McAfee-Secured Website

Exam Code: AI-300

Exam Name: Operationalizing Machine Learning and Generative AI Solutions

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Machine Learning Operations Engineer Associate

100% Microsoft AI-300 Guaranteed Success With Testking Training Materials

Get Real AI-300 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

213 Questions & Answers with Testing Engine

"Operationalizing Machine Learning and Generative AI Solutions Exam", also known as AI-300 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-300 Exam Engine. Your AI-300 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-300 Practice Test 80

Question 790

A deployment script references a fine-tuned model name without pinning its base version, so later deployments differ. What should be locked?

  1. Track training, evaluation, hosting, and inference costs for the tuned model lifecycle.
  2. Record the exact fine-tuned model artifact identifier and its base-model lineage.
  3. Confirm model-specific deployment region and SKU support before scheduling the production cutover.
  4. Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version.

Explanation

Correct Answer

B:

Model names alone do not guarantee identical weights or behavior across future production releases.

Incorrect Answers

A:

The alternative, Track training, evaluation, hosting, and inference costs for the tuned model lifecycle, belongs to a different technical condition: Training consumed large GPU resources but budget reports include only inference spending. There the source-specific mechanism or proof is: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision. The condition asked about here is: A deployment script references a fine-tuned model name without pinning its base version, so later deployments differ. A suitable correction is Record the exact fine-tuned model artifact identifier and its base-model lineage; the targeted justification or result is: Model names alone do not guarantee identical weights or behavior across future production releases.

C:

Confirm model-specific deployment region and SKU support before scheduling the production cutover. That operation would be relevant if: A team has an approved tuning recipe but the target Azure region lacks serving support for the selected model. Its corresponding reason or verification is: Training completion does not guarantee that the resulting model is deployable in every geography or serving tier. However the problem in this question is: A deployment script references a fine-tuned model name without pinning its base version, so later deployments differ. The requirement instead calls for Record the exact fine-tuned model artifact identifier and its base-model lineage; the question-specific check is: Model names alone do not guarantee identical weights or behavior across future production releases.

D:

Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version. A separate source case supports that approach when: An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. The supporting rationale or observable result is: Governance traceability requires evidence beyond a model display name when customized outputs are questioned. This question describes a different issue: A deployment script references a fine-tuned model name without pinning its base version, so later deployments differ. Address it through Record the exact fine-tuned model artifact identifier and its base-model lineage, with the required justification or evidence: Model names alone do not guarantee identical weights or behavior across future production releases.

Question 791

A team has an approved tuning recipe but the target Azure region lacks serving support for the selected model. What check comes first?

  1. Confirm model-specific deployment region and SKU support before scheduling the production cutover.
  2. Release a tested prompt-model configuration manifest with both exact identifiers.
  3. Require documented data-use, safety, and business-owner approvals on the immutable release candidate.
  4. Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed.

Explanation

Correct Answer

A:

Training completion does not guarantee that the resulting model is deployable in every geography or serving tier.

Incorrect Answers

B:

Release a tested prompt-model configuration manifest with both exact identifiers. It addresses another case: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. For that case, the distinguishing justification or acceptance check is: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies. The present case instead states: A team has an approved tuning recipe but the target Azure region lacks serving support for the selected model. It calls for Confirm model-specific deployment region and SKU support before scheduling the production cutover. Showing success for the other case would not establish: Training completion does not guarantee that the resulting model is deployable in every geography or serving tier.

C:

The alternative, Require documented data-use, safety, and business-owner approvals on the immutable release candidate, belongs to a different technical condition: A team collects promising offline results but has no security or product approval for tuning on customer records. There the source-specific mechanism or proof is: Quality benchmarks do not replace governance decisions about lawful data use and permitted production risk. The condition asked about here is: A team has an approved tuning recipe but the target Azure region lacks serving support for the selected model. A suitable correction is Confirm model-specific deployment region and SKU support before scheduling the production cutover; the targeted justification or result is: Training completion does not guarantee that the resulting model is deployable in every geography or serving tier.

