Microsoft AI-300 Practice Test 14
Question 130
During a Foundry generative AI quality and observability incident, the following condition occurs: A Foundry agent returns an unhelpful answer after a search tool timeout but logs only the final generated text, hiding the actual failing service and request correlation. The fix must preserve evaluation validity and avoid unnecessary user-data collection. What approved correction resolves the failure without modifying unrelated healthy services?
- Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers.
- Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.
- Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
- Mask sensitive tokens and personal data, apply least-privilege log access and retain only necessary structured debugging evidence.
Explanation
Correct Answer
A:Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers. This targets agent traces omit downstream tool failure reason. Multi-step AI agent troubleshooting needs per-tool traces, not just the final answer and an overall request status. Verify the failure through: Inspect agent instrumentation, OpenTelemetry trace propagation, tool span durations, error attributes and correlated resource logs. Validate success with: An induced tool timeout is attributable to the exact trace span and remediation can target the failing service.
Incorrect Answers
B:Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend. This could address token cost report omits embedding requests, but it does not solve the specified agent traces omit downstream tool failure reason failure. The matching remediation is: Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers.
C:Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run. This could address genai evaluation dataset lacks expected answer column mapping, but it does not solve the specified agent traces omit downstream tool failure reason failure. The matching remediation is: Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers.
D:Mask sensitive tokens and personal data, apply least-privilege log access and retain only necessary structured debugging evidence. This could address debug traces leak secrets and retrieved personal records, but it does not solve the specified agent traces omit downstream tool failure reason failure. The matching remediation is: Enable scoped distributed tracing for agent and tool steps with error tags and privacy-safe identifiers.
Question 131
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A distributed deep-learning job remains queued because the Azure ML compute cluster requests more regional GPU-family vCPU quota than the subscription currently has. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Select TWO actions: one decisive diagnostic and its corresponding targeted remediation.
- Compare the datastore URI and backing storage container, recorded data asset path, job input mount path and expected data version checksum.
- Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately.
- Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
- Gate production deployment behind approved GitHub environment protection, protected refs and status checks while keeping test deployments separate.
- Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability.
Explanation
Correct Answers
C:Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability. This targets compute cluster cannot allocate requested gpu quota. Azure ML computes draw from Azure regional VM family quota; declaring a cluster does not reserve unlimited GPU instances. Verify the failure through: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability. Validate success with: The training job acquires nodes inside documented quota and starts without creating unapproved capacity or dropping the distributed requirement.
E:Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability. This is the most relevant evidence for compute cluster cannot allocate requested gpu quota. Azure ML computes draw from Azure regional VM family quota; declaring a cluster does not reserve unlimited GPU instances. The matching controlled remediation is: Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
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 is associated with datastore uri selects the incorrect storage container, not the essential evidence-and-fix pair for compute cluster cannot allocate requested gpu quota. First examine: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability. Then correct: Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
B:Recreate or provision the workspace in the approved subscription and region through the reviewed deployment pipeline, preserving authorized datasets separately. This is associated with azure ml workspace deployed in the wrong subscription, not the essential evidence-and-fix pair for compute cluster cannot allocate requested gpu quota. First examine: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability. Then correct: Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
D:Gate production deployment behind approved GitHub environment protection, protected refs and status checks while keeping test deployments separate. This is associated with github deployment pipeline omits secure environment approvals, not the essential evidence-and-fix pair for compute cluster cannot allocate requested gpu quota. First examine: Inspect compute cluster requested SKU, max nodes, available family quota, current utilization, node provisioning events and region availability. Then correct: Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
Question 132
As an AI operations team checks production agent evaluation evidence, the following condition occurs: A Foundry agent evaluation job runs against a CSV whose expected_answer column was mapped as user_query, so relevance and groundedness results are invalid. The workflow must produce measurable quality, safety and performance evidence. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution.
- Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples.
- Inspect evaluation dataset field names, input-output column mapping, evaluator requirements, row samples and failed scoring logs.
- Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend.
- Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
Explanation
Correct Answers
C:Inspect evaluation dataset field names, input-output column mapping, evaluator requirements, row samples and failed scoring logs. This is the most relevant evidence for genai evaluation dataset lacks expected answer column mapping. Generative evaluation depends on the semantic role of each dataset field; a syntactically valid CSV is not necessarily an evaluable response set. The matching controlled remediation is: Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
E:Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run. This targets genai evaluation dataset lacks expected answer column mapping. Generative evaluation depends on the semantic role of each dataset field; a syntactically valid CSV is not necessarily an evaluable response set. Verify the failure through: Inspect evaluation dataset field names, input-output column mapping, evaluator requirements, row samples and failed scoring logs. Validate success with: Evaluation scores correspond to the intended example queries and reference answers without null or swapped fields.
Incorrect Answers
A:Add approved prompt-injection and tool-abuse evaluation cases with tool permission gates and block release on unsafe execution. This is associated with content safety evaluation omits prompt injection tests, not the essential evidence-and-fix pair for genai evaluation dataset lacks expected answer column mapping. First examine: Inspect evaluation dataset field names, input-output column mapping, evaluator requirements, row samples and failed scoring logs. Then correct: Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
B:Review user input field mapping, generated search query, response text, evaluator prompt and evaluation row samples. This is associated with relevance evaluator uses wrong intended user query, not the essential evidence-and-fix pair for genai evaluation dataset lacks expected answer column mapping. First examine: Inspect evaluation dataset field names, input-output column mapping, evaluator requirements, row samples and failed scoring logs. Then correct: Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
D:Instrument cost attribution across generation and embedding services, cache reusable embeddings and alert on unplanned spend. This is associated with token cost report omits embedding requests, not the essential evidence-and-fix pair for genai evaluation dataset lacks expected answer column mapping. First examine: Inspect evaluation dataset field names, input-output column mapping, evaluator requirements, row samples and failed scoring logs. Then correct: Correct the test dataset mapping for query, context, response and reference fields and validate a sampled evaluation before the full run.
Question 133
While preparing a secure Foundry model or agent release, the following condition occurs: A new troubleshooting prompt sounds polished in sample responses but fails to give actionable next steps because the evaluation measures only fluency and omits task completion and grounding. The remediation must meet enterprise access and reliability requirements without a broad bypass. Which investigation best verifies the cause before changing production configuration?
- Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds.
- Inspect the prompt template, target JSON schema, model output examples, parser exceptions and current evaluation dataset.
- Inspect prompt repository history, example dataset provenance, PII classification, branch access controls and affected downstream forks.
- Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
Explanation
Correct Answer
A:Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds. This is the most relevant evidence for prompt variant comparison measures fluency not task success. A fluent response can still be wrong or unhelpful; prompt tuning requires evaluation metrics tied to application goals. The matching controlled remediation is: Compare candidate prompts using task-specific relevance and groundedness metrics alongside fluency before promotion.
Incorrect Answers
B:Inspect the prompt template, target JSON schema, model output examples, parser exceptions and current evaluation dataset. This would investigate prompt omits explicit structured answer schema rather than the documented prompt variant comparison measures fluency not task success issue. The relevant investigation is: Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds.
C: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 prompt variant comparison measures fluency not task success issue. The relevant investigation is: Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds.
D:Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles. This would investigate provisioned throughput capacity insufficient at peak traffic rather than the documented prompt variant comparison measures fluency not task success issue. The relevant investigation is: Review prompt variant outputs, evaluator criteria, human task labels, relevance/groundedness scores and business acceptance thresholds.
Question 134
During a Microsoft Foundry generative AI deployment review, the following condition occurs: An agent receives a newly released model under full production traffic before anyone verifies tool calling and safety regressions. The deployment must retain governed identities, endpoint contracts and auditable release history. Which production-safe remediation should the engineer apply to this AI workflow?
- Separate trusted instructions from quoted retrieved context and enforce restricted tool permissions and injection-resilient prompt handling.
- Select an approved multimodal model or add a supported image-to-text preprocessing service before the text generation step.
- Route a controlled percentage of requests to the candidate and promote only after automated quality, safety and latency gates pass.
- Right-size the supported provisioned capacity or apply controlled request pacing and overflow strategy within budget and region limits.
