Topic 06 Practice Test 1 covers FM Deployment, Provisioned Throughput, Containers, GPUs, Cascading, and Hybrid Endpoints in Domain 2 (Task 2.2). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
A claims assistant has stable weekday demand and repeatedly hits on-demand throttling during its known peak window. Which approach best satisfies the requirement? Choose ONE.
- GPU memory headroom sizing
- Cost-aware routing threshold
- Application inference profile
- Bedrock Provisioned Throughput
Correct Answer: D
Correct Answer
Answer D is correct because Bedrock Provisioned Throughput: reserves model invocation capacity at a defined throughput level and fixed hourly cost for supported Bedrock models. T06P1Q01 needs obtain predictable dedicated model capacity for a sustained peak; it directly fits because it will purchase appropriately sized Amazon Bedrock Provisioned Throughput for the selected model.
Incorrect Answers
Answer A is incorrect because GPU memory headroom sizing: selects accelerator memory with room for model weights plus runtime structures such as KV cache and framework overhead. Best used when token generation needs memory beyond the static size of the model artifact; T06P1Q01 instead needs obtain predictable dedicated model capacity for a sustained peak, so it misses the deciding constraint.
Answer B is incorrect because Cost-aware routing threshold: uses measured task value and model cost to decide when an expensive model is justified. Best used when premium inference should be reserved for requests where added capability changes the outcome; T06P1Q01 instead needs obtain predictable dedicated model capacity for a sustained peak, so it misses the deciding constraint.
Answer C is incorrect because Application inference profile: provides a Bedrock invocation resource for tracking model usage and cost by application, optionally backed by cross-Region routing. Best used when teams need application-level usage attribution rather than new reserved capacity; T06P1Q01 instead needs obtain predictable dedicated model capacity for a sustained peak, so it misses the deciding constraint.
Question 2
A prototype is used by a small research group a few times each day, and request volume is unpredictable. What should the developer implement? Choose ONE.
- Fallback model for availability
- Token-throughput capacity planning
- Bedrock on-demand inference
- SageMaker endpoint autoscaling
Correct Answer: C
Correct Answer
Answer C is correct because Bedrock on-demand inference: uses shared Bedrock capacity without a reserved throughput commitment and is billed according to actual model use. T06P1Q02 needs avoid paying for continuously reserved model capacity during long idle periods; it directly fits because it will invoke the supported foundation model with Amazon Bedrock on-demand inference.
Incorrect Answers
Answer A is incorrect because Fallback model for availability: switches to an approved alternate model when the preferred model is unavailable or throttled, preserving a degraded but functional service. Best used when continuity matters more than using one exact model for every request; T06P1Q02 instead needs avoid paying for continuously reserved model capacity during long idle periods, so it misses the deciding constraint.
Answer B is incorrect because Token-throughput capacity planning: sizes serving capacity using input and output token rates, concurrency, and latency rather than request count alone. Best used when requests vary greatly in context and generated length; T06P1Q02 instead needs avoid paying for continuously reserved model capacity during long idle periods, so it misses the deciding constraint.
Answer D is incorrect because SageMaker endpoint autoscaling: adjusts hosted inference capacity in response to configured scaling metrics and policies. Best used when a managed endpoint has variable sustained load and should add or remove serving instances automatically; T06P1Q02 instead needs avoid paying for continuously reserved model capacity during long idle periods, so it misses the deciding constraint.
Question 3
A company must serve its own fine-tuned open model with a custom preprocessing container and VPC-controlled endpoint. Which design most directly addresses the constraint? Choose ONE.
- Lambda as thin on-demand invocation client
- Container dependency pinning
- Benchmark before model downsizing
- SageMaker real-time endpoint
Correct Answer: D
Correct Answer
Answer D is correct because SageMaker real-time endpoint: hosts a model behind a managed low-latency endpoint with configurable instances, containers, networking, and autoscaling. T06P1Q03 needs host custom model artifacts and inference code behind a managed interactive endpoint; it directly fits because it will deploy the model to an Amazon SageMaker AI real-time endpoint.