D:

Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed. That operation would be relevant if: The base model underlying a deployed tuned variant is approaching retirement. Its corresponding reason or verification is: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency. However the problem in this question is: A team has an approved tuning recipe but the target Azure region lacks serving support for the selected model. The requirement instead calls for Confirm model-specific deployment region and SKU support before scheduling the production cutover; the question-specific check is: Training completion does not guarantee that the resulting model is deployable in every geography or serving tier.

Question 792

A regulator requires that only approved applications call the tuned model. What endpoint configuration is needed?

  1. Require documented data-use, safety, and business-owner approvals on the immutable release candidate.
  2. Track training, evaluation, hosting, and inference costs for the tuned model lifecycle.
  3. Run shadow inference on mirrored approved traffic and compare outputs without returning experimental responses.
  4. Protect serving endpoints with managed identity or approved auth and least-privilege network access.

Explanation

Correct Answer

D:

A tuned model remains a service endpoint requiring normal identity and network protection controls.

Incorrect Answers

A:

Require documented data-use, safety, and business-owner approvals on the immutable release candidate. This is useful for the distinct issue: A team collects promising offline results but has no security or product approval for tuning on customer records. The relevant technical rationale or evidence for that issue is: Quality benchmarks do not replace governance decisions about lawful data use and permitted production risk. Here the reported issue is: A regulator requires that only approved applications call the tuned model. The relevant response is: Protect serving endpoints with managed identity or approved auth and least-privilege network access. The first result cannot verify this different correction: A tuned model remains a service endpoint requiring normal identity and network protection controls.

B:

Track training, evaluation, hosting, and inference costs for the tuned model lifecycle. It addresses another case: Training consumed large GPU resources but budget reports include only inference spending. For that case, the distinguishing justification or acceptance check is: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision. The present case instead states: A regulator requires that only approved applications call the tuned model. It calls for Protect serving endpoints with managed identity or approved auth and least-privilege network access. Showing success for the other case would not establish: A tuned model remains a service endpoint requiring normal identity and network protection controls.

C:

The alternative, Run shadow inference on mirrored approved traffic and compare outputs without returning experimental responses, belongs to a different technical condition: A new tuned model is deployed alongside the old one without serving live users; the team wants production comparisons without side effects. There the source-specific mechanism or proof is: Shadow evaluation measures real-input behavior while the established production model continues serving users. The condition asked about here is: A regulator requires that only approved applications call the tuned model. A suitable correction is Protect serving endpoints with managed identity or approved auth and least-privilege network access; the targeted justification or result is: A tuned model remains a service endpoint requiring normal identity and network protection controls.

Question 793

An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. What should accompany the artifact?

  1. Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed.
  2. Protect serving endpoints with managed identity or approved auth and least-privilege network access.
  3. Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version.
  4. Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment.

Explanation

Correct Answer

C:

Governance traceability requires evidence beyond a model display name when customized outputs are questioned.

Incorrect Answers

A:

Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed. A separate source case supports that approach when: The base model underlying a deployed tuned variant is approaching retirement. The supporting rationale or observable result is: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency. This question describes a different issue: An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. Address it through Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version, with the required justification or evidence: Governance traceability requires evidence beyond a model display name when customized outputs are questioned.

B:

Protect serving endpoints with managed identity or approved auth and least-privilege network access. This is useful for the distinct issue: A regulator requires that only approved applications call the tuned model. The relevant technical rationale or evidence for that issue is: A tuned model remains a service endpoint requiring normal identity and network protection controls. Here the reported issue is: An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. The relevant response is: Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version. The first result cannot verify this different correction: Governance traceability requires evidence beyond a model display name when customized outputs are questioned.

D:

Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment. It addresses another case: A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. For that case, the distinguishing justification or acceptance check is: Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved. The present case instead states: An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. It calls for Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version. Showing success for the other case would not establish: Governance traceability requires evidence beyond a model display name when customized outputs are questioned.

Question 794

A team wants to A/B test two fine-tuned releases but randomly alternates models on every retry of one user session. What must change?

  1. Release a tested prompt-model configuration manifest with both exact identifiers.
  2. Persist stable experiment assignment per eligible session or user cohort.
  3. Track training, evaluation, hosting, and inference costs for the tuned model lifecycle.
  4. Use staged canary deployment with task and safety acceptance thresholds.