Explanation
Correct Answer
C:Route a controlled percentage of requests to the candidate and promote only after automated quality, safety and latency gates pass. This targets new foundation model replaces production without canary. Gradual rollout with a known-good model mitigates production blast radius when foundation model behavior changes. Verify the failure through: Inspect deployment routing weights, evaluated model versions, incident timing, canary quality dashboards and rollback path. Validate success with: Failed canary metrics restore 100% of traffic to the last approved deployment without changing unrelated agent code.
Incorrect Answers
A:Separate trusted instructions from quoted retrieved context and enforce restricted tool permissions and injection-resilient prompt handling. This could address prompt includes untrusted retrieved instructions as system policy, but it does not solve the specified new foundation model replaces production without canary failure. The matching remediation is: Route a controlled percentage of requests to the candidate and promote only after automated quality, safety and latency gates pass.
B:Select an approved multimodal model or add a supported image-to-text preprocessing service before the text generation step. This could address vision input sent to text-only foundation model, but it does not solve the specified new foundation model replaces production without canary failure. The matching remediation is: Route a controlled percentage of requests to the candidate and promote only after automated quality, safety and latency gates pass.
D:Right-size the supported provisioned capacity or apply controlled request pacing and overflow strategy within budget and region limits. This could address provisioned throughput capacity insufficient at peak traffic, but it does not solve the specified new foundation model replaces production without canary failure. The matching remediation is: Route a controlled percentage of requests to the candidate and promote only after automated quality, safety and latency gates pass.
Question 135
While preparing a secure Foundry model or agent release, the following condition occurs: A chat assistant must return low-latency answers but the team selects an offline batch inference architecture that produces responses only after a scheduled job completes. 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?
- Inspect deployment routing weights, evaluated model versions, incident timing, canary quality dashboards and rollback path.
- Examine model deployment quota for selected region/type, Azure CLI errors, requested capacity units and existing model allocations.
- Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior.
- Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles.
Explanation
Correct Answer
C:Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior. This is the most relevant evidence for interactive inference deployed on unsuitable batch-style compute. Generative chat workloads ordinarily require online serving semantics; scheduled batch compute cannot satisfy an interactive response deadline. The matching controlled remediation is: Deploy an available supported online/serverless or managed model endpoint sized for synchronous conversational requests.
Incorrect Answers
A:Inspect deployment routing weights, evaluated model versions, incident timing, canary quality dashboards and rollback path. This would investigate new foundation model replaces production without canary rather than the documented interactive inference deployed on unsuitable batch-style compute issue. The relevant investigation is: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior.
B:Examine model deployment quota for selected region/type, Azure CLI errors, requested capacity units and existing model allocations. This would investigate azure cli foundry setup assumes existing model quota rather than the documented interactive inference deployed on unsuitable batch-style compute issue. The relevant investigation is: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior.
D:Inspect input/output token volume, peak concurrent requests, provisioned throughput allocation, throttling codes and deployment latency percentiles. This would investigate provisioned throughput capacity insufficient at peak traffic rather than the documented interactive inference deployed on unsuitable batch-style compute issue. The relevant investigation is: Check application response SLA, workload concurrency, chosen Foundry model endpoint type and current request queuing behavior.
Question 136
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: A withdrawn credit-risk model is marked deprecated in a wiki but remains exposed as an active registry version and continues to be selected by automated deployment tools. The operator must safeguard training reproducibility and serving continuity. Which production-safe remediation should the engineer apply to this AI workflow?
- Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates.
- Register a supported MLflow model with validated flavor, explicit input signature, pinned dependencies and a successful clean-environment load test.
- Enable supported model artifact logging or explicitly log a serializable MLflow model with its dependencies and signature.
- Archive or retire the version through a governed registry lifecycle and remove it from production deployment aliases after dependents migrate.
Explanation
Correct Answer
D:Archive or retire the version through a governed registry lifecycle and remove it from production deployment aliases after dependents migrate. This targets retired model kept active as discoverable deployment candidate. Model lifecycle management must reflect state in the model registry and deployment references, not just informal documentation. Verify the failure through: Inspect registry model lifecycle status, alias assignments, endpoint references, deployment logs and retirement approvals. Validate success with: New deployment workflows cannot select the retired version while required historic audit artifacts remain available.