Incorrect Answers
Answer A is incorrect because Lambda as thin on-demand invocation client: runs short-lived application logic that can call a managed FM API without hosting the model inside the function. Best used when bursty event-driven requests need lightweight orchestration around a managed FM; T06P1Q03 instead needs host custom model artifacts and inference code behind a managed interactive endpoint, so it misses the deciding constraint.
Answer B is incorrect because Container dependency pinning: locks compatible serving-framework, CUDA, driver-facing, and model-library versions inside the deployment artifact. Best used when runtime drift can break model loading or GPU execution after an image rebuild; T06P1Q03 instead needs host custom model artifacts and inference code behind a managed interactive endpoint, so it misses the deciding constraint.
Answer C is incorrect because Benchmark before model downsizing: compares candidate models against representative task data and acceptance criteria before replacing a larger model. Best used when cost reduction must not silently lower task quality; T06P1Q03 instead needs host custom model artifacts and inference code behind a managed interactive endpoint, so it misses the deciding constraint.
Question 4
An application purchased Bedrock Provisioned Throughput successfully but still invokes the base model ID and consumes on-demand capacity. What is the most appropriate implementation? Choose ONE.
- Hybrid Bedrock and SageMaker deployment
- SageMaker network-isolated endpoint container
- Model sharding across accelerators
- Provisioned model ARN invocation
Correct Answer: D
Correct Answer
Answer D is correct because Provisioned model ARN invocation: identifies a purchased Bedrock Provisioned Throughput resource that can be supplied as the model identifier for inference. T06P1Q04 needs route calls through the capacity that was actually provisioned; it directly fits because it will use the provisioned model ARN as the modelId for inference.
Incorrect Answers
Answer A is incorrect because Hybrid Bedrock and SageMaker deployment: combines managed Bedrock model APIs with SageMaker-hosted custom models when one hosting approach does not fit every model. Best used when an application needs both managed foundation models and a proprietary model with custom serving code; T06P1Q04 instead needs route calls through the capacity that was actually provisioned, so it misses the deciding constraint.
Answer B is incorrect because SageMaker network-isolated endpoint container: hosts inference code with controlled VPC placement and can restrict the model container’s network access. Best used when custom inference must run under tighter network controls than a generic public-service integration; T06P1Q04 instead needs route calls through the capacity that was actually provisioned, so it misses the deciding constraint.
Answer C is incorrect because Model sharding across accelerators: partitions large model state across accelerators so no single device must store every parameter. Best used when model size exceeds one accelerator’s memory even after reasonable optimization; T06P1Q04 instead needs route calls through the capacity that was actually provisioned, so it misses the deciding constraint.
Question 5
A release pipeline creates Bedrock Provisioned Throughput and immediately switches production traffic, causing early invocation failures. Which solution best fits this production requirement? Choose ONE.
- SageMaker endpoint startup timeout tuning
- Dedicated endpoint for regulated custom model
- Provisioned Throughput readiness check
- GPU utilization monitoring
Correct Answer: C
Correct Answer
Answer C is correct because Provisioned Throughput readiness check: uses the provisioning status to distinguish a usable reserved-capacity resource from one still being created or failed. T06P1Q05 needs gate traffic cutover on the reserved-capacity resource becoming ready; it directly fits because it will poll GetProvisionedModelThroughput until the status is InService before cutover.
Incorrect Answers
Answer A is incorrect because SageMaker endpoint startup timeout tuning: extends model-download and container health-check windows so large model loading can complete before the platform declares failure. Best used when large model artifacts legitimately need more time to download and initialize; T06P1Q05 instead needs gate traffic cutover on the reserved-capacity resource becoming ready, so it misses the deciding constraint.
Answer B is incorrect because Dedicated endpoint for regulated custom model: keeps a proprietary model on controlled SageMaker hosting where the team selects compute, networking, and deployment configuration. Best used when deployment policy requires direct control over the custom inference environment; T06P1Q05 instead needs gate traffic cutover on the reserved-capacity resource becoming ready, so it misses the deciding constraint.
