AWS AIP-C01 开源练习 · 困难 · 第 02 套
英文原题与中文机器翻译。本批暂不提供日文题文；日语界面显示英文原文。
开源练习，非 AWS 官方试题。按作者参考答案判分，解析附官方文档链接。
Chunks were configured to be very large, and retrieval now returns passages where the relevant sentence is buried among unrelated text, weakening match quality. Which change is MOST appropriate?
块被配置为非常大，检索现在返回的段落中相关句子被隐藏在不相关的文本中，从而削弱了匹配质量。哪种改变最合适？
- A
Reduce the chunk size so each chunk represents a more focused unit of meaning.
减少块大小，使每个块代表一个更集中的意义单元。
- B
Increase the chunk size further.
进一步增加块大小。
- C
Embed the whole document as one chunk.
将整个文档作为一个块嵌入。
- D
Raise the temperature.
提高温度。
解析 / Explanation
正确答案 / Answer: A
A. Smaller, focused chunks produce embeddings that better represent specific content, improving match precision.
B. Even larger chunks dilute the embedding more and worsen precision.
C. A single document-level vector is the extreme case of dilution and reduces precision.
D. Temperature is unrelated to chunk granularity.
A. 较小的、集中的块产生的嵌入可以更好地表示特定内容，从而提高匹配精度。
B. 更大的块会进一步稀释嵌入并降低精度。
C. 单个文档级向量是稀释的极端情况，会降低精度。
D. 温度与块粒度无关。
A knowledge base ingests a large set of short, self-contained FAQ entries, where each file already represents one complete question-and-answer. Which chunking choice is MOST appropriate?
知识库吸收大量简短的、独立的常见问题解答条目，其中每个文件已经代表一个完整的问题和答案。哪种分块选择最合适？
- A
Use a no-chunking strategy so each short file is treated as a single chunk.
使用无分块策略，以便将每个短文件视为单个块。
- B
Apply aggressive small fixed-size chunking to every file.
对每个文件应用积极的小型固定大小分块。
- C
Merge all FAQ files into one chunk.
将所有常见问题解答文件合并为一个块。
- D
Increase embedding dimensions to compensate.
增加嵌入尺寸以进行补偿。
解析 / Explanation
正确答案 / Answer: A
A. When each file is already an atomic, self-contained unit, no chunking preserves the complete meaning per entry without splitting it.
B. Splitting already-atomic FAQ entries fragments coherent answers and harms retrieval.
C. Combining unrelated entries into one chunk dilutes each entry's embedding.
D. Dimension size does not fix inappropriately split or merged atomic entries.
A. 当每个文件已经是一个原子的、独立的单元时，任何分块都不会在不分割它的情况下保留每个条目的完整含义。
B. 拆分已经原子的常见问题解答条目会破坏连贯的答案并损害检索。
C. 将不相关的条目组合成一个块会稀释每个条目的嵌入。
D. 维度大小不能修复不适当的分割或合并的原子条目。
An application must retrieve relevant items from a combined catalog of product images and text descriptions using a single query that may be text or an image. Which embedding approach supports this?
应用程序必须使用单个查询（可以是文本或图像）从产品图像和文本描述的组合目录中检索相关项目。哪种嵌入方法支持这一点？
- A
Use a multimodal embedding model (such as Amazon Titan Multimodal Embeddings) so images and text share one vector space.
使用多模态嵌入模型（例如 Amazon Titan Multimodal Embeddings），以便图像和文本共享一个向量空间。
- B
Use a text-only embedding model for everything.
对所有内容都使用纯文本嵌入模型。
- C
Use Amazon Rekognition labels as the only index.
使用 Amazon Rekognition 标签作为唯一索引。
- D
Store images as base64 strings and embed the strings as text.
将图像存储为 base64 字符串并将字符串嵌入为文本。
解析 / Explanation
正确答案 / Answer: A
A. Multimodal embeddings place images and text in a shared space, enabling cross-modal similarity search from either input type.
B. A text-only model cannot embed images for similarity search.
C. Label detection is not a shared embedding space for semantic text-and-image retrieval.
D. Embedding raw base64 as text does not capture visual semantics for retrieval.
A. 多模态嵌入将图像和文本放置在共享空间中，从而能够从任一输入类型进行跨模态相似性搜索。
B. 纯文本模型无法嵌入图像进行相似性搜索。
C. 标签检测不是语义文本和图像检索的共享嵌入空间。
D. 将原始 base64 嵌入为文本不会捕获用于检索的视觉语义。
A team wants to raise the overall quality of what their RAG system retrieves and ultimately answers. Which TWO techniques are most effective? (Select TWO.)
团队希望提高 RAG 系统检索和最终答案的整体质量。哪两种技术最有效？ （选择两项。）
- A
Choose a chunking strategy (semantic or hierarchical) aligned to the documents' structure.
选择与文档结构一致的分块策略（语义或分层）。
- B
Add a reranking step to reorder retrieved candidates by query relevance.
添加重新排名步骤以按查询相关性对检索到的候选者重新排序。
- C
Raise the generation temperature to add variety.
提高生成温度以增加品种。
- D
Increase max tokens on the response.
增加响应的最大标记。
- E
Always return the fewest possible results.
始终返回尽可能少的结果。
解析 / Explanation
正确答案 / Answer: A,B
A. Chunking that respects document structure produces more coherent, relevant chunks and better retrieval.
B. A reranker reorders the initial candidate set so the most relevant passages are used for generation.
C. Higher temperature increases randomness and does not improve retrieval quality.
D. Output length does not change which passages are retrieved or how relevant they are.
E. Minimizing results can drop relevant passages and reduce answer quality.
A. 尊重文档结构的分块会产生更连贯、相关的块和更好的检索。
B. 重新排序器对初始候选集重新排序，以便使用最相关的段落来生成。
C. 较高的温度会增加随机性，但不会提高检索质量。
D. 输出长度不会改变检索的段落或它们的相关程度。
E. 最小化结果可能会删除相关段落并降低答案质量。
A team wants to A/B test two wordings of a production prompt and roll back instantly if one underperforms, without redeploying application code. Which Amazon Bedrock capability supports this?
一个团队希望对生产提示的两个措辞进行 A/B 测试，如果其中一个措辞表现不佳，则立即回滚，而无需重新部署应用程序代码。哪种 Amazon Bedrock 功能支持此功能？
- A
Amazon Bedrock Prompt Management with versioned prompts and variants.
具有版本化提示和变体的 Amazon Bedrock 提示管理。
- B
Hard-code both prompts in the application and redeploy to switch.
在应用程序中对两个提示进行硬编码并重新部署以进行切换。
- C
Change the temperature parameter to switch prompts.
更改温度参数即可切换提示。
- D
Store prompts in environment variables only.
仅将提示存储在环境变量中。
解析 / Explanation
正确答案 / Answer: A
A. Prompt Management stores prompts as versioned, parameterized resources with variants, so teams can compare wordings and switch or roll back without code changes.
B. Hard-coding requires a code change and redeploy for every switch or rollback.
C. Temperature does not select between prompt wordings.
D. Environment variables lack versioning, variants, and managed rollback for prompts.
A. 提示管理将提示存储为带有变体的版本化、参数化资源，因此团队可以比较措辞并切换或回滚，而无需更改代码。
B. 硬编码需要对每次切换或回滚进行代码更改和重新部署。
C. 温度不选择提示文字。
D. 环境变量缺乏提示的版本控制、变体和托管回滚。
A developer must cap the length of model responses to control cost and prevent overly long outputs in a chat feature. Which inference setting should they use?
开发人员必须限制模型响应的长度，以控制成本并防止聊天功能中的输出过长。他们应该使用哪种推理设置？
- A
Set the maxTokens (maximum output tokens) parameter.
设置 maxTokens（最大输出令牌）参数。
- B
Raise the temperature.
提高温度。
- C
Increase top-k.
增加top-k。
- D
Add more stop sequences for unrelated phrases.
为不相关的短语添加更多停止序列。
解析 / Explanation
正确答案 / Answer: A
A. maxTokens bounds the number of tokens the model can generate, directly limiting response length and cost.
B. Temperature changes randomness, not output length or cost ceiling.
C. Top-k widens token choice diversity; it does not bound length.
D. Unrelated stop sequences do not reliably cap general response length.
A. maxTokens 限制了模型可以生成的令牌数量，直接限制了响应长度和成本。
B. 温度变化是随机的，而不是输出长度或成本上限。
C. Top-k 拓宽了代币选择的多样性；它不限制长度。
D. 不相关的停止序列不能可靠地限制一般响应长度。
An assistant must consistently follow durable behavior rules (role, tone, safety constraints) across every turn of a conversation, separate from each user's question. Where should these durable instructions be placed?
