Microsoft AI-300 Practice Test 15
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.
Question 141
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A training job reads a similarly named blob container containing last quarter's features because its Azure ML datastore URI targets the staging location rather than the approved production store. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Which observable postchange result proves the intended remediation succeeded?
- The job consumes the intended dataset version and staging files are not silently mixed into production runs.
- Idle costs fall outside approved usage windows and users can resume the instances for sanctioned experiments.
- Replaying a prior run uses identical input version and produces traceable comparison metrics before a new snapshot is promoted.
- The training job acquires nodes inside documented quota and starts without creating unapproved capacity or dropping the distributed requirement.
Explanation
Correct Answer
A:The job consumes the intended dataset version and staging files are not silently mixed into production runs. A datastore connection and data asset version identify where training inputs come from; paths must be validated independently from authentication success. a direct corrective action is to update the datastore or data asset binding to the governed storage path with an explicitly pinned version and validated read access.
Incorrect Answers
B:Idle costs fall outside approved usage windows and users can resume the instances for sanctioned experiments. That addresses a different configuration issue rather than the main requirement. Verify instead: The job consumes the intended dataset version and staging files are not silently mixed into production runs.
C:Replaying a prior run uses identical input version and produces traceable comparison metrics before a new snapshot is promoted. That addresses a different configuration issue rather than the main requirement. Verify instead: The job consumes the intended dataset version and staging files are not silently mixed into production runs.
D:The training job acquires nodes inside documented quota and starts without creating unapproved capacity or dropping the distributed requirement. That addresses a different configuration issue rather than the main requirement. Verify instead: The job consumes the intended dataset version and staging files are not silently mixed into production runs.
Question 142
As an MLOps engineer reviews a model training or inference pipeline, the following condition occurs: A churn model is registered and deployed, but online inference uses different feature lookup keys from training because the model artifact lacks the feature retrieval specification. The resolution must preserve quality gates, documented artifacts and rollback options. Which observable postchange result proves the intended remediation succeeded?
- A controlled API load test meets latency and correctness targets without requiring clients to poll batch jobs.
- Recreating a production deployment resolves the same approved model artifact until an authorized promotion changes the version.
- Historical inference replay uses only features available before prediction time and realistic performance survives evaluation.
- Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution.
Explanation
Correct Answer
D:Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution. Model weights alone do not specify how feature values were retrieved; production consistency requires a versioned feature contract. a direct corrective action is to package and version the approved feature retrieval specification with the model and enforce the same mapping at serving time.
Incorrect Answers
A:A controlled API load test meets latency and correctness targets without requiring clients to poll batch jobs. That addresses a different configuration issue rather than the main requirement. Verify instead: Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution.
B:Recreating a production deployment resolves the same approved model artifact until an authorized promotion changes the version. That addresses a different configuration issue rather than the main requirement. Verify instead: Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution.
C:Historical inference replay uses only features available before prediction time and realistic performance survives evaluation. That addresses a different configuration issue rather than the main requirement. Verify instead: Offline validation and online scoring use equivalent feature definitions and identical customer keys without silent missing-value substitution.
Question 143
During a retrieval or model-customization performance review, the following condition occurs: An agent's fine-tuning corpus is expanded with thousands of synthetic easy examples, making real rare technical failures a small fraction of training and reducing performance on those cases. The change must keep citations, privacy and previously approved behavior intact. What should the engineer examine first to confirm the specific AI-300 workload issue?
- Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals.
- Inspect fine-tuning records, role/message formatting, target response schemas, label consistency and curated holdout behavior.
- Compare synthetic-to-real sampling weights, rare-class coverage, holdout error breakdown and generated example diversity.
- Inspect representative code-based queries, BM25/keyword results, vector matches, analyzer configuration and fusion ranking behavior.
Explanation
Correct Answer
C:Compare synthetic-to-real sampling weights, rare-class coverage, holdout error breakdown and generated example diversity. This is the most relevant evidence for synthetic data overwhelms minority real-world edge cases. Synthetic data increases quantity but can distort the target distribution and crowd out the cases most important for deployment risk. The matching controlled remediation is: Rebalance curated synthetic and real cases to preserve rare operational scenarios and evaluate on an untouched real-world edge-case set.
Incorrect Answers
A:Examine synthetic-generation input provenance, PII detection scans, privacy transformations, output similarity to real records and review approvals. This would investigate synthetic training examples copy private customer identifiers rather than the documented synthetic data overwhelms minority real-world edge cases issue. The relevant investigation is: Compare synthetic-to-real sampling weights, rare-class coverage, holdout error breakdown and generated example diversity.
B:Inspect fine-tuning records, role/message formatting, target response schemas, label consistency and curated holdout behavior. This would investigate supervised fine-tuning dataset includes conflicting assistant behaviors rather than the documented synthetic data overwhelms minority real-world edge cases issue. The relevant investigation is: Compare synthetic-to-real sampling weights, rare-class coverage, holdout error breakdown and generated example diversity.
D:Inspect representative code-based queries, BM25/keyword results, vector matches, analyzer configuration and fusion ranking behavior. This would investigate pure vector search misses exact regulatory clause codes rather than the documented synthetic data overwhelms minority real-world edge cases issue. The relevant investigation is: Compare synthetic-to-real sampling weights, rare-class coverage, holdout error breakdown and generated example diversity.
Question 144
During a retrieval or model-customization performance review, the following condition occurs: An agent's fine-tuning corpus is expanded with thousands of synthetic easy examples, making real rare technical failures a small fraction of training and reducing performance on those cases. The change must keep citations, privacy and previously approved behavior intact. Which acceptance test shows the AI operations defect is corrected?
- Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence.
- A disaster-recovery deployment loads the reviewed customized model with matching tokenizer and runtime without repeating unrecoverable training.
- The model improves typical coverage while retaining or improving measured accuracy for rare high-impact cases.
- The selected retrieval strategy meets groundedness and task-success gates rather than optimizing click-through in isolation.
Explanation
Correct Answer
C:The model improves typical coverage while retaining or improving measured accuracy for rare high-impact cases. Synthetic data increases quantity but can distort the target distribution and crowd out the cases most important for deployment risk. a direct corrective action is to rebalance curated synthetic and real cases to preserve rare operational scenarios and evaluate on an untouched real-world edge-case set.
Incorrect Answers
A:Previously unseen product queries are monitored and subsequent vetted versions meet current quality targets without erasing old task competence. That addresses a different configuration issue rather than the main requirement. Verify instead: The model improves typical coverage while retaining or improving measured accuracy for rare high-impact cases.
B:A disaster-recovery deployment loads the reviewed customized model with matching tokenizer and runtime without repeating unrecoverable training. That addresses a different configuration issue rather than the main requirement. Verify instead: The model improves typical coverage while retaining or improving measured accuracy for rare high-impact cases.
D:The selected retrieval strategy meets groundedness and task-success gates rather than optimizing click-through in isolation. That addresses a different configuration issue rather than the main requirement. Verify instead: The model improves typical coverage while retaining or improving measured accuracy for rare high-impact cases.
Question 145
While preparing a secure Foundry model or agent release, the following condition occurs: A custom transformer model is deployed on managed compute but inference containers crash on startup because the image's tokenizer library is incompatible with the model artifact. The remediation must meet enterprise access and reliability requirements without a broad bypass. Which observable postchange result proves the intended remediation succeeded?
- Repeatable IaC releases create the expected project and model deployment before starting inference smoke tests.
- Clean inference workers initialize the model and produce valid outputs for representative tokenization inputs.
- Peak requests meet approved latency and throttling targets under a representative load test while capacity costs remain documented.
- A realistic image invoice yields extracted required details without unsupported request content or invented visual observations.
Explanation
Correct Answer
B:Clean inference workers initialize the model and produce valid outputs for representative tokenization inputs. Managed inference still requires the model weights and runtime dependencies to be compatible; compute provisioning cannot fix a broken model-serving image. a direct corrective action is to build and test a compatible version-pinned serving environment and promote the corrected managed model deployment.
Incorrect Answers
A:Repeatable IaC releases create the expected project and model deployment before starting inference smoke tests. That addresses a different configuration issue rather than the main requirement. Verify instead: Clean inference workers initialize the model and produce valid outputs for representative tokenization inputs.
C:Peak requests meet approved latency and throttling targets under a representative load test while capacity costs remain documented. That addresses a different configuration issue rather than the main requirement. Verify instead: Clean inference workers initialize the model and produce valid outputs for representative tokenization inputs.
D:A realistic image invoice yields extracted required details without unsupported request content or invented visual observations. That addresses a different configuration issue rather than the main requirement. Verify instead: Clean inference workers initialize the model and produce valid outputs for representative tokenization inputs.
Question 146
During a regulated Azure ML workspace infrastructure review, the following condition occurs: An Azure ML training job authenticates with a managed identity but fails to download input data because it has workspace rights without Storage Blob Data Reader on the data container. The team must preserve approved access, model provenance and private network constraints. Which targeted Azure AI operations change best addresses the documented problem?
- Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access.
- Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability.
- Replace embedded workspace-specific paths with parameterized, typed data inputs and bind them explicitly in each consuming pipeline.
- Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances.
Explanation
Correct Answer
A:Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access. This targets job identity lacks blob reader data-plane rights. Workspace RBAC and Azure Storage data-plane roles are distinct; successful workspace authentication does not authorize every blob read. Verify the failure through: Check job identity selection, storage data-plane authorization, private endpoint reachability and storage diagnostic access-denied records. Validate success with: The job reads the required training blobs and cannot write or list unrelated protected containers.
Incorrect Answers
B:Request approved additional quota or rescope the job to a permitted compute size and region while retaining necessary GPU capability. This could address compute cluster cannot allocate requested gpu quota, but it does not solve the specified job identity lacks blob reader data-plane rights failure. The matching remediation is: Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access.
C:Replace embedded workspace-specific paths with parameterized, typed data inputs and bind them explicitly in each consuming pipeline. This could address reusable component hardcodes workspace-specific asset path, but it does not solve the specified job identity lacks blob reader data-plane rights failure. The matching remediation is: Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access.
D:Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances. This could address interactive compute instances remain running while unused, but it does not solve the specified job identity lacks blob reader data-plane rights failure. The matching remediation is: Assign the minimum permitted storage data access role to the job identity at the relevant container scope without granting secret-key access.
Question 147
As an MLOps engineer reviews a model training or inference pipeline, 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 resolution must preserve quality gates, documented artifacts and rollback options. Select TWO actions: one decisive diagnostic and its corresponding targeted remediation.
- Inspect notebook input connection, extraction timestamp, data asset versions, query filter and persisted run metadata.
- Capture a governed immutable dataset or versioned data asset and reference that exact version in notebook and scheduled jobs.
- Monitor available proxy signals immediately and compute labeled performance only after verified outcomes arrive through a governed join.
- Review job output mappings, script save path, managed output directories, job artifacts and downstream register-component input.
- Allow the narrowly scoped supported compute-to-compute communication paths and correct the distributed training rendezvous configuration.
Explanation
Correct Answers
A: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.
B: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.
Incorrect Answers
C:Monitor available proxy signals immediately and compute labeled performance only after verified outcomes arrive through a governed join. This is associated with accuracy monitoring lacks delayed outcome labels, 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:Review job output mappings, script save path, managed output directories, job artifacts and downstream register-component input. This is associated with training job does not persist output artifacts, 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.
E:Allow the narrowly scoped supported compute-to-compute communication paths and correct the distributed training rendezvous configuration. 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.
Question 148
During a Microsoft Foundry generative AI deployment review, the following condition occurs: A GenAI application releases its Foundry project and calls a model name before that model deployment is created, causing intermittent first-run errors. The deployment must retain governed identities, endpoint contracts and auditable release history. Select TWO actions: one decisive diagnostic and its corresponding targeted remediation.
- Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations.
- Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration.
- Inspect agent tool definitions, outbound allow rules, service endpoints, private links and network-denied traces.
- Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
- Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review.
Explanation
Correct Answers
A:Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations. This is the most relevant evidence for foundry bicep rollout omits model deployment dependencies. Infrastructure-as-code can provision the platform consistently only when dependencies and deployment readiness are managed explicitly. The matching controlled remediation is: Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
D:Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration. This targets foundry bicep rollout omits model deployment dependencies. Infrastructure-as-code can provision the platform consistently only when dependencies and deployment readiness are managed explicitly. Verify the failure through: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations. Validate success with: Repeatable IaC releases create the expected project and model deployment before starting inference smoke tests.
Incorrect Answers
B:Grant the narrowly scoped search data access role to the correct agent identity and use identity-based connection configuration. This is associated with agent runtime managed identity cannot read retrieval data, not the essential evidence-and-fix pair for foundry bicep rollout omits model deployment dependencies. First examine: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations. Then correct: Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
C:Inspect agent tool definitions, outbound allow rules, service endpoints, private links and network-denied traces. This is associated with agent tool calls fail behind overbroad outbound block, not the essential evidence-and-fix pair for foundry bicep rollout omits model deployment dependencies. First examine: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations. Then correct: Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
E:Move the low-volume prototype to an appropriate supported consumption-based deployment or reduce provisioned capacity after review. This is associated with provisioned deployment pays for idle unused capacity, not the essential evidence-and-fix pair for foundry bicep rollout omits model deployment dependencies. First examine: Inspect Bicep template resource references, model deployment name outputs, CI step dependencies and Azure deployment operations. Then correct: Define the model and project deployment lifecycle dependencies and verify provisioning completion before dependent app configuration.
Question 149
While preparing a secure Foundry model or agent release, the following condition occurs: A team claims prompt B is better than prompt A, but B runs on a newer foundation model so the measured improvement cannot be attributed to the prompt change. The remediation must meet enterprise access and reliability requirements without a broad bypass. What measured outcome validates the solution while preserving the stated constraints?
- The workflow authenticates through a governed identity after old secrets are disabled and completes the intended evaluation without privileged keys in source control.
- The selected prompt improves successful issue resolution without factual regressions on the held-out support dataset.
- Adversarial test documents no longer cause unauthorized agent actions while the assistant still answers legitimate questions from the content.
- The comparison isolates prompt differences and produces reproducible evidence of quality and latency trade-offs.
Explanation
Correct Answer
D:The comparison isolates prompt differences and produces reproducible evidence of quality and latency trade-offs. Experimental prompt comparisons require controlling confounders; a simultaneous model upgrade prevents a valid prompt-only attribution. a direct corrective action is to evaluate both prompt variants against the same pinned model and representative dataset with controlled assignment and consistent metrics.
Incorrect Answers
A:The workflow authenticates through a governed identity after old secrets are disabled and completes the intended evaluation without privileged keys in source control. That addresses a different configuration issue rather than the main requirement. Verify instead: The comparison isolates prompt differences and produces reproducible evidence of quality and latency trade-offs.
B: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: The comparison isolates prompt differences and produces reproducible evidence of quality and latency trade-offs.
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: The comparison isolates prompt differences and produces reproducible evidence of quality and latency trade-offs.
Question 150
While deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A GitHub deployment workload that only uploads models receives broad subscription Contributor rights, enabling it to modify unrelated application infrastructure. The change must remain limited to authorized workspaces and avoid unreviewed production impact. Select TWO: the strongest initial evidence and the aligned controlled correction.
- Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
- Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances.
- Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema.
- Inspect the datastore authentication mode, storage key rotation event, job input failure, and available identity-based storage permission assignments.
- Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity.
Explanation
Correct Answers
A:Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant. This targets workspace contributor role granted to untrusted automation. An Azure resource role assignment grants management-plane capabilities according to scope and cannot be justified solely by pipeline convenience. Verify the failure through: Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. Validate success with: Automated model registration succeeds while the principal cannot change unrelated networks, secrets or subscriptions. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A GitHub….
E:Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. This is the most relevant evidence for workspace contributor role granted to untrusted automation. An Azure resource role assignment grants management-plane capabilities according to scope and cannot be justified solely by pipeline convenience. The matching controlled remediation is: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant. This matters here because while deploying reproducible Azure Machine Learning infrastructure, the following condition occurs: A GitHub….
Incorrect Answers
B:Apply supported automatic shutdown and business-hour schedules after confirming no required interactive or scheduled process depends on the instances. This is associated with interactive compute instances remain running while unused, not the essential evidence-and-fix pair for workspace contributor role granted to untrusted automation. First examine: Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. Then correct: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
C:Align the component output with the declared downstream input type or introduce a supported data conversion step with versioned schema. This is associated with pipeline component output type mismatches consumer, not the essential evidence-and-fix pair for workspace contributor role granted to untrusted automation. First examine: Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. Then correct: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.
D:Inspect the datastore authentication mode, storage key rotation event, job input failure, and available identity-based storage permission assignments. This is associated with datastore authentication relies on expiring account key, not the essential evidence-and-fix pair for workspace contributor role granted to untrusted automation. First examine: Inspect the workload principal, Azure ML workspace role assignment, subscription inheritance, required model-write actions and audit activity. Then correct: Scope an approved least-privilege Azure ML role to the intended resource group or workspace and remove the broad inherited Contributor grant.