Answer D is incorrect because GPU utilization monitoring: tracks accelerator use to reveal whether expensive GPU capacity is saturated, idle, or blocked by another bottleneck. Best used when operators need evidence before changing expensive inference instance capacity; T06P1Q05 instead needs gate traffic cutover on the reserved-capacity resource becoming ready, so it misses the deciding constraint.
Question 6
An EU application needs more resilient on-demand Bedrock capacity, but its policy forbids processing outside the EU geography. Which approach best satisfies the requirement? Choose ONE.
- AWS Large Model Inference container
- Geographic cross-Region inference profile
- Inference profile versus Provisioned Throughput distinction
- Separate model copies for concurrency
Correct Answer: B
Correct Answer
Answer B is correct because Geographic cross-Region inference profile: allows Bedrock to route on-demand requests among supported Regions inside a defined geography. T06P1Q06 needs expand available regional capacity without violating the geographic processing boundary; it directly fits because it will invoke the model through a geographic cross-Region inference profile.
Incorrect Answers
Answer A is incorrect because AWS Large Model Inference container: provides specialized serving components and libraries for deploying and optimizing LLM inference on SageMaker. Best used when large language model serving needs purpose-built loading and runtime optimizations; T06P1Q06 instead needs expand available regional capacity without violating the geographic processing boundary, so it misses the deciding constraint.
Answer C is incorrect because Inference profile versus Provisioned Throughput distinction: separates cross-Region on-demand routing from dedicated Bedrock provisioned capacity because the two mechanisms solve different capacity problems. Best used when architects must choose between regional routing breadth and reserved model capacity; T06P1Q06 instead needs expand available regional capacity without violating the geographic processing boundary, so it misses the deciding constraint.
Answer D is incorrect because Separate model copies for concurrency: runs multiple serving replicas so independent requests can be processed concurrently when one model copy becomes a bottleneck. Best used when the model fits comfortably but a single replica cannot meet concurrent request demand; T06P1Q06 instead needs expand available regional capacity without violating the geographic processing boundary, so it misses the deciding constraint.
Question 7
A public content generator has no regional residency requirement and often sees single-Region on-demand capacity pressure. What should the developer implement? Choose ONE.
- Global cross-Region inference profile
- Tensor parallelism
- Warm capacity for strict latency SLO
- Provisioned Throughput commitment sizing
Correct Answer: A
Correct Answer
Answer A is correct because Global cross-Region inference profile: lets Bedrock route supported on-demand requests across commercial Regions for broader capacity access. T06P1Q07 needs use a wider regional capacity pool without building custom failover; it directly fits because it will invoke a supported model through a global cross-Region inference profile.
Incorrect Answers
Answer B is incorrect because Tensor parallelism: splits model tensor computations across multiple accelerators so model execution is distributed beyond one GPU. Best used when a model is too large or compute-heavy for efficient execution on a single accelerator; T06P1Q07 instead needs use a wider regional capacity pool without building custom failover, so it misses the deciding constraint.
Answer C is incorrect because Warm capacity for strict latency SLO: keeps serving resources initialized so production requests do not pay model-loading cold-start time. Best used when interactive requests have a tight latency objective incompatible with loading a large model on demand; T06P1Q07 instead needs use a wider regional capacity pool without building custom failover, so it misses the deciding constraint.
Answer D is incorrect because Provisioned Throughput commitment sizing: aligns reserved capacity and commitment term with measured sustained demand so fixed capacity is not dramatically over- or under-sized. Best used when traffic is stable enough to forecast and the team is considering a paid commitment; T06P1Q07 instead needs use a wider regional capacity pool without building custom failover, so it misses the deciding constraint.
Question 8
Three internal products invoke the same Bedrock model, and finance needs usage and cost broken out by product. Which design most directly addresses the constraint? Choose ONE.
- Application inference profile
- Smaller specialist model
- Quantized model serving
- Scale on serving bottleneck metric
Correct Answer: A
Correct Answer
Answer A is correct because Application inference profile: provides a Bedrock invocation resource for tracking model usage and cost by application, optionally backed by cross-Region routing. T06P1Q08 needs attribute model consumption to one application while keeping the underlying model deployment unchanged; it directly fits because it will create and invoke through an application inference profile for that product.
