edge inference trades centralized elasticity for local latency and offline availability with hardware constraints. The cause in the stated scenario is different: Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium.
Question 761A tabular model runs cheaply on CPU while a vision transformer needs an accelerator for throughput targets. What is the most relevant acceptance check for the proposed ML change? Benchmark on-device accuracy, latency, memory and recovery after network disconnection. Confirm approved services can score requests over expected private paths and unauthorized callers are denied. Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations. Replay the same raw test records through training and serving and compare transformed tensor values and outputs. ExplanationCorrect Answer C:Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations. This tests the underlying failure: Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. Other Answers A:Edge inference trades centralized elasticity for local latency and offline availability with hardware constraints. That does not establish the required outcome; verify it by checking: Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations. B:Endpoint network reachability and cloud identity authorization are separate security boundaries. That does not establish the required outcome; verify it by checking: Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations. D:Training-serving feature skew can arise when model inputs are transformed inconsistently across environments. That does not establish the required outcome; verify it by checking: Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations.
Question 762A tabular model runs cheaply on CPU while a vision transformer needs an accelerator for throughput targets. Which repeatable process control should the ML engineering team adopt? Require a documented latency justification before reserving always-on warm serving resources. Keep container image policies, pod security and service-mesh observability under centralized platform ownership. Review endpoint exposure and identity policies before enabling any new inference consumer. Add hardware re-benchmarking to model-version releases because model architecture changes performance requirements. ExplanationCorrect Answer D:Add hardware re-benchmarking to model-version releases because model architecture changes performance requirements. This protects the selected approach because Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. The immediate solution is: Benchmark compatible CPU and accelerator inference configurations on representative traffic and select the least-cost compliant option. Other Answers A:Warm replicas reduce first-request initialization latency but incur charges while receiving no traffic. The continuing safeguard required here is: Add hardware re-benchmarking to model-version releases because model architecture changes performance requirements. B:Kubernetes-based serving supports flexible sidecars and orchestration while increasing operational maintenance obligations. The continuing safeguard required here is: Add hardware re-benchmarking to model-version releases because model architecture changes performance requirements. C:Endpoint network reachability and cloud identity authorization are separate security boundaries. The continuing safeguard required here is: Add hardware re-benchmarking to model-version releases because model architecture changes performance requirements.
Question 763An organization must justify inference hardware selection against production p95 and cost budgets. Which proposed approach best resolves the described model engineering problem? Benchmark compatible CPU and accelerator inference configurations on representative traffic and select the least-cost compliant option. Deploy a suitably optimized edge model and establish controlled synchronization and rollout mechanisms. Choose serving accelerator memory and concurrency limits based on measured token or batch demand. Package and version inference preprocessing and postprocessing together with the approved model artifact. ExplanationCorrect Answer A:Hardware acceleration is beneficial only when supported operators and workload intensity justify the resource premium. To validate the correction: Compare per-request cost, p95 latency, memory saturation and measured throughput across hardware configurations. Other Answers B:Edge inference trades centralized elasticity for local latency and offline availability with hardware constraints. It does not resolve the specific failure; the targeted correction is: Benchmark compatible CPU and accelerator inference configurations on representative traffic and select the least-cost compliant option. C:Generative inference memory depends on model parameters, KV-cache and concurrent sequence lengths, not only average CPU use. It does not resolve the specific failure; the targeted correction is: Benchmark compatible CPU and accelerator inference configurations on representative traffic and select the least-cost compliant option. D:Training-serving feature skew can arise when model inputs are transformed inconsistently across environments. It does not resolve the specific failure; the targeted correction is: Benchmark compatible CPU and accelerator inference configurations on representative traffic and select the least-cost compliant option.
Question 764A company needs a reproducible pipeline that validates data, trains a classifier, evaluates it and registers an artifact. Which Google Cloud machine-learning design choice best satisfies the stated requirement? Configure a Cloud Build CI pipeline that tests and constructs signed versioned inference images from controlled source commits. Package preprocessing as a versioned pipeline artifact and reference its exact digest in both training and serving components. Add an explicit evaluation and approval component that blocks promotion unless documented acceptance thresholds pass. Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. ExplanationCorrect Answer D:ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. To validate the correction: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. Other Answers A:Continuous integration converts source changes into reproducible artifacts but cannot establish model quality on its own. It does not resolve the specific failure; the targeted correction is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. B:Training-serving skew arises when logically equivalent features are computed with different code or fitted parameters. It does not resolve the specific failure; the targeted correction is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. C:Quality gates compare candidate evidence against policy before triggering a serving deployment. It does not resolve the specific failure; the targeted correction is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks.
Question 765A company needs a reproducible pipeline that validates data, trains a classifier, evaluates it and registers an artifact. Which underlying ML principle makes the proposed approach effective? ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. Training-serving skew arises when logically equivalent features are computed with different code or fitted parameters. Registry governance separates successful training from authorization to expose a model in production. Retraining should be driven by meaningful quality change and valid fresh data, not solely a clock-based assumption. ExplanationCorrect Answer A:ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. This supports the specific implementation: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. A suitable independent check is: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. Other Answers B:Training-serving skew arises when logically equivalent features are computed with different code or fitted parameters. The cause in the stated scenario is different: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. C:Registry governance separates successful training from authorization to expose a model in production. The cause in the stated scenario is different: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. D:Retraining should be driven by meaningful quality change and valid fresh data, not solely a clock-based assumption. The cause in the stated scenario is different: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently.
Question 766A company needs a reproducible pipeline that validates data, trains a classifier, evaluates it and registers an artifact. What is the most relevant acceptance check for the proposed ML change? Trigger a DAG and inspect correct dependency ordering, retries and success/failure propagation across ML tasks. Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. Check worker scale-out, completed task outputs, retry semantics and resource limits under a representative test load. Verify build logs, test status, image digest and artifact provenance for a known commit. ExplanationCorrect Answer B:Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. This tests the underlying failure: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. Other Answers A:Airflow manages DAG scheduling and dependencies across many services but does not remove task-level model evaluation needs. That does not establish the required outcome; verify it by checking: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. C:Ray coordinates distributed Python tasks and actors while pipeline orchestration tracks broader workflow dependencies. That does not establish the required outcome; verify it by checking: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. D:Continuous integration converts source changes into reproducible artifacts but cannot establish model quality on its own. That does not establish the required outcome; verify it by checking: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry.
Question 767A company needs a reproducible pipeline that validates data, trains a classifier, evaluates it and registers an artifact. Which ongoing standard would make this result reliable for future model versions? Version allowed data contracts and monitor recurring validation failures before relaxing any acceptance thresholds. Document outcome delay and dataset freeze periods in continuous-training controls. Require integration tests for parallel dependency changes and version each component interface. Require version-controlled pipeline definitions and explicit promotion gates for artifact quality and source integrity. ExplanationCorrect Answer D:Require version-controlled pipeline definitions and explicit promotion gates for artifact quality and source integrity. This protects the selected approach because ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. The immediate solution is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. Other Answers A:Automated data validation prevents invalid distributions and schema changes from silently contaminating downstream models. The continuing safeguard required here is: Require version-controlled pipeline definitions and explicit promotion gates for artifact quality and source integrity. B:Delayed labels make recent training rows incomplete and can introduce leakage or misleading performance estimates. The continuing safeguard required here is: Require version-controlled pipeline definitions and explicit promotion gates for artifact quality and source integrity. C:Directed acyclic pipeline graphs can run independent steps concurrently while waiting at synchronization dependencies. The continuing safeguard required here is: Require version-controlled pipeline definitions and explicit promotion gates for artifact quality and source integrity.
Question 768A fraud model currently depends on manual notebook steps and the business needs a governed end-to-end release workflow. What should the team change to address the specific technical requirement? Make pipeline steps idempotent using run identifiers, atomic outputs and explicit retry-safe artifact publication. Choose a supported Ray-on-Agent-Platform workflow when distributed task execution requirements warrant Ray. Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. Implement staged model delivery that requires registry approval, evaluation pass and guarded rollout before full traffic. ExplanationCorrect Answer C:ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. To validate the correction: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. Other Answers A:At-least-once workflow execution can repeat steps, so side effects need idempotency safeguards. It does not resolve the specific failure; the targeted correction is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. B:Ray coordinates distributed Python tasks and actors while pipeline orchestration tracks broader workflow dependencies. It does not resolve the specific failure; the targeted correction is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. D:ML continuous delivery joins software deployment controls with model-specific validation and traffic management. It does not resolve the specific failure; the targeted correction is: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks.
Question 769A fraud model currently depends on manual notebook steps and the business needs a governed end-to-end release workflow. Which underlying ML principle makes the proposed approach effective? Airflow manages DAG scheduling and dependencies across many services but does not remove task-level model evaluation needs. ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. Delayed labels make recent training rows incomplete and can introduce leakage or misleading performance estimates. Registry governance separates successful training from authorization to expose a model in production. ExplanationCorrect Answer B:ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. This supports the specific implementation: Build a versioned Agent Platform Pipeline with modular data checks, training, evaluation and registry tasks. A suitable independent check is: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. Other Answers A:Airflow manages DAG scheduling and dependencies across many services but does not remove task-level model evaluation needs. The cause in the stated scenario is different: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. C:Delayed labels make recent training rows incomplete and can introduce leakage or misleading performance estimates. The cause in the stated scenario is different: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. D:Registry governance separates successful training from authorization to expose a model in production. The cause in the stated scenario is different: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently.
Question 770A fraud model currently depends on manual notebook steps and the business needs a governed end-to-end release workflow. What controlled validation most directly confirms the approach worked? Inject a retry after a partial write and confirm output counts and model registry versions remain correct. Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. Test that a deliberately inferior candidate remains staged while an approved candidate obtains a versioned release. Trigger a DAG and inspect correct dependency ordering, retries and success/failure propagation across ML tasks. ExplanationCorrect Answer B:Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. This tests the underlying failure: ML pipelines encode execution dependencies and artifacts so the same documented process can run consistently. Other Answers A:At-least-once workflow execution can repeat steps, so side effects need idempotency safeguards. That does not establish the required outcome; verify it by checking: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. C:Continuous training creates candidate versions; deployment decisions need independent performance and safety evidence. That does not establish the required outcome; verify it by checking: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry. D:Airflow manages DAG scheduling and dependencies across many services but does not remove task-level model evaluation needs. That does not establish the required outcome; verify it by checking: Trigger the pipeline with a fixed dataset and inspect component states, lineage, model metrics and generated registry entry.