助理必须在每次对话中始终遵循持久的行为规则（角色、语气、安全限制），与每个用户的问题分开。这些持久的说明应该放在哪里？
- A
In the system prompt.
系统提示中。
- B
Appended to each user message by the user.
由用户附加到每条用户消息中。
- C
In the temperature setting.
在温度设定中。
- D
In the stop sequences list.
在停止序列列表中。
解析 / Explanation
正确答案 / Answer: A
A. The system prompt sets persistent instructions and context that govern the model's behavior across the conversation, distinct from per-turn user messages.
B. Relying on users to repeat rules is unreliable and inconsistent.
C. Temperature is a sampling parameter, not a place for behavior instructions.
D. Stop sequences end generation; they do not encode behavior rules.
A. 系统提示设置持久指令和上下文，用于控制模型在整个对话中的行为，这与每轮用户消息不同。
B. 依赖用户重复规则是不可靠且不一致的。
C. 温度是一个采样参数，而不是行为指令的地方。
D. 停止序列结束生成；他们不编码行为规则。
Outputs must follow a precise structured format, and instructions alone have not produced consistent results. Which prompt-engineering technique most improves format adherence?
输出必须遵循精确的结构化格式，仅靠指令并不能产生一致的结果。哪种即时工程技术最能提高格式的一致性？
- A
Provide a few worked examples (few-shot exemplars) demonstrating the exact desired format.
提供一些有效的示例（少数样本）来演示确切的所需格式。
- B
Increase the temperature for more creative formatting.
提高温度以实现更具创意的格式化。
- C
Remove all instructions and let the model decide.
删除所有指令并让模型决定。
- D
Shorten the prompt to a single sentence.
将提示缩短为单个句子。
解析 / Explanation
正确答案 / Answer: A
A. Few-shot examples show the model the precise structure to imitate, improving consistency beyond instructions alone.
B. Higher temperature increases variability and undermines consistent formatting.
C. Removing guidance reduces, not improves, format adherence.
D. Over-shortening removes the structure cues the model needs.
A. 少量示例向模型展示了可模仿的精确结构，从而提高了指令的一致性。
B. 较高的温度会增加变异性并破坏一致的格式。
C. 删除指导会减少而不是改善格式的遵守情况。
D. 过度缩短会消除模型所需的结构线索。
Several applications reuse the same prompt with only a few values changing per request (customer name, product, locale). The team wants to manage this prompt centrally and inject values at runtime. Which approach fits best?
多个应用程序重复使用相同的提示，每个请求仅更改几个值（客户名称、产品、区域设置）。团队希望集中管理此提示并在运行时注入值。哪种方法最适合？
- A
Use Amazon Bedrock Prompt Management templates with input variables.
将 Amazon Bedrock Prompt Management 模板与输入变量结合使用。
- B
Copy the prompt text into each application and edit by hand.
将提示文本复制到每个应用程序中并手动编辑。
- C
Concatenate strings ad hoc in every service.
在每个服务中临时连接字符串。
- D
Raise the temperature to make the prompt flexible.
提高温度，使提示灵活。
解析 / Explanation
正确答案 / Answer: A
A. Prompt Management supports parameterized templates with input variables, so a single managed prompt is reused and values are supplied at invocation time.
B. Duplicated, hand-edited prompts drift apart and are hard to maintain or update centrally.
C. Ad hoc concatenation gives no central governance, versioning, or reuse.
D. Temperature does not parameterize or manage prompts.
A. 提示管理支持带有输入变量的参数化模板，因此可以重用单个托管提示并在调用时提供值。
B. 重复的、手工编辑的提示分散开来，并且很难集中维护或更新。
C. 即席串联不提供集中治理、版本控制或重用。
D. 温度不会参数化或管理提示。
A developer needs to orchestrate a multi-step generative workflow (classify input, retrieve context, then generate a response) with branching, and wants a managed, low-code way to build and version it. Which Amazon Bedrock feature is designed for this?
开发人员需要通过分支编排多步骤生成工作流（对输入进行分类、检索上下文，然后生成响应），并需要一种托管的低代码方式来构建和版本控制它。 Amazon Bedrock 的哪些功能是为此设计的？
- A
Amazon Bedrock Flows (prompt flows).
亚马逊基岩流（即时流）。
- B
A single InvokeModel call.
单个 InvokeModel 调用。
- C
Manually wiring Lambda functions for every step from scratch.
从头开始为每个步骤手动连接 Lambda 函数。
- D
Increasing max tokens.
增加最大令牌。
解析 / Explanation
正确答案 / Answer: A
A. Bedrock Flows let you visually build and version multi-step, branching generative workflows that chain prompts, knowledge bases, and other components with little code.
B. One model call cannot orchestrate a multi-step, branching workflow.
C. Hand-built orchestration adds development and maintenance overhead versus a managed flow builder.
D. Output length has nothing to do with workflow orchestration.
A. Bedrock Flows 让您可以直观地构建和版本化多步骤、分支生成工作流程，只需很少的代码即可链接提示、知识库和其他组件。
B. 一个模型调用无法编排多步骤、分支工作流程。
C. 与托管流程构建器相比，手工构建的编排增加了开发和维护开销。
D. 输出长度与工作流程编排无关。
For a regression test suite, a team needs the same prompt to return as close to the same output as possible on each run. Which single inference parameter change most increases determinism?
对于回归测试套件，团队需要相同的提示，以便在每次运行时返回尽可能接近相同的输出。哪个单一推理参数的变化最能增加确定性？
- A
Set the temperature to 0 (or very near 0).
将温度设置为 0（或非常接近 0）。
- B
Set a very high top-k.
设置非常高的top-k。
- C
Increase max tokens.
增加最大令牌数。
- D
Raise the temperature to 1.
将温度升至 1。
解析 / Explanation
正确答案 / Answer: A
A. A temperature at or near 0 makes sampling effectively greedy, producing the most consistent, repeatable outputs.
B. A high top-k widens the sampling pool and increases variability.
C. Output length does not control run-to-run consistency.
D. A high temperature maximizes randomness, the opposite of determinism.
A. 0 或接近 0 的温度使采样有效地贪婪，产生最一致、可重复的输出。
B. 高 top-k 扩大了采样池并增加了可变性。
C. 输出长度不控制运行之间的一致性。
D. 高温使随机性最大化，这与决定论相反。
In a RAG assistant, the team wants to reduce fabricated answers by making the model rely only on the supplied context and admit when the answer is not present. Which prompt technique helps MOST?
在 RAG 助手中，团队希望通过使模型仅依赖于提供的上下文来减少捏造的答案，并在答案不存在时承认。哪种提示技术最有帮助？
- A
Instruct the model to answer only from the provided context and to say it does not know when the context lacks the answer.
指示模型仅根据提供的上下文进行回答，并表示它不知道上下文何时缺少答案。
- B
Raise the temperature so answers are more creative.
提高温度，让答案更有创意。
- C
Remove the retrieved context to keep prompts short.
删除检索到的上下文以使提示保持简短。
- D
Tell the model to always produce an answer no matter what.
告诉模型无论如何都要给出答案。
解析 / Explanation
正确答案 / Answer: A
A. Explicit grounding instructions plus an allowed 'I don't know' response reduce fabrication when the retrieved context does not contain the answer.
B. More creativity increases the chance of fabricated content.
C. Without context the model must guess, increasing hallucination.
D. Forcing an answer encourages fabrication when information is missing.
A. 当检索到的上下文不包含答案时，明确的基础指令加上允许的“我不知道”响应可以减少捏造。
B. 更多的创造力增加了捏造内容的机会。
C. 在没有上下文的情况下，模型必须猜测，从而增加幻觉。
D. 当信息缺失时，强行给出答案会鼓励捏造事实。
A developer wants to prevent unbounded, runaway generations that drive up cost and sometimes ramble past the intended answer. Which TWO inference controls directly help? (Select TWO.)
开发人员希望防止无限制的、失控的生成，这会增加成本，有时甚至会超出预期的答案。哪两个推理控制有直接帮助？ （选择两项。）
- A
Set a sensible maxTokens limit.
设置合理的 maxTokens 限制。
- B
Define stop sequences that end generation at a known marker.
定义在已知标记处结束生成的停止序列。
- C
Increase the temperature.
提高温度。
- D
Increase top-k substantially.
大幅增加top-k。
- E
Enlarge the context window.
放大上下文窗口。
解析 / Explanation
正确答案 / Answer: A,B
A. maxTokens caps how many tokens can be generated, bounding cost and preventing runaway length.
B. Stop sequences halt generation when a defined phrase appears, preventing the model from rambling past the answer.
C. Higher temperature does not limit length and can make output more verbose and variable.
D. A larger top-k affects token diversity, not output length.
E. A bigger input window does not constrain how long the output can be.
A. maxTokens 限制了可以生成的代币数量，限制成本并防止长度失控。
B. 当出现定义的短语时，停止序列会停止生成，从而防止模型杂乱无章地跳过答案。
C. 较高的温度不会限制长度，并且会使输出更加冗长和可变。
D. 较大的 top-k 会影响 token 多样性，而不是输出长度。
E. 更大的输入窗口不会限制输出的长度。
A developer is building an Amazon Bedrock agent that must call an internal order-management API to look up and update orders. What is the standard way to give the agent this capability?
开发人员正在构建一个 Amazon Bedrock 代理，该代理必须调用内部订单管理 API 来查找和更新订单。为代理提供此功能的标准方法是什么？
- A
Define an action group backed by an AWS Lambda function, described with an API (OpenAPI) or function schema the agent can call.
定义由 AWS Lambda 函数支持的操作组，使用代理可以调用的 API (OpenAPI) 或函数架构进行描述。
- B
Paste the API documentation into the agent's instructions and hope it calls the API.
将API文档粘贴到代理的说明中并希望它调用API。
- C
Increase the model's temperature so it can reach the API.
提高模型的温度，使其能够达到 API。
- D
Store the API responses in the prompt as static text.
将提示中的 API 响应存储为静态文本。
解析 / Explanation
正确答案 / Answer: A
A. Bedrock agents use action groups: a schema declares the available operations and a Lambda function executes them when the agent decides to call the API.
B. Instructions alone do not give the agent an executable tool; action groups provide the actual invocation path.
C. Temperature has nothing to do with tool or API access.
D. Static text cannot perform live lookups or updates against the order system.
A. Bedrock 代理使用操作组：架构声明可用的操作，并在代理决定调用 API 时由 Lambda 函数执行这些操作。
B. 仅指令并不能为代理提供可执行工具；操作组提供实际的调用路径。
C. 温度与工具或 API 访问无关。
D. 静态文本无法针对订单系统执行实时查找或更新。
For compliance, an organization wants its own application to execute tool/API calls rather than letting the Bedrock agent invoke a Lambda directly, while still letting the agent decide which action to take. Which agent capability supports this?
为了合规性，组织希望自己的应用程序执行工具/API 调用，而不是让 Bedrock 代理直接调用 Lambda，同时仍然让代理决定采取哪个操作。哪种代理功能支持此功能？
- A
Configure the action group to return control to the caller so the application performs the action and returns results to the agent.
配置操作组以将控制权返回给调用者，以便应用程序执行操作并将结果返回给代理。
- B
Disable the agent's reasoning entirely.
完全禁用代理的推理。
- C
Hard-code the action in the system prompt.
对系统提示中的操作进行硬编码。
- D
Lower the temperature to 0.
将温度降至 0。
解析 / Explanation
正确答案 / Answer: A
A. Return of control hands the chosen action and parameters back to the calling application to execute, instead of the agent invoking a Lambda directly.
B. Disabling reasoning would remove the agent's ability to decide which action to take.
C. A hard-coded action removes the agent's dynamic decision-making and is not the return-of-control mechanism.
D. Temperature does not control where tool execution happens.
A. 控制权返回将所选操作和参数交回调用应用程序以执行，而不是代理直接调用 Lambda。
B. 禁用推理将消除智能体决定采取哪种行动的能力。
C. 硬编码操作消除了代理的动态决策，并且不是控制权返回机制。
D. 温度并不控制工具执行的位置。
An agent should ground its answers in the company's policy documents in addition to taking actions. With the LEAST custom code, how should the developer add this knowledge?
除了采取行动外，代理人还应该根据公司的政策文件给出答案。使用最少的自定义代码，开发人员应该如何添加这些知识？
- A
Associate an Amazon Bedrock knowledge base with the agent.
将 Amazon Bedrock 知识库与代理关联。
- B
Embed the entire policy library into the agent's instructions.
将整个策略库嵌入到代理的指令中。
- C
Fine-tune the agent's model on the policies each week.
每周根据保单微调代理模型。
- D
Store policies in DynamoDB and query them with custom code.
将策略存储在 DynamoDB 中并使用自定义代码查询它们。
解析 / Explanation
正确答案 / Answer: A
A. Associating a knowledge base lets the agent retrieve and ground responses on the documents without custom retrieval code.
B. Instructions have limited size and cannot scale to a full document library or stay current.
C. Repeated fine-tuning is heavy and unnecessary when a knowledge base provides current grounding.
D. A custom store and retrieval code adds work that the built-in knowledge base association avoids.
A. 关联知识库使代理可以检索文档并根据文档做出响应，而无需自定义检索代码。
B. 说明的大小有限，无法扩展到完整的文档库或保持最新状态。
C. 当知识库提供当前的基础时，重复的微调是繁重且不必要的。
D. 自定义存储和检索代码增加了内置知识库关联所避免的工作。
A team is deploying agents on Amazon Bedrock AgentCore and wants each user's agent to remember preferences and prior context across separate sessions, as a managed capability. Which AgentCore component should they use?
一个团队正在 Amazon Bedrock AgentCore 上部署代理，并希望每个用户的代理能够记住各个会话中的首选项和先前上下文，作为一项托管功能。他们应该使用哪个 AgentCore 组件？
- A
AgentCore Memory.
代理核心内存。
- B
AgentCore Browser.
AgentCore 浏览器。
- C
AgentCore Code Interpreter.
AgentCore 代码解释器。
- D
AgentCore Gateway.
AgentCore 网关。
解析 / Explanation
正确答案 / Answer: A
A. AgentCore Memory provides managed short-term and long-term memory so agents can persist and recall context across sessions.
B. The Browser tool provides managed web browsing, not persistent agent memory.
C. Code Interpreter runs code in a sandbox; it does not store cross-session memory.
D. Gateway exposes APIs as tools; it is not the memory store.
A. AgentCore Memory 提供托管的短期和长期内存，因此代理可以跨会话保留和调用上下文。
B. 浏览器工具提供托管 Web 浏览，而不是持久代理内存。
C. 代码解释器在沙箱中运行代码；它不存储跨会话内存。
D. 网关将 API 作为工具公开；它不是内存存储。
A company already has many internal REST APIs and Lambda functions and wants to expose them to its agents as secure, governed tools without writing custom integration code for each one. Which AgentCore capability is designed for this?
一家公司已经拥有许多内部 REST API 和 Lambda 函数，并希望将它们作为安全、受管控的工具公开给其代理，而无需为每个函数编写自定义集成代码。哪种 AgentCore 功能是为此设计的？
- A
AgentCore Gateway.
AgentCore 网关。
- B
AgentCore Memory.
代理核心内存。
- C
AgentCore Code Interpreter.
AgentCore 代码解释器。
- D
AgentCore Observability.
AgentCore 可观察性。
解析 / Explanation
正确答案 / Answer: A
A. AgentCore Gateway turns existing APIs, Lambda functions, and services into agent-ready tools (for example via MCP) with built-in authentication and governance.
B. Memory persists context; it does not expose APIs as tools.
C. Code Interpreter executes code, not API-to-tool exposure.
D. Observability provides tracing and monitoring, not tool integration.
A. AgentCore Gateway 将现有 API、Lambda 函数和服务转变为具有内置身份验证和治理的代理就绪工具（例如通过 MCP）。
B. 记忆保留上下文；它不将 API 作为工具公开。
C. 代码解释器执行代码，而不是 API 到工具的暴露。
D. 可观察性提供跟踪和监控，而不是工具集成。
An analytics agent must run user-provided Python to perform calculations and data transformations safely, isolated from other workloads. Which AgentCore capability should the developer use?
分析代理必须运行用户提供的 Python 才能安全地执行计算和数据转换，并与其他工作负载隔离。开发人员应该使用哪种 AgentCore 功能？
- A
AgentCore Code Interpreter.
AgentCore 代码解释器。
- B
AgentCore Gateway.
AgentCore 网关。
- C
AgentCore Browser.
AgentCore 浏览器。
- D
AgentCore Memory.
代理核心内存。
解析 / Explanation
正确答案 / Answer: A
A. Code Interpreter gives the agent a secure, isolated sandbox to execute code for calculations and data processing.
B. Gateway exposes external tools; it is not a code execution sandbox.
C. Browser automates web interactions, not arbitrary code execution.
D. Memory stores context and does not execute code.
A. 代码解释器为代理提供了一个安全、隔离的沙箱来执行计算和数据处理的代码。
B. 网关暴露外部工具；它不是代码执行沙箱。
C. 浏览器自动执行网络交互，而不是任意代码执行。
D. 内存存储上下文并且不执行代码。
An agent needs to navigate websites and interact with web pages to complete tasks, in a managed and secure environment rather than self-hosted browser infrastructure. Which AgentCore capability fits?
代理需要在托管且安全的环境中而不是自托管的浏览器基础设施中导航网站并与网页交互来完成任务。哪种 AgentCore 功能适合？
- A
AgentCore Browser.
AgentCore 浏览器。
- B
AgentCore Code Interpreter.
AgentCore 代码解释器。
- C
AgentCore Memory.
代理核心内存。
- D
AgentCore Identity.
代理核心身份。
解析 / Explanation
正确答案 / Answer: A
A. The AgentCore Browser tool provides a managed, secure, cloud-based browser the agent can use to navigate and interact with web pages.
B. Code Interpreter runs code; it is not a managed browser.
C. Memory persists context, not web interaction.
D. Identity manages secure access to resources, not browser automation.
A. AgentCore 浏览器工具提供了一个托管、安全、基于云的浏览器，代理可以使用它来导航网页并与网页交互。
B. 代码解释器运行代码；它不是受管理的浏览器。
C. 内存保留的是上下文，而不是网络交互。
D. 身份管理的是对资源的安全访问，而不是浏览器自动化。
Agents must securely access third-party SaaS services on behalf of users, including handling tokens and consent, without the developer building a custom credential broker. Which AgentCore capability addresses this?
代理必须代表用户安全地访问第三方 SaaS 服务，包括处理令牌和同意，而无需开发人员构建自定义凭证代理。哪种 AgentCore 功能可以解决这个问题？
- A
AgentCore Identity.
代理核心身份。
- B
AgentCore Browser.
AgentCore 浏览器。
- C
AgentCore Memory.
代理核心内存。
- D
AgentCore Code Interpreter.
AgentCore 代码解释器。
解析 / Explanation
正确答案 / Answer: A
A. AgentCore Identity provides secure identity and access management for agents, including authenticating to and accessing AWS and third-party services on a user's behalf.
B. Browser automates web pages; it is not an identity and access broker.
C. Memory persists context, not credentials and consent flows.
D. Code Interpreter executes code and does not manage outbound identity.
A. AgentCore Identity 为代理提供安全的身份和访问管理，包括代表用户对 AWS 和第三方服务进行身份验证和访问。
B. 浏览器使网页自动化；它不是身份和访问代理。
C. 内存保留的是上下文，而不是凭证和同意流。
D. 代码解释器执行代码并且不管理出站身份。
A complex workflow needs one coordinating agent to delegate subtasks to several specialized agents (billing, shipping, returns) and combine their results. Which Amazon Bedrock approach supports this?
复杂的工作流程需要一名协调代理将子任务委托给多个专门代理（计费、运输、退货）并合并他们的结果。哪种 Amazon Bedrock 方法支持此功能？
- A
Use multi-agent collaboration with a supervisor agent coordinating specialized sub-agents.
使用多代理协作，由主管代理协调专门的子代理。
- B
Put all capabilities into a single prompt and hope the model separates them.
将所有功能放入一个提示中，并希望模型将它们分开。
- C
Run unrelated agents with no coordination.
在没有协调的情况下运行不相关的代理。
- D
Increase max tokens so one agent can do everything.
增加最大令牌数，以便一名代理可以完成所有工作。
解析 / Explanation
正确答案 / Answer: A
A. Bedrock multi-agent collaboration lets a supervisor agent route subtasks to specialized agents and aggregate their outputs.
B. One monolithic prompt does not provide structured delegation or specialization.
C. Without coordination there is no way to combine sub-agent results into one workflow.
D. Output length does not provide multi-agent delegation.
A. Bedrock 多智能体协作让主管智能体将子任务路由给专业智能体并聚合其输出。
B. 一个单一的提示不提供结构化的委派或专业化。
C. 如果没有协调，就无法将子代理结果合并到一个工作流程中。
D. 输出长度不提供多代理委托。
A developer wants an open-source, model-driven framework to build agents with tools and reasoning in a few lines of Python, and then deploy them to Amazon Bedrock AgentCore Runtime. Which option fits?
开发人员需要一个开源、模型驱动的框架来使用工具和几行 Python 推理来构建代理，然后将它们部署到 Amazon Bedrock AgentCore Runtime。哪个选项适合？
- A
The Strands Agents SDK.
Strands Agent SDK。
- B
Amazon QuickSight.
亚马逊 QuickSight。
- C
Amazon Comprehend.
亚马逊理解。
- D
AWS Glue.
AWS 胶水。
解析 / Explanation
正确答案 / Answer: A
A. Strands Agents is AWS's open-source, model-driven agent SDK for building tool-using agents that can be deployed to AgentCore Runtime.
B. QuickSight is a business-intelligence service, not an agent framework.
C. Comprehend is an NLP service, not an agent-building SDK.
D. Glue is an ETL service and is unrelated to building agents.
A. Strands Agents 是 AWS 的开源模型驱动代理 SDK，用于构建可部署到 AgentCore Runtime 的使用工具的代理。
B. QuickSight 是一项商业智能服务，而不是代理框架。
C. Comprehend 是一个 NLP 服务，而不是一个代理构建 SDK。
D. Glue 是一项 ETL 服务，与构建代理无关。
An operations team must debug why an agent chose a particular tool and trace each step, tool call, and latency in production. Which AgentCore capability provides this visibility?
运营团队必须调试代理选择特定工具的原因，并跟踪生产中的每个步骤、工具调用和延迟。哪种 AgentCore 功能提供了这种可见性？
- A
AgentCore Observability, which provides tracing and metrics for agent steps and tool calls (integrating with CloudWatch/OpenTelemetry).
AgentCore Observability，为代理步骤和工具调用提供跟踪和指标（与 CloudWatch/OpenTelemetry 集成）。
- B
AgentCore Memory.
代理核心内存。
- C
AgentCore Gateway.
AgentCore 网关。
- D
Raising the temperature.
提高温度。
解析 / Explanation
正确答案 / Answer: A
A. AgentCore Observability captures detailed traces of an agent's reasoning steps and tool invocations for debugging and monitoring.
B. Memory stores context; it is not a tracing and monitoring tool.
C. Gateway exposes tools; it does not provide step-level tracing.
D. Temperature is unrelated to observability.
A. AgentCore Observability 捕获代理推理步骤和工具调用的详细跟踪，以进行调试和监控。
B. 内存存储上下文；它不是跟踪和监控工具。
C. 网关公开工具；它不提供步骤级跟踪。
D. 温度与可观测性无关。
An agent can issue refunds, and the business requires a human to approve before any refund is executed. Which Bedrock agent feature enforces this in the action flow?
代理可以发放退款，并且企业需要人工批准才能执行任何退款。哪个基岩代理功能在操作流程中强制执行此操作？
- A
Enable user confirmation (require confirmation) on the sensitive action so the agent must obtain approval before executing it.
对敏感操作启用用户确认（需要确认），以便代理在执行之前必须获得批准。
- B
Rely on the system prompt to ask the model to be careful.
依靠系统提示要求模型小心。
- C
Set temperature to 0 so the agent never makes mistakes.
将温度设置为 0，这样代理就不会犯错误。
- D
Disable the action group entirely.
完全禁用操作组。
解析 / Explanation
正确答案 / Answer: A
A. Bedrock agents can require user confirmation for an action, pausing for explicit approval before the operation runs.
B. A prompt request does not enforce a hard approval gate before execution.
C. Temperature does not create an approval step.
D. Disabling the action removes the capability rather than gating it with approval.
A. Bedrock 代理可以要求用户确认操作，并在操作运行之前暂停以获取明确批准。
B. 提示请求在执行前不会强制执行硬批准门。
C. 温度不会产生批准步骤。
D. 禁用该操作会删除该功能，而不是通过批准对其进行控制。
An application needs synchronous, low-latency predictions for small payloads with steady, predictable traffic. Which Amazon SageMaker AI deployment option is the best fit?
应用程序需要对具有稳定、可预测流量的小负载进行同步、低延迟预测。哪种 Amazon SageMaker AI 部署选项最适合？
- A
A real-time inference endpoint.
实时推理端点。
- B
A batch transform job.
批量转换作业。
- C
An asynchronous inference endpoint.
异步推理端点。
- D
A SageMaker Ground Truth job.
SageMaker Ground Truth 作业。
解析 / Explanation
正确答案 / Answer: A
A. Real-time endpoints provide persistent, low-latency synchronous inference, ideal for steady traffic with small payloads.
B. Batch transform is for offline, asynchronous scoring of datasets, not low-latency synchronous requests.
C. Asynchronous endpoints suit long-running requests and large payloads, adding queueing latency unnecessary for small, fast requests.
D. Ground Truth is for data labeling, not model serving.
A. 实时端点提供持久、低延迟的同步推理，非常适合小负载的稳定流量。
B. 批量转换适用于数据集的离线、异步评分，而不是低延迟同步请求。
C. 异步端点适合长时间运行的请求和大的负载，增加了小而快速的请求不必要的排队延迟。
D. Ground Truth 用于数据标记，而不是模型服务。
A model serves intermittent traffic with idle periods, does not require a GPU, and the team can tolerate occasional cold starts. They want to avoid paying for idle capacity. Which option is MOST cost-effective?
模型可服务具有空闲期的间歇性流量，不需要 GPU，并且团队可以容忍偶尔的冷启动。他们希望避免为闲置产能付费。哪种选择最具成本效益？
- A
SageMaker Serverless Inference.
SageMaker 无服务器推理。
- B
A large always-on real-time endpoint.
大型始终在线的实时端点。
- C
An asynchronous endpoint sized for peak.
适合峰值大小的异步端点。
- D
A 24/7 multi-GPU instance.
24/7 多 GPU 实例。
解析 / Explanation
正确答案 / Answer: A
A. Serverless Inference automatically provisions and scales (including to zero) for intermittent CPU workloads, so you pay only when serving requests.
B. An always-on endpoint incurs cost during the long idle periods.
C. Provisioning for peak wastes capacity for intermittent traffic.
D. A continuously running GPU instance is the most expensive and is unnecessary for a CPU workload.
A. 无服务器推理会针对间歇性 CPU 工作负载自动配置和扩展（包括为零），因此您只需在处理请求时付费。
B. 始终在线的端点在长时间空闲期间会产生成本。
C. 为间歇性流量提供高峰浪费容量。
D. 连续运行的 GPU 实例是最昂贵的，对于 CPU 工作负载来说是不必要的。
A company must host hundreds of small, similar models (one per customer) cost-effectively behind as few endpoints as possible, loading each model on demand. Which SageMaker feature fits best?
公司必须在尽可能少的端点后面经济高效地托管数百个小型相似模型（每个客户一个），并按需加载每个模型。哪种 SageMaker 功能最适合？
- A
A multi-model endpoint (MME).
多模型端点 (MME)。
- B
One dedicated real-time endpoint per model.
每个模型一个专用的实时端点。
- C
A batch transform job per model.
每个模型一个批量转换作业。
- D
Serverless inference with one function per model and no sharing.
无服务器推理，每个模型一个函数且不共享。
解析 / Explanation
正确答案 / Answer: A
A. Multi-model endpoints host many models behind a single endpoint, loading them on demand and sharing resources to reduce cost.
B. Hundreds of separate endpoints multiply cost and management overhead.
C. Batch transform is for offline scoring, not on-demand hosting of many models.
D. This does not provide the on-demand shared hosting that multi-model endpoints are designed for.
A. 多模型端点在单个端点后面托管许多模型，按需加载它们并共享资源以降低成本。
B. 数百个独立端点会增加成本和管理开销。
C. 批量转换用于离线评分，而不是许多模型的按需托管。
D. 这不提供多模型端点设计用于的按需共享托管。
A team deploys several large language models to GPU instances and wants to pack multiple models and copies onto shared accelerators, scale each independently, and scale unused ones to zero to cut cost. Which capability supports this?
一个团队将多个大型语言模型部署到 GPU 实例，并希望将多个模型打包并复制到共享加速器上，独立扩展每个模型，并将未使用的模型扩展为零以降低成本。哪种能力支持这个？
- A
SageMaker inference components.
SageMaker 推理组件。
- B
A single-model real-time endpoint per model only.
仅每个模型一个单模型实时端点。
- C
Batch transform.
批量变换。
- D
Ground Truth.
地面真相。
解析 / Explanation
正确答案 / Answer: A
A. Inference components let you deploy multiple models/copies on shared instances, allocate resources per component, scale them independently, and scale to zero.
B. Single-model endpoints cannot pack multiple models onto shared accelerators with independent scaling.
C. Batch transform is offline scoring, not GPU packing of online models.
D. Ground Truth labels data and is unrelated to serving.
A. 推理组件允许您在共享实例上部署多个模型/副本，为每个组件分配资源，独立扩展它们，并扩展到零。
B. 单模型端点无法将多个模型打包到具有独立扩展的共享加速器上。
C. 批量转换是离线评分，而不是在线模型的 GPU 打包。
D. Ground Truth 标记数据并且与服务无关。
Traffic to a real-time endpoint varies widely between business hours and nights. The team wants the endpoint to add and remove instances automatically based on load. What should they configure?
实时端点的流量在工作时间和夜间之间差异很大。该团队希望端点能够根据负载自动添加和删除实例。他们应该配置什么？
- A
Automatic scaling (for example, target-tracking on invocations per instance) for the endpoint.
端点的自动扩展（例如，对每个实例的调用进行目标跟踪）。
- B
A fixed, manually chosen instance count sized for peak.
手动选择的固定实例计数，大小适合峰值。
- C
Switch to batch transform during the day.
白天切换到批量转换。
- D
Raise the model temperature during peak hours.
在高峰时段提高模型温度。
解析 / Explanation
正确答案 / Answer: A
A. SageMaker endpoint auto scaling adjusts instance count based on a target metric such as invocations per instance, matching capacity to demand.
B. A fixed peak size wastes money at night and cannot adapt automatically.
C. Batch transform is offline and cannot serve real-time daytime traffic.
D. Temperature does not affect endpoint capacity.
A. SageMaker 端点自动扩展根据目标指标（例如每个实例的调用次数）调整实例计数，将容量与需求相匹配。
B. 固定的峰值大小在晚上会浪费金钱，而且无法自动适应。
C. 批量转换是离线的，无法服务于白天的实时流量。
D. 温度不影响端点容量。
Before promoting a new model version, a team wants to send a copy of live production traffic to it and compare results without affecting users. Which SageMaker capability provides this?
在推广新模型版本之前，团队希望向其发送实时生产流量的副本并比较结果而不影响用户。哪种 SageMaker 功能可以提供此功能？
- A
Shadow testing (shadow variants).
影子测试（影子变体）。
- B
Batch transform on yesterday's data.
对昨天的数据进行批量转换。
- C
Deleting the old endpoint first.
首先删除旧端点。
- D
Increasing max tokens on the new version.
增加新版本的最大令牌数。
解析 / Explanation
正确答案 / Answer: A
A. Shadow testing mirrors live traffic to a shadow variant so you can validate a new version against real requests without impacting production responses.
B. Offline batch scoring does not exercise the new version against live production traffic.
C. Removing the current endpoint would disrupt users and provides no safe comparison.
D. Output length settings are unrelated to safely comparing versions on live traffic.
A. 影子测试将实时流量镜像到影子变体，以便您可以根据实际请求验证新版本，而不会影响生产响应。
B. 离线批量评分不会针对实时生产流量使用新版本。
C. 删除当前端点会扰乱用户并且无法提供安全比较。
D. 输出长度设置与安全比较实时流量的版本无关。
A developer wants to quickly deploy a popular open foundation model with minimal setup, including a tested container and example notebooks, to get an endpoint running fast. Which option is best?
开发人员希望以最少的设置快速部署流行的开放基础模型，包括经过测试的容器和示例笔记本，以使端点快速运行。哪个选项最好？
- A
Deploy the model from Amazon SageMaker JumpStart.
从 Amazon SageMaker JumpStart 部署模型。
- B
Build a custom container from scratch and write all serving code.
从头开始构建自定义容器并编写所有服务代码。
- C
Use batch transform to expose an endpoint.
使用批量转换来公开端点。
- D
Use Amazon Comprehend.
使用 Amazon Comprehend。
解析 / Explanation
正确答案 / Answer: A
A. JumpStart offers prebuilt, deployable foundation and task models with curated containers and example notebooks for fast, low-effort deployment.
B. Building everything yourself is the slowest, highest-effort path.
C. Batch transform does not create an online endpoint.
D. Comprehend is a managed NLP service, not a way to deploy an arbitrary foundation model.
A. JumpStart 提供预构建、可部署的基础和任务模型，以及精选的容器和示例笔记本，以实现快速、省力的部署。
B. 自己建造一切是最慢、最费力的道路。
C. 批量转换不会创建在线端点。
D. Comprehend 是一项托管 NLP 服务，而不是部署任意基础模型的方法。
A team must score a large, fixed dataset that already sits in Amazon S3 in a single offline run, with no need for a persistent endpoint. Which SageMaker option is MOST appropriate and cost-effective?
团队必须在一次离线运行中对已位于 Amazon S3 中的大型固定数据集进行评分，而无需持久端点。哪种 SageMaker 选项最合适且最具成本效益？
- A
A batch transform job.
批量转换作业。
- B
A real-time endpoint invoked in a loop.
循环调用的实时端点。
- C
A serverless endpoint with manual paging.
具有手动寻呼功能的无服务器端点。
- D
An asynchronous endpoint for each record.
每条记录的异步端点。
解析 / Explanation
正确答案 / Answer: A
A. Batch transform processes a whole dataset from S3 in one offline run and tears down compute afterward, with no persistent endpoint to pay for.
B. Looping requests over a persistent endpoint is more expensive and complex than a single batch job.
C. Serverless endpoints target intermittent online requests, not bulk offline scoring of a fixed dataset.
D. Per-record async calls add unnecessary overhead versus one batch transform over the dataset.
A. 批量转换在一次离线运行中处理来自 S3 的整个数据集，然后拆除计算，无需支付持久端点费用。
B. 通过持久端点循环请求比单个批处理作业更加昂贵和复杂。
C. 无服务器端点针对间歇性在线请求，而不是固定数据集的批量离线评分。
D. 与对数据集进行批量转换相比，每条记录的异步调用会增加不必要的开销。
When updating a production endpoint to a new model, a team wants a controlled rollout that shifts a small percentage of traffic first and automatically rolls back if errors spike. Which SageMaker feature provides this?
将生产端点更新为新模型时，团队希望进行受控的部署，首先转移一小部分流量，并在错误激增时自动回滚。哪个 SageMaker 功能可以提供此功能？
- A
Deployment guardrails with a canary (or linear) traffic-shifting strategy and automatic rollback.
具有金丝雀（或线性）流量转移策略和自动回滚的部署护栏。
- B
Deleting and recreating the endpoint in place.
删除并重新创建端点。
- C
Batch transform validation only.
仅批量转换验证。
- D
Raising the instance count during the switch.
在切换期间增加实例计数。
解析 / Explanation
正确答案 / Answer: A
A. SageMaker deployment guardrails shift traffic gradually (canary/linear), monitor alarms, and roll back automatically if the new version misbehaves.
B. An in-place replace gives no gradual shift or automatic rollback and risks downtime.
C. Offline batch validation does not control live traffic shifting or rollback.
D. More instances does not provide controlled traffic shifting or automatic rollback.
A. SageMaker 部署护栏会逐渐转移流量（金丝雀/线性）、监控警报，并在新版本出现问题时自动回滚。
B. 就地替换不会导致逐渐转移或自动回滚，并且存在停机风险。
C. 离线批量验证不控制实时流量转移或回滚。
D. 更多实例不提供受控流量转移或自动回滚。
Developers want inline, context-aware code suggestions and the ability to generate functions from natural-language comments directly inside their IDE to speed up routine coding. Which capability provides this?
开发人员需要内联、上下文感知的代码建议以及直接在 IDE 中从自然语言注释生成函数的能力，以加快日常编码速度。哪种功能可以提供此功能？
- A
Amazon Q Developer in-IDE code suggestions and generation.
Amazon Q Developer IDE 内代码建议和生成。
- B
Amazon QuickSight Q.
亚马逊 QuickSight Q。
- C
Amazon Comprehend.
亚马逊理解。
- D
Amazon Kendra.
亚马逊肯德拉。
解析 / Explanation
正确答案 / Answer: A
A. Amazon Q Developer integrates with the IDE to provide real-time inline completions and generate code from natural-language prompts.
B. QuickSight Q answers natural-language questions about BI data, not IDE code generation.
C. Comprehend is an NLP analysis service, not a coding assistant.
D. Kendra is enterprise search, not an IDE coding assistant.
A. Amazon Q Developer 与 IDE 集成，提供实时内联补全并根据自然语言提示生成代码。
B. QuickSight Q 回答有关 BI 数据的自然语言问题，而不是 IDE 代码生成。
C. Comprehend 是一种 NLP 分析服务，而不是编码助手。
D. Kendra 是企业搜索，而不是 IDE 编码助手。
A team wants an assistant that can take a natural-language feature request, plan the change, and implement it across multiple files in their project. Which Amazon Q Developer capability is designed for this?
团队需要一名助理来接受自然语言功能请求、规划变更并在项目中的多个文件中实施变更。哪种 Amazon Q Developer 功能是为此设计的？
- A
The Amazon Q Developer feature-development agent (/dev), which plans and applies multi-file changes.
Amazon Q Developer 功能开发代理 (/dev)，用于计划和应用多文件更改。
- B
Single-line autocomplete only.
仅单行自动完成。
- C
A CloudWatch dashboard.
CloudWatch 仪表板。
- D
Increasing the model temperature.
提高模型温度。
解析 / Explanation
正确答案 / Answer: A
A. The Q Developer agent for software development analyzes the codebase, proposes a plan, and generates coordinated changes across multiple files.
B. Basic autocomplete cannot plan and implement a multi-file feature.
C. Dashboards visualize metrics; they do not implement code features.
D. Temperature does not provide multi-file feature implementation.
A. 用于软件开发的 Q Developer 代理分析代码库、提出计划并跨多个文件生成协调的更改。
B. 基本自动完成无法规划和实现多文件功能。
C. 仪表板可视化指标；他们不实现代码功能。
D. 温度不提供多文件功能实现。
An enterprise must upgrade many applications from an old Java version to a current LTS release and wants to automate the repetitive code changes at scale. Which Amazon Q Developer capability helps MOST?
企业必须将许多应用程序从旧的 Java 版本升级到当前的 LTS 版本，并希望大规模地自动执行重复的代码更改。哪种 Amazon Q Developer 功能最有帮助？
- A
Amazon Q Developer code transformation (automated language/framework upgrades).
Amazon Q Developer 代码转换（自动语言/框架升级）。
- B
Manually editing each file by hand.
手动编辑每个文件。
- C
Amazon Macie.
亚马逊麦西.
- D
Amazon Bedrock Guardrails.
亚马逊基岩护栏。
解析 / Explanation
正确答案 / Answer: A
A. Q Developer's transformation capability automates large-scale upgrades, such as moving codebases to a newer Java version, reducing manual effort.
B. Manual edits across many applications are slow and error-prone, which automation avoids.
C. Macie classifies sensitive data and has nothing to do with code upgrades.
D. Guardrails govern model output content, not code transformations.
A. Q Developer 的转换功能可自动进行大规模升级，例如将代码库移动到较新的 Java 版本，从而减少手动工作。
B. 许多应用程序中的手动编辑速度缓慢且容易出错，而自动化可以避免这种情况。
C. Macie 对敏感数据进行分类，与代码升级无关。
D. 护栏管理模型输出内容，而不是代码转换。
A security-conscious team wants their assistant to scan code for vulnerabilities and suggest remediations as part of development. Which Amazon Q Developer capability provides this?
具有安全意识的团队希望他们的助手能够扫描代码中的漏洞并提出修复建议，作为开发的一部分。哪种 Amazon Q Developer 功能可以提供此功能？
- A
Amazon Q Developer security scanning, which detects vulnerabilities and suggests fixes.
Amazon Q Developer 安全扫描，可检测漏洞并提出修复建议。
- B
Amazon GuardDuty agent on the laptop.
笔记本电脑上的 Amazon GuardDuty 代理。
- C
Raising the temperature parameter.
提高温度参数。
- D
Amazon Athena queries.
亚马逊雅典娜查询。
解析 / Explanation
正确答案 / Answer: A
A. Q Developer can scan code for security issues and recommend remediations directly in the developer workflow.
B. GuardDuty monitors AWS accounts and workloads for threats, not in-IDE source code scanning for developers.
C. Temperature is unrelated to code security scanning.
D. Athena queries data in S3 and does not scan code for vulnerabilities.
A. Q Developer 可以扫描代码中的安全问题，并直接在开发人员工作流程中建议修复措施。
B. GuardDuty 监控 AWS 账户和工作负载是否存在威胁，而不是对开发人员进行 IDE 源代码扫描。
C. 温度与代码安全扫描无关。
D. Athena 查询 S3 中的数据，并且不扫描代码中的漏洞。
To improve coverage and velocity, a team wants to automatically generate unit tests for existing functions. Which capability supports this with the least manual effort?
为了提高覆盖率和速度，团队希望自动为现有功能生成单元测试。哪种功能可以用最少的手动操作来支持此操作？
- A
Amazon Q Developer unit test generation.
Amazon Q Developer 单元测试生成。
- B
Writing every test by hand with no tooling.
无需任何工具，手动编写每个测试。
- C
Amazon QuickSight.
亚马逊 QuickSight。
- D
Amazon Comprehend.
亚马逊理解。
解析 / Explanation
正确答案 / Answer: A
A. Q Developer can generate unit tests for existing code, increasing coverage with minimal manual effort.
B. Hand-writing all tests is the slowest option and is what the team wants to avoid.
C. QuickSight is for dashboards, not test generation.
D. Comprehend performs NLP analysis, not unit test generation.
A. Q Developer 可以为现有代码生成单元测试，以最少的手动工作增加覆盖范围。
B. 手写所有测试是最慢的选择，也是团队想要避免的。
C. QuickSight 适用于仪表板，而不是测试生成。
D. Comprehend 执行 NLP 分析，而不是生成单元测试。
A team must retain the full request and response content of every Amazon Bedrock model call for troubleshooting and audit. Which capability captures this with the LEAST custom code?
团队必须保留每个 Amazon Bedrock 模型调用的完整请求和响应内容，以进行故障排除和审核。哪种功能可以用最少的自定义代码捕获此内容？
- A
Enable Amazon Bedrock model invocation logging to Amazon S3 (and/or CloudWatch Logs).
启用 Amazon Bedrock 模型调用日志记录到 Amazon S3（和/或 CloudWatch Logs）。
- B
Add print statements in the application and hope they are captured.
在应用程序中添加打印语句并希望它们被捕获。
- C
Enable AWS CloudTrail only.
仅启用 AWS CloudTrail。
- D
Use Amazon Macie on the application logs.
在应用程序日志上使用 Amazon Macie。
解析 / Explanation
正确答案 / Answer: A
A. Model invocation logging is a built-in feature that records full prompts, responses, and metadata to S3 and/or CloudWatch with no custom instrumentation.
B. Ad hoc logging is incomplete and unreliable compared with built-in invocation logging.
C. CloudTrail records API activity metadata, not the full prompt and response content.
D. Macie classifies sensitive data; it does not capture model request/response content.
A. 模型调用日志记录是一项内置功能，无需自定义工具即可将完整的提示、响应和元数据记录到 S3 和/或 CloudWatch。
B. 与内置调用日志记录相比，临时日志记录不完整且不可靠。
C. CloudTrail 记录 API 活动元数据，而不是完整的提示和响应内容。
D. Macie 对敏感数据进行分类；它不捕获模型请求/响应内容。
A finance team needs to attribute Amazon Bedrock usage and cost to specific applications and teams that share the same models. Which approach is purpose-built for this?
财务团队需要将 Amazon Bedrock 使用情况和成本归因于共享相同模型的特定应用程序和团队。哪种方法是专门为此构建的？
- A
Use Amazon Bedrock application inference profiles with cost allocation tags to track usage and cost per application.
使用带有成本分配标签的 Amazon Bedrock 应用程序推理配置文件来跟踪每个应用程序的使用情况和成本。
- B
Guess each team's share from total spend.
猜测每个团队在总支出中所占的份额。
- C
Raise the temperature to reduce cost.
提高温度以降低成本。
- D
Disable logging to lower the bill.
禁用日志记录以降低费用。
解析 / Explanation
正确答案 / Answer: A
A. Application inference profiles let you route and tag invocations so usage and cost can be tracked and attributed per application or team.
B. Estimation without tagging is inaccurate and not auditable.
C. Temperature does not attribute or reduce cost.
D. Disabling logging removes visibility and does not attribute cost.
A. 应用程序推理配置文件允许您路由和标记调用，以便可以跟踪每个应用程序或团队的使用情况和成本并归因。
B. 没有标记的估计是不准确且不可审计的。
C. 温度不会归因或降低成本。
D. 禁用日志记录会消除可见性并且不会产生成本。
An application intermittently receives throttling errors from Amazon Bedrock, and operators want to be alerted automatically when throttling rises. What should they configure?
应用程序间歇性地收到来自 Amazon Bedrock 的限制错误，操作员希望在限制上升时自动收到警报。他们应该配置什么？
- A
A CloudWatch alarm on the Bedrock throttling metric for the model.
针对模型的 Bedrock 限制指标的 CloudWatch 警报。
- B
Nothing; throttling cannot be monitored.
没有什么;无法监控节流。
- C
Increase max tokens to avoid throttling.
增加最大令牌以避免限制。
- D
Lower the temperature to reduce throttling.
降低温度以减少节流。
解析 / Explanation
正确答案 / Answer: A
A. Bedrock publishes CloudWatch metrics (including throttling), so an alarm on the throttle metric notifies operators when throttling increases.
B. Throttling is observable through CloudWatch metrics and can be alarmed on.
C. Output length does not address request-rate throttling or provide alerting.
D. Temperature has no effect on throttling or its monitoring.
A. Bedrock 发布 CloudWatch 指标（包括限制），因此当限制增加时，有关限制指标的警报会通知操作员。
B. 可以通过 CloudWatch 指标观察限制并发出警报。
C. 输出长度不解决请求速率限制或提供警报。
D. 温度对节流或其监控没有影响。
A generative AI request flows through API Gateway, several Lambda functions, a knowledge base, and Bedrock, and the team needs to find which component causes high end-to-end latency. Which service provides distributed tracing to localize the bottleneck?
生成式 AI 请求流经 API Gateway、多个 Lambda 函数、知识库和 Bedrock，团队需要找到哪个组件导致较高的端到端延迟。哪个服务提供分布式跟踪来定位瓶颈？
- A
AWS X-Ray.
AWS X 射线。
- B
Amazon Macie.
亚马逊麦西.
- C
Amazon Comprehend.
亚马逊理解。
- D
AWS Glue.
AWS 胶水。
解析 / Explanation
正确答案 / Answer: A
A. X-Ray traces requests across distributed components, producing a service map and segment timings that pinpoint the slow part of the chain.
B. Macie classifies sensitive data and does not trace request latency.
C. Comprehend analyzes text and is not a tracing tool.
D. Glue is for ETL, not distributed tracing.
A. X-Ray 跨分布式组件跟踪请求，生成服务图和分段计时，以精确定位链中的慢速部分。
B. Macie 对敏感数据进行分类，并且不跟踪请求延迟。
C. Comprehend 分析文本，而不是跟踪工具。
D. Glue 用于 ETL，而不是分布式跟踪。
Invocation logs are being written to CloudWatch Logs, and an engineer needs to interactively query them to find error patterns and aggregate failures. Which tool should they use?
调用日志被写入 CloudWatch Logs，工程师需要以交互方式查询它们以查找错误模式和聚合故障。他们应该使用哪种工具？
- A
Amazon CloudWatch Logs Insights.
Amazon CloudWatch 日志见解。
- B
Amazon QuickSight Q on raw log files.
原始日志文件上的 Amazon QuickSight Q。
- C
Amazon Rekognition.
亚马逊重新识别。
- D
Increasing the log retention period only.
仅增加日志保留期限。
解析 / Explanation
正确答案 / Answer: A
A. Logs Insights provides an interactive query language to search, filter, and aggregate log data such as invocation logs to surface error patterns.
B. QuickSight Q answers questions over prepared BI datasets, not ad hoc CloudWatch log querying.
C. Rekognition analyzes images and video, not logs.
D. Longer retention stores logs but does not provide query and aggregation capabilities.
A. Logs Insights 提供交互式查询语言来搜索、过滤和聚合日志数据（例如调用日志）以显示错误模式。
B. QuickSight Q 回答有关准备好的 BI 数据集的问题，而不是临时 CloudWatch 日志查询。
C. Rekognition 分析图像和视频，而不是日志。
D. 较长的保留期会存储日志，但不提供查询和聚合功能。
A model deployed on a SageMaker real-time endpoint must be monitored for data drift and degradation in input/prediction quality over time, with automated alerts. Which capability is designed for this?
必须通过自动警报来监控部署在 SageMaker 实时端点上的模型的数据漂移和输入/预测质量随时间的下降。哪种功能是为此设计的？
- A
Amazon SageMaker Model Monitor.
Amazon SageMaker 模型监视器。
- B
AWS CloudTrail.
AWS CloudTrail。
- C
Amazon Bedrock Guardrails.
亚马逊基岩护栏。
- D
Amazon Athena.
亚马逊雅典娜。
解析 / Explanation
正确答案 / Answer: A
A. Model Monitor continuously checks deployed endpoints for data quality, drift, and other issues against a baseline and can raise alerts.
B. CloudTrail records API calls, not model data drift.
C. Guardrails filter content; they do not monitor endpoint data drift.
D. Athena is a query service and does not monitor model drift.
A. 模型监视器根据基线持续检查已部署端点的数据质量、偏差和其他问题，并可以发出警报。
B. CloudTrail 记录 API 调用，而不是模型数据漂移。
C. 护栏过滤内容；他们不监控端点数据漂移。
D. Athena 是一项查询服务，不监控模型漂移。
After launch, a team wants to continuously assure the quality of a generative feature in production, catching regressions as prompts and models change. Which practice is MOST effective?
发布后，团队希望不断确保生产中生成功能的质量，并随着提示和模型的变化捕获回归。哪种做法最有效？
- A
Log responses and run periodic automated evaluations on sampled production traffic, alerting on quality regressions.
记录响应并定期对采样的生产流量进行自动评估，并就质量回归发出警报。
- B
Only watch CPU and memory metrics.
仅查看 CPU 和内存指标。
- C
Wait for users to report problems.
等待用户报告问题。
- D
Run a single pre-launch test and never re-check.
运行一次启动前测试，无需重新检查。
解析 / Explanation
正确答案 / Answer: A
A. Combining response logging with scheduled automated evaluation on samples provides ongoing, data-driven quality monitoring and early regression detection.
B. Infrastructure metrics do not reveal degradations in generation quality.
C. Relying on complaints is reactive and lets defects reach users.
D. A one-time test misses regressions introduced by later changes.
A. 将响应记录与样本的计划自动评估相结合，可以提供持续的、数据驱动的质量监控和早期回归检测。
B. 基础设施指标并未揭示发电质量的下降。
C. 依赖投诉是被动的，会让缺陷传达给用户。
D. 一次性测试会错过后续更改引入的回归。
A mobile app must let many end users call Amazon Bedrock without embedding AWS credentials in the app. The team wants each user to receive scoped, temporary AWS credentials after signing in. Which approach fits best?
移动应用程序必须允许许多最终用户调用 Amazon Bedrock，而无需在应用程序中嵌入 AWS 凭证。该团队希望每个用户在登录后都能收到有范围的临时 AWS 凭证。哪种方法最适合？
- A
Use Amazon Cognito identity pools to exchange a sign-in for temporary, scoped AWS credentials.
使用 Amazon Cognito 身份池将登录信息交换为临时的、限定范围的 AWS 凭证。
- B
Embed a shared IAM access key in the mobile app.
在移动应用程序中嵌入共享 IAM 访问密钥。
- C
Give every user the root account credentials.
为每个用户提供 root 帐户凭据。
- D
Store one IAM user's secret in the app config for everyone.
将一个 IAM 用户的秘密存储在每个人的应用程序配置中。
解析 / Explanation
正确答案 / Answer: A
A. Cognito identity pools federate authenticated users to temporary AWS credentials tied to an IAM role, avoiding embedded long-lived keys.
B. Embedding long-lived keys in a distributed app is a serious security risk.
C. Sharing root credentials is never acceptable and grossly violates least privilege.
D. A shared long-lived secret cannot be scoped per user and is easily compromised.
A. Cognito 身份池将经过身份验证的用户联合到与 IAM 角色关联的临时 AWS 凭证，从而避免嵌入长期密钥。
B. 在分布式应用程序中嵌入长期密钥是一个严重的安全风险。
C. 共享 root 凭据是绝对不可接受的，并且严重违反了最低权限。
D. 共享的长期秘密无法按用户确定范围，并且很容易被泄露。
A regulated company must call Amazon Bedrock from workloads in its VPC without traversing the public internet. Which solution provides private connectivity?
受监管的公司必须从其 VPC 中的工作负载中调用 Amazon Bedrock，而无需穿越公共互联网。哪种解决方案提供专用连接？
- A
Create VPC interface endpoints (AWS PrivateLink) for Amazon Bedrock.
为 Amazon Bedrock 创建 VPC 接口终端节点 (AWS PrivateLink)。
- B
Route all Bedrock calls through a public NAT gateway only.
仅通过公共 NAT 网关路由所有 Bedrock 呼叫。
- C
Disable encryption to simplify networking.
禁用加密以简化网络。
- D
Use the model temperature to control routing.
使用模型温度来控制布线。
解析 / Explanation
正确答案 / Answer: A
A. Interface VPC endpoints via PrivateLink keep traffic to Bedrock on the AWS network, avoiding the public internet.
B. A NAT gateway still sends traffic over the public internet, not a private path.
C. Disabling encryption weakens security and does not provide private connectivity.
D. Temperature is unrelated to network connectivity.
A. 通过 PrivateLink 的接口 VPC 终端节点将流量保留到 AWS 网络上的 Bedrock，从而避开公共互联网。
B. NAT 网关仍然通过公共互联网而不是专用路径发送流量。
C. 禁用加密会削弱安全性并且无法提供私人连接。
D. 温度与网络连接无关。
A company must encrypt its Bedrock custom model artifacts and knowledge base data at rest using keys it controls and can audit and rotate. Which approach should it use?
公司必须使用其控制且可以审核和轮换的密钥对其静态的 Bedrock 自定义模型工件和知识库数据进行加密。应该使用哪种方法？
- A
Encrypt with an AWS KMS customer managed key (CMK).
使用 AWS KMS 客户管理密钥 (CMK) 进行加密。
- B
Rely on no encryption to reduce latency.
不依靠加密来减少延迟。
- C
Store the data in a public S3 bucket.
将数据存储在公共 S3 存储桶中。
- D
Hard-code an encryption passphrase in the app.
在应用程序中硬编码加密密码。
解析 / Explanation
正确答案 / Answer: A
A. A customer managed KMS key gives the company control over key policies, rotation, and CloudTrail auditing of key usage for data at rest.
B. Operating without encryption fails the data-protection requirement.
C. Public exposure is the opposite of the security requirement.
D. A hard-coded passphrase is insecure and is not a managed, auditable key.
A. 客户管理的 KMS 密钥使公司能够控制密钥策略、轮换以及静态数据密钥使用情况的 CloudTrail 审核。
B. 不加密的操作不符合数据保护要求。
C. 公开暴露与安全要求相反。
D. 硬编码的密码是不安全的，并且不是托管的、可审计的密钥。
An application in Account A must call Amazon Bedrock resources governed by Account B, using temporary credentials and no long-lived keys. Which mechanism is appropriate?
账户 A 中的应用程序必须使用临时凭证且无长期密钥来调用由账户 B 管理的 Amazon Bedrock 资源。哪种机制合适？
- A
Create an IAM role in Account B and have Account A assume it with AWS STS AssumeRole.
在账户 B 中创建 IAM 角色，并让账户 A 使用 AWS STS AssumeRole 代入该角色。
- B
Email an IAM access key from Account B to Account A.
将 IAM 访问密钥从账户 B 通过电子邮件发送到账户 A。
- C
Make all Bedrock resources public.
公开所有基岩资源。
- D
Use the same root user for both accounts.
两个帐户使用相同的根用户。
解析 / Explanation
正确答案 / Answer: A
A. Cross-account IAM roles assumed via STS provide temporary credentials and scoped permissions without sharing long-lived keys.
B. Sharing long-lived access keys is insecure and unmanageable.
C. Public access removes the access control the requirement depends on.
D. Sharing root credentials across accounts is a severe security violation.
A. 通过 STS 承担的跨账户 IAM 角色提供临时凭证和范围权限，而无需共享长期密钥。
B. 共享长期访问密钥是不安全且难以管理的。
C. 公共访问消除了需求所依赖的访问控制。
D. 跨帐户共享根凭据是一种严重的安全违规行为。