Incorrect Answers
Answer B is incorrect because Smaller specialist model: uses a lower-cost or lower-latency model that is sufficient for a narrow, well-bounded task. Best used when a simple task does not require the reasoning capability of the largest available FM; T06P1Q08 instead needs attribute model consumption to one application while keeping the underlying model deployment unchanged, so it misses the deciding constraint.
Answer C is incorrect because Quantized model serving: reduces numerical precision of model weights or activations to lower memory use and often improve inference efficiency. Best used when the selected model nearly exceeds accelerator memory and an acceptable lower-precision variant is validated; T06P1Q08 instead needs attribute model consumption to one application while keeping the underlying model deployment unchanged, so it misses the deciding constraint.
Answer D is incorrect because Scale on serving bottleneck metric: uses a metric tied to inference saturation or concurrency rather than an unrelated host signal when expanding serving capacity. Best used when autoscaling should react to the limiting resource that causes request queuing; T06P1Q08 instead needs attribute model consumption to one application while keeping the underlying model deployment unchanged, so it misses the deciding constraint.
Question 9
A custom LLM endpoint has predictable daytime surges and overnight lows, and operators want capacity to follow demand. What is the most appropriate implementation? Choose ONE.
- Load-test maximum context behavior
- Confidence-based model cascade
- Continuous batching
- SageMaker endpoint autoscaling
Correct Answer: D
Correct Answer
Answer D is correct because SageMaker endpoint autoscaling: adjusts hosted inference capacity in response to configured scaling metrics and policies. T06P1Q09 needs change SageMaker serving capacity automatically as endpoint load changes; it directly fits because it will configure automatic scaling for the SageMaker real-time endpoint.
Incorrect Answers
Answer A is incorrect because Load-test maximum context behavior: tests long prompts and outputs because memory use and token latency can be far worse than short synthetic requests. Best used when production traffic includes large contexts that materially change capacity needs; T06P1Q09 instead needs change SageMaker serving capacity automatically as endpoint load changes, so it misses the deciding constraint.
Answer B is incorrect because Confidence-based model cascade: lets a cheaper model handle routine cases and escalates uncertain or difficult cases to a stronger model. Best used when most requests are simple but a minority need more capable reasoning; T06P1Q09 instead needs change SageMaker serving capacity automatically as endpoint load changes, so it misses the deciding constraint.
Answer C is incorrect because Continuous batching: dynamically groups token-generation work from concurrent requests to improve accelerator utilization during LLM decoding. Best used when many concurrent generation requests leave GPU execution inefficient when processed independently; T06P1Q09 instead needs change SageMaker serving capacity automatically as endpoint load changes, so it misses the deciding constraint.
Question 10
An S3 event occasionally triggers a short summarization task using a Bedrock model; there is no need to host model weights. Which solution best fits this production requirement? Choose ONE.
- Model preload before readiness
- Separate interactive and batch deployment paths
- Lambda as thin on-demand invocation client
- Task-based model routing
Correct Answer: C
Correct Answer
Answer C is correct because Lambda as thin on-demand invocation client: runs short-lived application logic that can call a managed FM API without hosting the model inside the function. T06P1Q10 needs keep invocation glue serverless while leaving FM hosting to Bedrock; it directly fits because it will use a Lambda function to call the Bedrock runtime API on demand.
Incorrect Answers
Answer A is incorrect because Model preload before readiness: loads required model weights and runtime state before the endpoint advertises itself as ready to receive production traffic. Best used when large model initialization would otherwise cause cold requests to fail or time out; T06P1Q10 instead needs keep invocation glue serverless while leaving FM hosting to Bedrock, so it misses the deciding constraint.
Answer B is incorrect because Separate interactive and batch deployment paths: keeps latency-sensitive online traffic from competing directly with throughput-oriented offline generation. Best used when online and offline workloads have different latency and scheduling objectives; T06P1Q10 instead needs keep invocation glue serverless while leaving FM hosting to Bedrock, so it misses the deciding constraint.
Answer D is incorrect because Task-based model routing: selects different models according to request type so each workload uses an appropriate capability and cost profile. Best used when distinct task classes benefit from different model strengths; T06P1Q10 instead needs keep invocation glue serverless while leaving FM hosting to Bedrock, so it misses the deciding constraint.
Question 11
A regulated workload must host a custom model in a VPC and prevent the inference container from reaching the public internet. Which approach best satisfies the requirement? Choose ONE.
- Latency-first model selection
- SageMaker network-isolated endpoint container
- Explicit model version pinning
- GPU memory headroom sizing
Correct Answer: B
Correct Answer
Answer B is correct because SageMaker network-isolated endpoint container: hosts inference code with controlled VPC placement and can restrict the model container’s network access. T06P1Q11 needs serve the custom model while constraining container network access; it directly fits because it will deploy it on SageMaker with VPC configuration and network isolation.
Incorrect Answers
Answer A is incorrect because Latency-first model selection: chooses a model whose response-time characteristics satisfy the interactive experience even if a slower model has marginally higher benchmark quality. Best used when user experience has a hard response-time target; T06P1Q11 instead needs serve the custom model while constraining container network access, so it misses the deciding constraint.
Answer C is incorrect because Explicit model version pinning: keeps production behavior tied to a validated model version until a new version passes evaluation and rollout controls. Best used when uncontrolled model changes could alter output quality or compatibility; T06P1Q11 instead needs serve the custom model while constraining container network access, so it misses the deciding constraint.
Answer D is incorrect because GPU memory headroom sizing: selects accelerator memory with room for model weights plus runtime structures such as KV cache and framework overhead. Best used when token generation needs memory beyond the static size of the model artifact; T06P1Q11 instead needs serve the custom model while constraining container network access, so it misses the deciding constraint.
Question 12
A large model container is healthy after several minutes, but deployment repeatedly fails because the default startup health window expires first. What should the developer implement? Choose ONE.
- SageMaker endpoint startup timeout tuning
- Token-throughput capacity planning
- Quality-first model selection
- Bedrock Provisioned Throughput
Correct Answer: A
Correct Answer
Answer A is correct because SageMaker endpoint startup timeout tuning: extends model-download and container health-check windows so large model loading can complete before the platform declares failure. T06P1Q12 needs allow legitimate large-model initialization time without masking runtime failures; it directly fits because it will set appropriate model download and container startup health-check timeouts.
Incorrect Answers
Answer B is incorrect because Token-throughput capacity planning: sizes serving capacity using input and output token rates, concurrency, and latency rather than request count alone. Best used when requests vary greatly in context and generated length; T06P1Q12 instead needs allow legitimate large-model initialization time without masking runtime failures, so it misses the deciding constraint.
Answer C is incorrect because Quality-first model selection: chooses the higher-quality model when task errors are materially more costly than added latency or inference expense. Best used when the application can tolerate slower responses but not weaker reasoning on complex cases; T06P1Q12 instead needs allow legitimate large-model initialization time without masking runtime failures, so it misses the deciding constraint.
Answer D is incorrect because Bedrock Provisioned Throughput: reserves model invocation capacity at a defined throughput level and fixed hourly cost for supported Bedrock models. Best used when steady, business-critical traffic needs predictable capacity; T06P1Q12 instead needs allow legitimate large-model initialization time without masking runtime failures, so it misses the deciding constraint.
Question 13
A team is moving a multi-billion-parameter model to SageMaker and wants AWS-supported tooling for optimized LLM serving rather than a generic web container. Which design most directly addresses the constraint? Choose ONE.
- Cost-aware routing threshold
- Bedrock on-demand inference
- Container dependency pinning
- AWS Large Model Inference container
Correct Answer: D
Correct Answer
Answer D is correct because AWS Large Model Inference container: provides specialized serving components and libraries for deploying and optimizing LLM inference on SageMaker. T06P1Q13 needs use a serving stack designed for large-model inference characteristics; it directly fits because it will deploy with an appropriate SageMaker Large Model Inference container.
Incorrect Answers
Answer A is incorrect because Cost-aware routing threshold: uses measured task value and model cost to decide when an expensive model is justified. Best used when premium inference should be reserved for requests where added capability changes the outcome; T06P1Q13 instead needs use a serving stack designed for large-model inference characteristics, so it misses the deciding constraint.
Answer B is incorrect because Bedrock on-demand inference: uses shared Bedrock capacity without a reserved throughput commitment and is billed according to actual model use. Best used when traffic is intermittent or uncertain and dedicated capacity would sit idle; T06P1Q13 instead needs use a serving stack designed for large-model inference characteristics, so it misses the deciding constraint.
Answer C is incorrect because Container dependency pinning: locks compatible serving-framework, CUDA, driver-facing, and model-library versions inside the deployment artifact. Best used when runtime drift can break model loading or GPU execution after an image rebuild; T06P1Q13 instead needs use a serving stack designed for large-model inference characteristics, so it misses the deciding constraint.
Question 14
A custom LLM cannot fit its working model state on one GPU, but the selected instance provides several GPUs with high-speed interconnects. What is the most appropriate implementation? Choose ONE.
- Tensor parallelism
- Fallback model for availability
- SageMaker real-time endpoint
- Model sharding across accelerators
Correct Answer: A
Correct Answer
Answer A is correct because Tensor parallelism: splits model tensor computations across multiple accelerators so model execution is distributed beyond one GPU. T06P1Q14 needs distribute a single model’s computation across multiple GPUs; it directly fits because it will configure a supported tensor-parallel large-model serving strategy.
Incorrect Answers
Answer B is incorrect because Fallback model for availability: switches to an approved alternate model when the preferred model is unavailable or throttled, preserving a degraded but functional service. Best used when continuity matters more than using one exact model for every request; T06P1Q14 instead needs distribute a single model’s computation across multiple GPUs, so it misses the deciding constraint.
Answer C is incorrect because SageMaker real-time endpoint: hosts a model behind a managed low-latency endpoint with configurable instances, containers, networking, and autoscaling. Best used when a team needs control of model artifacts or custom inference code; T06P1Q14 instead needs distribute a single model’s computation across multiple GPUs, so it misses the deciding constraint.
Answer D is incorrect because Model sharding across accelerators: partitions large model state across accelerators so no single device must store every parameter. Best used when model size exceeds one accelerator’s memory even after reasonable optimization; T06P1Q14 instead needs distribute a single model’s computation across multiple GPUs, so it misses the deciding constraint.
Question 15
A tested 8-bit version of an open model meets quality requirements while the full-precision build does not fit economically on the target GPU fleet. Which solution best fits this production requirement? Choose ONE.
- Benchmark before model downsizing
- Quantized model serving
- Provisioned model ARN invocation
- GPU utilization monitoring
Correct Answer: B
Correct Answer
Answer B is correct because Quantized model serving: reduces numerical precision of model weights or activations to lower memory use and often improve inference efficiency. T06P1Q15 needs reduce memory pressure while preserving validated task quality; it directly fits because it will deploy the validated quantized model representation.
Incorrect Answers
Answer A is incorrect because Benchmark before model downsizing: compares candidate models against representative task data and acceptance criteria before replacing a larger model. Best used when cost reduction must not silently lower task quality; T06P1Q15 instead needs reduce memory pressure while preserving validated task quality, so it misses the deciding constraint.
Answer C is incorrect because Provisioned model ARN invocation: identifies a purchased Bedrock Provisioned Throughput resource that can be supplied as the model identifier for inference. Best used when a provisioned resource is already InService and applications must use that reserved capacity; T06P1Q15 instead needs reduce memory pressure while preserving validated task quality, so it misses the deciding constraint.
Answer D is incorrect because GPU utilization monitoring: tracks accelerator use to reveal whether expensive GPU capacity is saturated, idle, or blocked by another bottleneck. Best used when operators need evidence before changing expensive inference instance capacity; T06P1Q15 instead needs reduce memory pressure while preserving validated task quality, so it misses the deciding constraint.
Question 16
An LLM endpoint serves many simultaneous chats and GPU utilization is low because requests are scheduled one at a time. Which approach best satisfies the requirement? Choose ONE.
- Separate model copies for concurrency
- Continuous batching
- Hybrid Bedrock and SageMaker deployment
- Provisioned Throughput readiness check
Correct Answer: B
Correct Answer
Answer B is correct because Continuous batching: dynamically groups token-generation work from concurrent requests to improve accelerator utilization during LLM decoding. T06P1Q16 needs improve token-serving throughput across concurrent generation requests; it directly fits because it will enable a supported continuous-batching serving engine.
Incorrect Answers
Answer A is incorrect because Separate model copies for concurrency: runs multiple serving replicas so independent requests can be processed concurrently when one model copy becomes a bottleneck. Best used when the model fits comfortably but a single replica cannot meet concurrent request demand; T06P1Q16 instead needs improve token-serving throughput across concurrent generation requests, so it misses the deciding constraint.
Answer C is incorrect because Hybrid Bedrock and SageMaker deployment: combines managed Bedrock model APIs with SageMaker-hosted custom models when one hosting approach does not fit every model. Best used when an application needs both managed foundation models and a proprietary model with custom serving code; T06P1Q16 instead needs improve token-serving throughput across concurrent generation requests, so it misses the deciding constraint.
Answer D is incorrect because Provisioned Throughput readiness check: uses the provisioning status to distinguish a usable reserved-capacity resource from one still being created or failed. Best used when deployment automation must wait until reserved capacity is available before cutting traffic; T06P1Q16 instead needs improve token-serving throughput across concurrent generation requests, so it misses the deciding constraint.
Question 17
A new container passes its process health check before model weights finish loading, so the first user requests time out. What should the developer implement? Choose ONE.
- Geographic cross-Region inference profile
- Dedicated endpoint for regulated custom model
- Model preload before readiness
- Warm capacity for strict latency SLO
Correct Answer: C
Correct Answer
Answer C is correct because Model preload before readiness: loads required model weights and runtime state before the endpoint advertises itself as ready to receive production traffic. T06P1Q17 needs withhold readiness until the model is actually loaded for inference; it directly fits because it will make the serving container report ready only after model initialization completes.
Incorrect Answers
Answer A is incorrect because Geographic cross-Region inference profile: allows Bedrock to route on-demand requests among supported Regions inside a defined geography. Best used when a workload can use multiple Regions but must keep processing within a permitted geography; T06P1Q17 instead needs withhold readiness until the model is actually loaded for inference, so it misses the deciding constraint.
Answer B is incorrect because Dedicated endpoint for regulated custom model: keeps a proprietary model on controlled SageMaker hosting where the team selects compute, networking, and deployment configuration. Best used when deployment policy requires direct control over the custom inference environment; T06P1Q17 instead needs withhold readiness until the model is actually loaded for inference, so it misses the deciding constraint.
Answer D is incorrect because Warm capacity for strict latency SLO: keeps serving resources initialized so production requests do not pay model-loading cold-start time. Best used when interactive requests have a tight latency objective incompatible with loading a large model on demand; T06P1Q17 instead needs withhold readiness until the model is actually loaded for inference, so it misses the deciding constraint.
Question 18
A model’s weights technically fit on a GPU, but long-context requests cause out-of-memory failures during generation. Which design most directly addresses the constraint? Choose ONE.
- Inference profile versus Provisioned Throughput distinction
- GPU memory headroom sizing
- Global cross-Region inference profile
- Smaller specialist model
Correct Answer: B
Correct Answer
Answer B is correct because GPU memory headroom sizing: selects accelerator memory with room for model weights plus runtime structures such as KV cache and framework overhead. T06P1Q18 needs size GPU memory for weights plus inference-time cache and runtime overhead; it directly fits because it will choose an accelerator configuration with sufficient inference memory headroom.
Incorrect Answers
Answer A is incorrect because Inference profile versus Provisioned Throughput distinction: separates cross-Region on-demand routing from dedicated Bedrock provisioned capacity because the two mechanisms solve different capacity problems. Best used when architects must choose between regional routing breadth and reserved model capacity; T06P1Q18 instead needs size GPU memory for weights plus inference-time cache and runtime overhead, so it misses the deciding constraint.
Answer C is incorrect because Global cross-Region inference profile: lets Bedrock route supported on-demand requests across commercial Regions for broader capacity access. Best used when data residency does not restrict processing and the workload values broad regional capacity and routing; T06P1Q18 instead needs size GPU memory for weights plus inference-time cache and runtime overhead, so it misses the deciding constraint.
Answer D is incorrect because Smaller specialist model: uses a lower-cost or lower-latency model that is sufficient for a narrow, well-bounded task. Best used when a simple task does not require the reasoning capability of the largest available FM; T06P1Q18 instead needs size GPU memory for weights plus inference-time cache and runtime overhead, so it misses the deciding constraint.
Question 19
Two applications issue the same number of requests, but one uses 20-times-longer prompts and outputs and repeatedly saturates the endpoint. What is the most appropriate implementation? Choose ONE.
- Token-throughput capacity planning
- Provisioned Throughput commitment sizing
- Application inference profile
- Confidence-based model cascade
Correct Answer: A
Correct Answer
Answer A is correct because Token-throughput capacity planning: sizes serving capacity using input and output token rates, concurrency, and latency rather than request count alone. T06P1Q19 needs plan capacity around token processing demand rather than raw request rate; it directly fits because it will measure and size for input/output token throughput and concurrency.
Incorrect Answers
Answer B is incorrect because Provisioned Throughput commitment sizing: aligns reserved capacity and commitment term with measured sustained demand so fixed capacity is not dramatically over- or under-sized. Best used when traffic is stable enough to forecast and the team is considering a paid commitment; T06P1Q19 instead needs plan capacity around token processing demand rather than raw request rate, so it misses the deciding constraint.
Answer C is incorrect because Application inference profile: provides a Bedrock invocation resource for tracking model usage and cost by application, optionally backed by cross-Region routing. Best used when teams need application-level usage attribution rather than new reserved capacity; T06P1Q19 instead needs plan capacity around token processing demand rather than raw request rate, so it misses the deciding constraint.
Answer D is incorrect because Confidence-based model cascade: lets a cheaper model handle routine cases and escalates uncertain or difficult cases to a stronger model. Best used when most requests are simple but a minority need more capable reasoning; T06P1Q19 instead needs plan capacity around token processing demand rather than raw request rate, so it misses the deciding constraint.
Question 20
A rebuilt inference image unexpectedly upgrades a model-serving library and a previously working LLM no longer loads. Which solution best fits this production requirement? Choose ONE.
- Task-based model routing
- SageMaker endpoint autoscaling
- Container dependency pinning
- Scale on serving bottleneck metric
Correct Answer: C
Correct Answer
Answer C is correct because Container dependency pinning: locks compatible serving-framework, CUDA, driver-facing, and model-library versions inside the deployment artifact. T06P1Q20 needs make the serving environment reproducible across releases; it directly fits because it will pin tested container dependencies and image versions.
Incorrect Answers
Answer A is incorrect because Task-based model routing: selects different models according to request type so each workload uses an appropriate capability and cost profile. Best used when distinct task classes benefit from different model strengths; T06P1Q20 instead needs make the serving environment reproducible across releases, so it misses the deciding constraint.
Answer B is incorrect because SageMaker endpoint autoscaling: adjusts hosted inference capacity in response to configured scaling metrics and policies. Best used when a managed endpoint has variable sustained load and should add or remove serving instances automatically; T06P1Q20 instead needs make the serving environment reproducible across releases, so it misses the deciding constraint.
Answer D is incorrect because Scale on serving bottleneck metric: uses a metric tied to inference saturation or concurrency rather than an unrelated host signal when expanding serving capacity. Best used when autoscaling should react to the limiting resource that causes request queuing; T06P1Q20 instead needs make the serving environment reproducible across releases, so it misses the deciding constraint.