Explanation

Correct Answer

B:

Consistent allocation prevents retries from mixing model behaviors and corrupting outcome attribution.

Incorrect Answers

A:

Release a tested prompt-model configuration manifest with both exact identifiers. That operation would be relevant if: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. Its corresponding reason or verification is: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies. However the problem in this question is: A team wants to A/B test two fine-tuned releases but randomly alternates models on every retry of one user session. The requirement instead calls for Persist stable experiment assignment per eligible session or user cohort; the question-specific check is: Consistent allocation prevents retries from mixing model behaviors and corrupting outcome attribution.

C:

Track training, evaluation, hosting, and inference costs for the tuned model lifecycle. A separate source case supports that approach when: Training consumed large GPU resources but budget reports include only inference spending. The supporting rationale or observable result is: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision. This question describes a different issue: A team wants to A/B test two fine-tuned releases but randomly alternates models on every retry of one user session. Address it through Persist stable experiment assignment per eligible session or user cohort, with the required justification or evidence: Consistent allocation prevents retries from mixing model behaviors and corrupting outcome attribution.

D:

Use staged canary deployment with task and safety acceptance thresholds. This is useful for the distinct issue: A new tuned model is ready, but the team proposes replacing all production traffic instantly. The relevant technical rationale or evidence for that issue is: Controlled traffic increases allow detection of production regressions before the entire customer population is affected. Here the reported issue is: A team wants to A/B test two fine-tuned releases but randomly alternates models on every retry of one user session. The relevant response is: Persist stable experiment assignment per eligible session or user cohort. The first result cannot verify this different correction: Consistent allocation prevents retries from mixing model behaviors and corrupting outcome attribution.

Question 795

The base model underlying a deployed tuned variant is approaching retirement. What continuity plan is appropriate?

  1. Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed.
  2. Track training, evaluation, hosting, and inference costs for the tuned model lifecycle.
  3. Release a tested prompt-model configuration manifest with both exact identifiers.
  4. Persist stable experiment assignment per eligible session or user cohort.

Explanation

Correct Answer

A:

Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency.

Incorrect Answers

B:

The alternative, Track training, evaluation, hosting, and inference costs for the tuned model lifecycle, belongs to a different technical condition: Training consumed large GPU resources but budget reports include only inference spending. There the source-specific mechanism or proof is: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision. The condition asked about here is: The base model underlying a deployed tuned variant is approaching retirement. A suitable correction is Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed; the targeted justification or result is: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency.

C:

Release a tested prompt-model configuration manifest with both exact identifiers. That operation would be relevant if: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. Its corresponding reason or verification is: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies. However the problem in this question is: The base model underlying a deployed tuned variant is approaching retirement. The requirement instead calls for Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed; the question-specific check is: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency.

D:

Persist stable experiment assignment per eligible session or user cohort. A separate source case supports that approach when: A team wants to A/B test two fine-tuned releases but randomly alternates models on every retry of one user session. The supporting rationale or observable result is: Consistent allocation prevents retries from mixing model behaviors and corrupting outcome attribution. This question describes a different issue: The base model underlying a deployed tuned variant is approaching retirement. Address it through Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed, with the required justification or evidence: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency.

Question 796

Training consumed large GPU resources but budget reports include only inference spending. What should the release cost record include?

  1. Confirm model-specific deployment region and SKU support before scheduling the production cutover.
  2. Keep the prior validated model deployment available with a tested traffic reversal procedure.
  3. Persist stable experiment assignment per eligible session or user cohort.
  4. Track training, evaluation, hosting, and inference costs for the tuned model lifecycle.

Explanation

Correct Answer

D:

Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision.

Incorrect Answers

A:

Confirm model-specific deployment region and SKU support before scheduling the production cutover. It addresses another case: A team has an approved tuning recipe but the target Azure region lacks serving support for the selected model. For that case, the distinguishing justification or acceptance check is: Training completion does not guarantee that the resulting model is deployable in every geography or serving tier. The present case instead states: Training consumed large GPU resources but budget reports include only inference spending. It calls for Track training, evaluation, hosting, and inference costs for the tuned model lifecycle. Showing success for the other case would not establish: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision.

