Microsoft AI-300 Practice Test 69
Question 680
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A new Azure ML endpoint deployment is created for a new model but traffic is mistakenly set to 100% immediately, exposing all users before canary evaluation. The resolution must preserve quality gates, documented artifacts and rollback options. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Align the published endpoint contract and client payload mapping, then deploy the reviewed scoring code and test representative requests.
- Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features.
- Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment.
- Apply a controlled small traffic allocation to the candidate, monitor quality and latency, then gradually increase only after passing gates.
- Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results.
Explanation
Correct Answers
C:Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment. This is the most relevant evidence for canary rollout sends all traffic to unvalidated candidate. Managed online endpoints can route traffic among deployments; safe releases require explicit percentages and stop criteria. The matching controlled remediation is: Apply a controlled small traffic allocation to the candidate, monitor quality and latency, then gradually increase only after passing gates. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A new Azure ML….
D:Apply a controlled small traffic allocation to the candidate, monitor quality and latency, then gradually increase only after passing gates. This targets canary rollout sends all traffic to unvalidated candidate. Managed online endpoints can route traffic among deployments; safe releases require explicit percentages and stop criteria. Verify the failure through: Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment. Validate success with: Canary traffic follows the approved fraction and rollback to the old deployment is possible when health thresholds fail. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A new Azure ML….
Incorrect Answers
A:Align the published endpoint contract and client payload mapping, then deploy the reviewed scoring code and test representative requests. This is associated with endpoint scoring script expects wrong json schema, not the essential evidence-and-fix pair for canary rollout sends all traffic to unvalidated candidate. First examine: Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment. Then correct: Apply a controlled small traffic allocation to the candidate, monitor quality and latency, then gradually increase only after passing gates.
B:Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features. This is associated with automated ml experiment uses leakage feature, not the essential evidence-and-fix pair for canary rollout sends all traffic to unvalidated candidate. First examine: Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment. Then correct: Apply a controlled small traffic allocation to the candidate, monitor quality and latency, then gradually increase only after passing gates.
E:Review deployment container logs, init() traceback, model mount path, environment image and readiness/liveness probe results. This is associated with endpoint deployment health probe fails before model initialization, not the essential evidence-and-fix pair for canary rollout sends all traffic to unvalidated candidate. First examine: Inspect online endpoint traffic weights, deployment versions, request tracing, rollout approval and available prior stable deployment. Then correct: Apply a controlled small traffic allocation to the candidate, monitor quality and latency, then gradually increase only after passing gates.
Question 681
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: An engineer promotes the model with the lowest training loss even though the held-out validation loss worsens, indicating overfitting. The resolution must preserve quality gates, documented artifacts and rollback options. Which TWO complementary choices identify the issue and correct it? Select TWO.
- Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold.
- Align the published endpoint contract and client payload mapping, then deploy the reviewed scoring code and test representative requests.
- Configure the framework's appropriate distributed data sampler and synchronize epoch handling and metrics aggregation.
- Inspect class prevalence, AutoML primary metric, confusion matrix, recall and precision of candidate runs and business error costs.
- Review training and validation learning curves, selection rule, dataset splits and registered model candidate metrics.
Explanation
Correct Answers
A:Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold. This targets new candidate chosen from training instead of validation results. A training objective measures fit on observed examples; generalization claims require independent validation and selection rules. Verify the failure through: Review training and validation learning curves, selection rule, dataset splits and registered model candidate metrics. Validate success with: The selected version meets validation criteria without depending on apparent fit to the training examples alone. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: An engineer….
E:Review training and validation learning curves, selection rule, dataset splits and registered model candidate metrics. This is the most relevant evidence for new candidate chosen from training instead of validation results. A training objective measures fit on observed examples; generalization claims require independent validation and selection rules. The matching controlled remediation is: Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: An engineer….
Incorrect Answers
B:Align the published endpoint contract and client payload mapping, then deploy the reviewed scoring code and test representative requests. This is associated with endpoint scoring script expects wrong json schema, not the essential evidence-and-fix pair for new candidate chosen from training instead of validation results. First examine: Review training and validation learning curves, selection rule, dataset splits and registered model candidate metrics. Then correct: Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: An engineer….
C:Configure the framework's appropriate distributed data sampler and synchronize epoch handling and metrics aggregation. This is associated with distributed training uses one shard for every worker, not the essential evidence-and-fix pair for new candidate chosen from training instead of validation results. First examine: Review training and validation learning curves, selection rule, dataset splits and registered model candidate metrics. Then correct: Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold.
D:Inspect class prevalence, AutoML primary metric, confusion matrix, recall and precision of candidate runs and business error costs. This is associated with automl primary metric does not match rare-event goal, not the essential evidence-and-fix pair for new candidate chosen from training instead of validation results. First examine: Review training and validation learning curves, selection rule, dataset splits and registered model candidate metrics. Then correct: Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold.
Question 682
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: An Azure ML online endpoint reports HTTP 400/500 errors because the deployed scoring code expects a named features field but the calling service sends a raw nested array. The operator must safeguard training reproducibility and serving continuity. Which observable postchange result proves the intended remediation succeeded?
- Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption.
- Deployment reaches healthy readiness and handles a representative test request without restart loops.
- Canary traffic follows the approved fraction and rollback to the old deployment is possible when health thresholds fail.
- A historical notebook rerun uses the recorded snapshot and yields traceable feature statistics.
Explanation
Correct Answer
A:Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption. Endpoint infrastructure can be healthy while the caller and scoring script disagree on their data contract. a direct corrective action is to align the published endpoint contract and client payload mapping, then deploy the reviewed scoring code and test representative requests.
Incorrect Answers
B:Deployment reaches healthy readiness and handles a representative test request without restart loops. That addresses a different configuration issue rather than the main requirement. Verify instead: Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption.
C:Canary traffic follows the approved fraction and rollback to the old deployment is possible when health thresholds fail. That addresses a different configuration issue rather than the main requirement. Verify instead: Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption.
D:A historical notebook rerun uses the recorded snapshot and yields traceable feature statistics. That addresses a different configuration issue rather than the main requirement. Verify instead: Valid payloads return the correct prediction shape and malformed requests return documented validation errors rather than silent feature corruption.
Question 683
During a Foundry generative AI quality and observability incident, the following condition occurs: A retrieval assistant generates fluent answers citing unrelated documents, yet the release dashboard declares success based solely on readability. The fix must preserve evaluation validity and avoid unnecessary user-data collection. What approved correction resolves the failure without modifying unrelated healthy services?
- Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
- Make required quality/safety criteria explicit CI gates and stop deployment when evaluated scores fail the approved tolerances.
- Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers.
- Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution.
Explanation
Correct Answer
A:Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency. This targets groundedness metric conflated with response fluency. Readability and factual support are distinct dimensions of GenAI output quality; polished prose is not evidence that retrieved sources justify a claim. Verify the failure through: Inspect evaluator definitions, retrieved source contexts, response citations, groundedness scoring and sampled human reference judgments. Validate success with: Unsupported claims fail the release gate even when the answers are grammatically smooth.
Incorrect Answers
B:Make required quality/safety criteria explicit CI gates and stop deployment when evaluated scores fail the approved tolerances. This could address evaluation workflow fails to block degraded agent deployment, but it does not solve the specified groundedness metric conflated with response fluency failure. The matching remediation is: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
C:Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers. This could address agent traces omit downstream tool failure reason, but it does not solve the specified groundedness metric conflated with response fluency failure. The matching remediation is: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
D:Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution. This could address content safety evaluation omits prompt injection tests, but it does not solve the specified groundedness metric conflated with response fluency failure. The matching remediation is: Add context-groundedness and relevance evaluation on representative answers alongside coherence and fluency.
Question 684
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A reusable feature-engineering component emits a URI folder while a downstream training step declares a tabular MLTable input, causing type validation to fail. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema.
- Compare environment image digest, Conda or pip dependencies, Docker build logs, locally resolved versions and job import traceback.
- Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow.
- Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error.
- Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates.
Explanation
Correct Answers
A:Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema. This targets pipeline component output type mismatches consumer. Azure ML pipeline components have explicit typed ports; connecting incompatible artifact interfaces does not automatically transform the data. Verify the failure through: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. Validate success with: The pipeline resolves component dependencies and the training step reads the expected typed asset without implicit file guessing.
D:Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. This is the most relevant evidence for pipeline component output type mismatches consumer. Azure ML pipeline components have explicit typed ports; connecting incompatible artifact interfaces does not automatically transform the data. The matching controlled remediation is: Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema.
Incorrect Answers
B:Compare environment image digest, Conda or pip dependencies, Docker build logs, locally resolved versions and job import traceback. This is associated with training environment lacks pinned package versions, not the essential evidence-and-fix pair for pipeline component output type mismatches consumer. First examine: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. Then correct: Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema.
C:Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow. This is associated with production training script commit cannot be traced, not the essential evidence-and-fix pair for pipeline component output type mismatches consumer. First examine: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. Then correct: Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema.
E:Pin the pipeline input to an approved versioned data asset and advance versions through controlled evaluation gates. This is associated with versionless dataset reference breaks reproducibility, not the essential evidence-and-fix pair for pipeline component output type mismatches consumer. First examine: Compare both component input/output interface schemas, YAML bindings, produced artifacts and the specific Azure ML pipeline validation error. Then correct: Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema.
Question 685
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A notebook computes features from a mutable SQL export every time it runs, but the submitted model job cannot reproduce yesterday's feature distribution. The operator must safeguard training reproducibility and serving continuity. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment.
- Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata.
- Inspect distributed job settings, node count, rendezvous endpoint configuration, compute network security rules and worker log timeouts.
- Add suitable data quality and schema drift signals with thresholds for missingness alongside distribution drift.
- Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs.
Explanation
Correct Answers
B:Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata. This is the most relevant evidence for exploration notebook reads production data without reproducible snapshot. Notebook experimentation is flexible but reliable model comparison requires stable, recorded data inputs rather than live mutable queries. The matching controlled remediation is: Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: A notebook….
E:Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs. This targets exploration notebook reads production data without reproducible snapshot. Notebook experimentation is flexible but reliable model comparison requires stable, recorded data inputs rather than live mutable queries. Verify the failure through: Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata. Validate success with: A historical notebook rerun uses the recorded snapshot and yields traceable feature statistics. This matters here because during a production Azure Machine Learning model lifecycle incident, the following condition occurs: A notebook….
Incorrect Answers
A:Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment. This is associated with mlflow experiment logs omit run parameters, not the essential evidence-and-fix pair for exploration notebook reads production data without reproducible snapshot. First examine: Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata. Then correct: Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs.
C:Inspect distributed job settings, node count, rendezvous endpoint configuration, compute network security rules and worker log timeouts. This is associated with distributed job workers cannot communicate over required network path, not the essential evidence-and-fix pair for exploration notebook reads production data without reproducible snapshot. First examine: Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata. Then correct: Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs.
D:Add suitable data quality and schema drift signals with thresholds for missingness alongside distribution drift. This is associated with missing feature values escape input drift checks, not the essential evidence-and-fix pair for exploration notebook reads production data without reproducible snapshot. First examine: Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata. Then correct: Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs.
Question 686
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: Two Azure ML training runs produce different validation scores but the team cannot determine whether learning rate or data version changed because the Python scripts log only final accuracy. The resolution must preserve quality gates, documented artifacts and rollback options. What approved correction resolves the failure without modifying unrelated healthy services?
- Fix the image or artifact mount path and ensure init() loads the registered model before promoting the deployment.
- Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs.
- Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment.
- Allow the narrowly scoped supported compute-to-compute communication paths and correct the distributed training rendezvous configuration.
Explanation
Correct Answer
C:Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment. This targets mlflow experiment logs omit run parameters. MLflow experiment tracking associates metadata with run IDs, but unlogged hyperparameters and datasets cannot be recovered reliably from a final metric alone. Verify the failure through: Inspect MLflow tracking URI and experiment name, run parameter history, metric logs, artifact URIs and the input data version of each job. Validate success with: A reviewer compares both runs and identifies the exact code, data, hyperparameters and metrics used without reconstructing them by hand. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: Two Azure ML….
Incorrect Answers
A:Fix the image or artifact mount path and ensure init() loads the registered model before promoting the deployment. This could address endpoint deployment health probe fails before model initialization, but it does not solve the specified mlflow experiment logs omit run parameters failure. The matching remediation is: Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment.
B:Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs. This could address exploration notebook reads production data without reproducible snapshot, but it does not solve the specified mlflow experiment logs omit run parameters failure. The matching remediation is: Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment. This matters here because as an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: Two Azure ML….
D:Allow the narrowly scoped supported compute-to-compute communication paths and correct the distributed training rendezvous configuration. This could address distributed job workers cannot communicate over required network path, but it does not solve the specified mlflow experiment logs omit run parameters failure. The matching remediation is: Add MLflow parameter and dataset-version logging to the training script and group runs under a stable named experiment.
Question 687
During a Microsoft Foundry generative AI deployment review, the following condition occurs: A GenAI team commits real customer tickets containing identifiers as few-shot prompt examples in a shared Git repository, exposing data beyond the approved processing boundary. The deployment must retain governed identities, endpoint contracts and auditable release history. What measured outcome validates the solution while preserving the stated constraints?
- Prompt tests retain representative behavior without exposing customer identities in accessible Git history or future deployments.
- Requests from the approved VNet resolve the private address and succeed, while public connectivity remains blocked.
- Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content.
- The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset.
Explanation
Correct Answer
A:Prompt tests retain representative behavior without exposing customer identities in accessible Git history or future deployments. Source control is not an appropriate unreviewed store for sensitive customer data; prompt versioning requires privacy-preserving examples and access governance. a direct corrective action is to remove direct identifiers from ongoing prompt examples, rotate affected credentials if necessary and use sanitized synthetic cases under review. This matters here because a GenAI team commits real customer tickets containing identifiers as few-shot prompt examples in a shared Git….
Incorrect Answers
B:Requests from the approved VNet resolve the private address and succeed, while public connectivity remains blocked. That addresses a different configuration issue rather than the main requirement. Verify instead: Prompt tests retain representative behavior without exposing customer identities in accessible Git history or future deployments.
C:Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content. That addresses a different configuration issue rather than the main requirement. Verify instead: Prompt tests retain representative behavior without exposing customer identities in accessible Git history or future deployments.
D:The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset. That addresses a different configuration issue rather than the main requirement. Verify instead: Prompt tests retain representative behavior without exposing customer identities in accessible Git history or future deployments.
Question 688
While preparing a secure Foundry model or agent release, the following condition occurs: An enterprise disables public network access to Foundry but its container app still resolves the model service hostname to a public endpoint, resulting in failed inference calls. The remediation must meet enterprise access and reliability requirements without a broad bypass. What should the engineer examine first to confirm the specific AI-300 workload issue?
- Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
- Inspect Foundry project and deployment inventory, pipeline environment variables, identities, API endpoints and change audit logs.
- Inspect agent tool definitions, outbound allow rules, service endpoints, private links and network-denied traces.
- Inspect prompt repository history, example dataset provenance, PII classification, branch access controls and affected downstream forks.
Explanation
Correct Answer
A:Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. This is the most relevant evidence for foundry private endpoint dns resolves to public ip. Private network security requires both endpoint approval and correct DNS resolution; a private endpoint creation event does not change every client's name resolution. The matching controlled remediation is: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity. This matters here because while preparing a secure Foundry model or agent release, the following condition occurs: An enterprise disables….
Incorrect Answers
B:Inspect Foundry project and deployment inventory, pipeline environment variables, identities, API endpoints and change audit logs. This would investigate foundry environment mixes development and production agent resources rather than the documented foundry private endpoint dns resolves to public ip issue. The relevant investigation is: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
C:Inspect agent tool definitions, outbound allow rules, service endpoints, private links and network-denied traces. This would investigate agent tool calls fail behind overbroad outbound block rather than the documented foundry private endpoint dns resolves to public ip issue. The relevant investigation is: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
D:Inspect prompt repository history, example dataset provenance, PII classification, branch access controls and affected downstream forks. This would investigate prompt repository references secret customer examples rather than the documented foundry private endpoint dns resolves to public ip issue. The relevant investigation is: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces.
Question 689
During a regulated Azure ML workspace infrastructure review, the following condition occurs: A model platform pipeline applies Bicep changes immediately and breaks a production workspace because it never checks syntax, resource drift or the expected scope. The team must preserve approved access, model provenance and private network constraints. Which diagnostic evidence most directly distinguishes the described AI operations failure?
- Compare the datastore URI and backing storage container, recorded data asset path, job input mount path and expected data version checksum.
- Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals.
- Inspect az account subscription context, deployment scope flags, tenant identity, Bicep parameter file and cloud audit evidence.
- Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history.
Explanation
Correct Answer
B:Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals. This is the most relevant evidence for iac workflow lacks repeatable infrastructure validation. Infrastructure automation should detect configuration drift and unexpected destructive changes before a production deployment proceeds. The matching controlled remediation is: Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
Incorrect Answers
A:Compare the datastore URI and backing storage container, recorded data asset path, job input mount path and expected data version checksum. This would investigate datastore uri selects the incorrect storage container rather than the documented iac workflow lacks repeatable infrastructure validation issue. The relevant investigation is: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals.
C:Inspect az account subscription context, deployment scope flags, tenant identity, Bicep parameter file and cloud audit evidence. This would investigate azure cli deployment silently targets sandbox defaults rather than the documented iac workflow lacks repeatable infrastructure validation issue. The relevant investigation is: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals.
D:Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. This would investigate notebook merge bypasses protected evaluation tests rather than the documented iac workflow lacks repeatable infrastructure validation issue. The relevant investigation is: Inspect GitHub workflow jobs, Bicep build/what-if steps, test deployment history, parameter values and release approvals.
Question 690
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A production deployment references a mutable latest model alias and picks up a candidate model that bypassed validation when another team registered a new version. The resolution must preserve quality gates, documented artifacts and rollback options. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events.
- Evaluate candidate and baseline on the same governed holdout dataset using the same business-aligned metrics and thresholds.
- Implement point-in-time correct feature retrieval and package its temporal lookup logic with the registered model version.
- Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
- Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness.
Explanation
Correct Answers
A:Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events. This is the most relevant evidence for model registry latest alias points to unapproved experiment. Mutable model aliases simplify discovery but are unsafe production identifiers when deployment reproducibility and review are required. The matching controlled remediation is: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
D:Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates. This targets model registry latest alias points to unapproved experiment. Mutable model aliases simplify discovery but are unsafe production identifiers when deployment reproducibility and review are required. Verify the failure through: Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events. Validate success with: Recreating a production deployment resolves the same approved model artifact until an authorized promotion changes the version.
Incorrect Answers
B:Evaluate candidate and baseline on the same governed holdout dataset using the same business-aligned metrics and thresholds. This is associated with run comparison uses incompatible validation populations, not the essential evidence-and-fix pair for model registry latest alias points to unapproved experiment. First examine: Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events. Then correct: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
C:Implement point-in-time correct feature retrieval and package its temporal lookup logic with the registered model version. This is associated with online feature retrieval pulls values from after prediction time, not the essential evidence-and-fix pair for model registry latest alias points to unapproved experiment. First examine: Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events. Then correct: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
E:Examine cohort labels, fairness metrics, stratified error rates, consent constraints and model evaluation dataset representativeness. This is associated with responsible ai evaluation omits demographic cohort slices, not the essential evidence-and-fix pair for model registry latest alias points to unapproved experiment. First examine: Inspect model version IDs, labels/aliases, approval evidence, endpoint deployment configuration and registry audit events. Then correct: Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.