Incorrect Answers
A:Pin production deployments to an immutable reviewed model version and promote newer versions only after explicit approval gates. This could address model registry latest alias points to unapproved experiment, but it does not solve the specified retired model kept active as discoverable deployment candidate failure. The matching remediation is: Archive or retire the version through a governed registry lifecycle and remove it from production deployment aliases after dependents migrate.
B:Register a supported MLflow model with validated flavor, explicit input signature, pinned dependencies and a successful clean-environment load test. This could address mlflow registry artifact cannot load on serving target, but it does not solve the specified retired model kept active as discoverable deployment candidate failure. The matching remediation is: Archive or retire the version through a governed registry lifecycle and remove it from production deployment aliases after dependents migrate.
C:Enable supported model artifact logging or explicitly log a serializable MLflow model with its dependencies and signature. This could address autologging excludes required model artifact, but it does not solve the specified retired model kept active as discoverable deployment candidate failure. The matching remediation is: Archive or retire the version through a governed registry lifecycle and remove it from production deployment aliases after dependents migrate.
Question 137
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A feature-engineering change is merged directly to the main branch without running schema and leakage tests, and the next scheduled model training uses incompatible features. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Configure GitHub OIDC federated identity with a scoped Azure service principal and replace the long-lived secret in the deployment workflow.
- Require protected pull requests and automated component tests before changes to production training code are merged.
- Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs.
- Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history.
- Grant controlled registry consumption rights and use the fully qualified versioned registry asset reference in the target workspace.
Explanation
Correct Answers
B:Require protected pull requests and automated component tests before changes to production training code are merged. This targets notebook merge bypasses protected evaluation tests. Git repositories need enforced review and CI checks; simply storing a notebook in source control does not prevent defective changes. Verify the failure through: Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. Validate success with: A breaking feature change is rejected before deployment and approved commits are tied to passed evaluation evidence.
D:Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. This is the most relevant evidence for notebook merge bypasses protected evaluation tests. Git repositories need enforced review and CI checks; simply storing a notebook in source control does not prevent defective changes. The matching controlled remediation is: Require protected pull requests and automated component tests before changes to production training code are merged.
Incorrect Answers
A:Configure GitHub OIDC federated identity with a scoped Azure service principal and replace the long-lived secret in the deployment workflow. This is associated with github action uses persistent azure client secret, not the essential evidence-and-fix pair for notebook merge bypasses protected evaluation tests. First examine: Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. Then correct: Require protected pull requests and automated component tests before changes to production training code are merged.
C:Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs. This is associated with aml private workspace storage access fails from compute, not the essential evidence-and-fix pair for notebook merge bypasses protected evaluation tests. First examine: Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. Then correct: Require protected pull requests and automated component tests before changes to production training code are merged.
E:Grant controlled registry consumption rights and use the fully qualified versioned registry asset reference in the target workspace. This is associated with registry model asset not available to second workspace, not the essential evidence-and-fix pair for notebook merge bypasses protected evaluation tests. First examine: Inspect Git branch protection, pull-request review requirements, test workflow triggers, dataset schema checks and recent commit history. Then correct: Require protected pull requests and automated component tests before changes to production training code are merged.
Question 138
During a Microsoft Foundry generative AI deployment review, 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 deployment must retain governed identities, endpoint contracts and auditable release history. What approved correction resolves the failure without modifying unrelated healthy services?
- Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
- Separate governed development and production environments with distinct project configuration, least-privilege credentials and deployment gates.
- Pin an approved model/deployment version where supported and pass candidate upgrades through evaluation and controlled promotion.
- Deploy the governed Foundry resource and project in a supported compliant region with an available model option.
Explanation
Correct Answer
A:Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity. This targets 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. Verify the failure through: Compare approved private endpoint status, private DNS zones and links, container VNet DNS answers, effective routes and model request traces. Validate success with: Requests from the approved VNet resolve the private address and succeed, while public connectivity remains blocked.
Incorrect Answers
B:Separate governed development and production environments with distinct project configuration, least-privilege credentials and deployment gates. This could address foundry environment mixes development and production agent resources, but it does not solve the specified foundry private endpoint dns resolves to public ip failure. The matching remediation is: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
C:Pin an approved model/deployment version where supported and pass candidate upgrades through evaluation and controlled promotion. This could address foundry deployment alias changes model behavior unexpectedly, but it does not solve the specified foundry private endpoint dns resolves to public ip failure. The matching remediation is: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
D:Deploy the governed Foundry resource and project in a supported compliant region with an available model option. This could address foundry project created in mismatched resource location, but it does not solve the specified foundry private endpoint dns resolves to public ip failure. The matching remediation is: Implement the correct supported Foundry private networking and DNS integration for the inference endpoint before denying public connectivity.
Question 139
During a regulated Azure ML workspace infrastructure review, the following condition occurs: An ML team edits training notebooks directly on a shared VM and deploys model versions without recording the corresponding Git commit, preventing a reproducible incident rollback. The team must preserve approved access, model provenance and private network constraints. Select TWO actions: one decisive diagnostic and its corresponding targeted remediation.
- Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow.
- Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply.
- Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers.
- Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access.
- Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs.
Explanation
Correct Answers
A:Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow. This targets production training script commit cannot be traced. Source control alone does not provide provenance unless the pipeline records which revision produced the deployed asset. Verify the failure through: Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers. Validate success with: An incident analyst can reproduce the model version's code and dependencies from its recorded source commit.
C:Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers. This is the most relevant evidence for production training script commit cannot be traced. Source control alone does not provide provenance unless the pipeline records which revision produced the deployed asset. The matching controlled remediation is: Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow.
Incorrect Answers
B:Add Bicep validation, targeted what-if reviews and a controlled environment promotion stage before production apply. This is associated with iac workflow lacks repeatable infrastructure validation, not the essential evidence-and-fix pair for production training script commit cannot be traced. First examine: Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers. Then correct: Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow.
D:Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access. This is associated with job identity lacks blob reader data-plane rights, not the essential evidence-and-fix pair for production training script commit cannot be traced. First examine: Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers. Then correct: Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow.
E:Inspect storage private endpoint approvals, privatelink DNS A records, compute subnet DNS, routes and outbound access test logs. This is associated with aml private workspace storage access fails from compute, not the essential evidence-and-fix pair for production training script commit cannot be traced. First examine: Inspect model metadata, job source snapshot, repository history, CI release manifest and available recorded commit identifiers. Then correct: Bind every training job and model registration to a reviewed Git commit SHA and immutable environment/data version in the workflow.
Question 140
During a production Azure Machine Learning model lifecycle incident, the following condition occurs: An AutoML classification search reports near-perfect accuracy because the supplied training dataset includes a field recorded only after the real customer outcome is known. The operator must safeguard training reproducibility and serving continuity. Which targeted Azure AI operations change best addresses the documented problem?
- Register a supported MLflow model with validated flavor, explicit input signature, pinned dependencies and a successful clean-environment load test.
- Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold.
- Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features.
- Version the transformation component and inputs explicitly and disable or invalidate reuse where the effective computation has changed.
Explanation
Correct Answer
C:Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features. This targets automated ml experiment uses leakage feature. Automated model search cannot correct information leakage caused by improperly constructed input data and validation boundaries. Verify the failure through: Review AutoML featurization, data collection timestamps, target leakage candidates, split strategy and training-versus-production feature availability. Validate success with: Holdout performance is credible against production-like inputs and candidate selection no longer uses post-outcome fields.
Incorrect Answers
A:Register a supported MLflow model with validated flavor, explicit input signature, pinned dependencies and a successful clean-environment load test. This could address mlflow registry artifact cannot load on serving target, but it does not solve the specified automated ml experiment uses leakage feature failure. The matching remediation is: Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features.
B:Use a documented holdout or cross-validation-based model selection procedure and require a production-relevant promotion threshold. This could address new candidate chosen from training instead of validation results, but it does not solve the specified automated ml experiment uses leakage feature failure. The matching remediation is: Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features.
D:Version the transformation component and inputs explicitly and disable or invalidate reuse where the effective computation has changed. This could address pipeline cache reuses stale transform artifact, but it does not solve the specified automated ml experiment uses leakage feature failure. The matching remediation is: Remove target-leakage columns and rerun the AutoML search with an appropriate time-aware validation split and realistic available features.