B:

The alternative, Keep the prior validated model deployment available with a tested traffic reversal procedure, belongs to a different technical condition: A newly promoted tuned version starts returning invalid JSON, and the old model passed production tests. There the source-specific mechanism or proof is: An operational rollback must restore known-good serving behavior without rebuilding or retraining the old artifact. The condition asked about here is: Training consumed large GPU resources but budget reports include only inference spending. A suitable correction is Track training, evaluation, hosting, and inference costs for the tuned model lifecycle; the targeted justification or result is: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision.

C:

Persist stable experiment assignment per eligible session or user cohort. That operation would be relevant if: A team wants to A/B test two fine-tuned releases but randomly alternates models on every retry of one user session. Its corresponding reason or verification is: Consistent allocation prevents retries from mixing model behaviors and corrupting outcome attribution. However the problem in this question is: Training consumed large GPU resources but budget reports include only inference spending. The requirement instead calls for Track training, evaluation, hosting, and inference costs for the tuned model lifecycle; the question-specific check is: Total ownership cost extends beyond per-token inference charges and should inform the build-versus-buy decision.

Question 797

A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. What should be versioned together?

  1. Use staged canary deployment with task and safety acceptance thresholds.
  2. Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version.
  3. Release a tested prompt-model configuration manifest with both exact identifiers.
  4. Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed.

Explanation

Correct Answer

C:

Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies.

Incorrect Answers

A:

Use staged canary deployment with task and safety acceptance thresholds. This is useful for the distinct issue: A new tuned model is ready, but the team proposes replacing all production traffic instantly. The relevant technical rationale or evidence for that issue is: Controlled traffic increases allow detection of production regressions before the entire customer population is affected. Here the reported issue is: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. The relevant response is: Release a tested prompt-model configuration manifest with both exact identifiers. The first result cannot verify this different correction: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies.

B:

Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version. It addresses another case: An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. For that case, the distinguishing justification or acceptance check is: Governance traceability requires evidence beyond a model display name when customized outputs are questioned. The present case instead states: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. It calls for Release a tested prompt-model configuration manifest with both exact identifiers. Showing success for the other case would not establish: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies.

D:

The alternative, Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed, belongs to a different technical condition: The base model underlying a deployed tuned variant is approaching retirement. There the source-specific mechanism or proof is: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency. The condition asked about here is: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. A suitable correction is Release a tested prompt-model configuration manifest with both exact identifiers; the targeted justification or result is: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies.

Question 798

A team collects promising offline results but has no security or product approval for tuning on customer records. What must happen before promotion?

  1. Release a tested prompt-model configuration manifest with both exact identifiers.
  2. Require documented data-use, safety, and business-owner approvals on the immutable release candidate.
  3. Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version.
  4. Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment.

Explanation

Correct Answer

B:

Quality benchmarks do not replace governance decisions about lawful data use and permitted production risk.

Incorrect Answers

A:

Release a tested prompt-model configuration manifest with both exact identifiers. A separate source case supports that approach when: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. The supporting rationale or observable result is: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies. This question describes a different issue: A team collects promising offline results but has no security or product approval for tuning on customer records. Address it through Require documented data-use, safety, and business-owner approvals on the immutable release candidate, with the required justification or evidence: Quality benchmarks do not replace governance decisions about lawful data use and permitted production risk.

C:

Attach dataset lineage, training job identifiers, evaluation results, and approval records to the deployed version. This is useful for the distinct issue: An auditor asks which source corpus produced the current regulated assistant. The deployment has a model ID but no training record. The relevant technical rationale or evidence for that issue is: Governance traceability requires evidence beyond a model display name when customized outputs are questioned. Here the reported issue is: A team collects promising offline results but has no security or product approval for tuning on customer records. The relevant response is: Require documented data-use, safety, and business-owner approvals on the immutable release candidate. The first result cannot verify this different correction: Quality benchmarks do not replace governance decisions about lawful data use and permitted production risk.

D:

Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment. It addresses another case: A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. For that case, the distinguishing justification or acceptance check is: Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved. The present case instead states: A team collects promising offline results but has no security or product approval for tuning on customer records. It calls for Require documented data-use, safety, and business-owner approvals on the immutable release candidate. Showing success for the other case would not establish: Quality benchmarks do not replace governance decisions about lawful data use and permitted production risk.

Question 799

A new tuned model is deployed alongside the old one without serving live users; the team wants production comparisons without side effects. Which technique fits?

  1. Run shadow inference on mirrored approved traffic and compare outputs without returning experimental responses.
  2. Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment.
  3. Release a tested prompt-model configuration manifest with both exact identifiers.
  4. Protect serving endpoints with managed identity or approved auth and least-privilege network access.

Explanation

Correct Answer

A:

Shadow evaluation measures real-input behavior while the established production model continues serving users.

Incorrect Answers

B:

Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment. That operation would be relevant if: A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. Its corresponding reason or verification is: Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved. However the problem in this question is: A new tuned model is deployed alongside the old one without serving live users; the team wants production comparisons without side effects. The requirement instead calls for Run shadow inference on mirrored approved traffic and compare outputs without returning experimental responses; the question-specific check is: Shadow evaluation measures real-input behavior while the established production model continues serving users.

C:

Release a tested prompt-model configuration manifest with both exact identifiers. A separate source case supports that approach when: A prompt version and a tuned model are deployed independently, causing production prompts to mismatch trained behavior. The supporting rationale or observable result is: Reproducible application behavior requires compatibility across model weights, prompt instructions, and retrieval dependencies. This question describes a different issue: A new tuned model is deployed alongside the old one without serving live users; the team wants production comparisons without side effects. Address it through Run shadow inference on mirrored approved traffic and compare outputs without returning experimental responses, with the required justification or evidence: Shadow evaluation measures real-input behavior while the established production model continues serving users.

D:

Protect serving endpoints with managed identity or approved auth and least-privilege network access. This is useful for the distinct issue: A regulator requires that only approved applications call the tuned model. The relevant technical rationale or evidence for that issue is: A tuned model remains a service endpoint requiring normal identity and network protection controls. Here the reported issue is: A new tuned model is deployed alongside the old one without serving live users; the team wants production comparisons without side effects. The relevant response is: Run shadow inference on mirrored approved traffic and compare outputs without returning experimental responses. The first result cannot verify this different correction: Shadow evaluation measures real-input behavior while the established production model continues serving users.

Question 800

A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. What retirement action is needed?

  1. Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed.
  2. Use staged canary deployment with task and safety acceptance thresholds.
  3. Bind each trained artifact to immutable data, prompt, base model, and training configuration versions.
  4. Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment.

Explanation

Correct Answer

D:

Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved.

Incorrect Answers

A:

Plan migration to a supported base and rerun training, evaluation, and cutover tests as needed. A separate source case supports that approach when: The base model underlying a deployed tuned variant is approaching retirement. The supporting rationale or observable result is: Tuned artifacts can depend on a specific base lineage, so model retirement creates a lifecycle dependency. This question describes a different issue: A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. Address it through Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment, with the required justification or evidence: Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved.

B:

Use staged canary deployment with task and safety acceptance thresholds. This is useful for the distinct issue: A new tuned model is ready, but the team proposes replacing all production traffic instantly. The relevant technical rationale or evidence for that issue is: Controlled traffic increases allow detection of production regressions before the entire customer population is affected. Here the reported issue is: A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. The relevant response is: Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment. The first result cannot verify this different correction: Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved.

C:

Bind each trained artifact to immutable data, prompt, base model, and training configuration versions. It addresses another case: A fine-tuning run uses a dataset that changes in place and a mutable prompt repository. For that case, the distinguishing justification or acceptance check is: Without fixed inputs and parameters, engineers cannot reproduce model behavior or attribute later regressions. The present case instead states: A deprecated tuned endpoint still accepts production traffic months after migration, creating inconsistent outcomes. It calls for Drain callers, verify zero approved traffic, revoke access, and decommission the old deployment. Showing success for the other case would not establish: Completing model promotion includes retiring obsolete serving endpoints after dependencies and rollback requirements are resolved.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests