AWS AIP-C01 开源练习 · 基础 · 第 03 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。
A team has small, self-contained records where each document is already a single coherent unit. Which Knowledge Base chunking option may be appropriate?
团队拥有小型、独立的记录,其中每个文档已经是一个连贯的单元。哪种知识库分块选项可能合适?
- A
No chunking — treat each document as one chunk
无分块——将每个文档视为一个块
- B
Hierarchical chunking with many levels
具有多个级别的分层分块
- C
Semantic chunking with tiny windows
使用小窗口进行语义分块
- D
Fixed-size chunking at 10 tokens
固定大小分块为 10 个标记
解析 / Explanation
正确答案 / Answer: A
A. When documents are short and self-contained, the "no chunking" option indexes each as a single chunk.
B. Hierarchy adds needless complexity for already-atomic records.
C. Over-splitting atomic records can fragment meaning.
D. Extremely small fixed chunks destroy context for whole-record content.
A. 当文档较短且独立时,“无分块”选项将每个文档作为单个块进行索引。
B. 层次结构给已经原子的记录增加了不必要的复杂性。
C. 过度分割原子记录可能会导致意义支离破碎。
D. 极小的固定块会破坏整个记录内容的上下文。
A team needs full control over how content is split (e.g., by their own delimiters or a Lambda transformation) in a Bedrock Knowledge Base. Which option supports this?
团队需要完全控制基岩知识库中内容的分割方式(例如,通过他们自己的分隔符或 Lambda 转换)。哪个选项支持这个?
- A
Custom chunking (e.g., via a Lambda transformation)
自定义分块(例如,通过 Lambda 转换)
- B
Fixed-size chunking only
仅固定大小分块
- C
No chunking
无分块
- D
Sharding
分片
解析 / Explanation
正确答案 / Answer: A
A. Custom chunking lets you define your own splitting logic, including a Lambda-based transformation.
B. Fixed-size cannot express custom delimiter logic.
C. No chunking is the opposite of fine-grained custom control.
D. Sharding is a storage-layer concept, not content splitting.
A. 自定义分块允许您定义自己的拆分逻辑,包括基于 Lambda 的转换。
B. 固定大小无法表达自定义分隔符逻辑。
C. 无分块与细粒度自定义控制相反。
D. 分片是一个存储层的概念,而不是内容分割。
Amazon Titan Text Embeddings supports configurable output dimensions. Which set of dimensions is supported?
Amazon Titan 文本嵌入支持可配置的输出尺寸。支持哪组维度?
- A
256, 512, or 1024 dimensions
256、512 或 1024 尺寸
- B
16, 32, or 64 dimensions
16、32 或 64 尺寸
- C
Exactly 1 dimension
恰好为 1 维
- D
Unlimited arbitrary dimensions
无限任意尺寸
解析 / Explanation
正确答案 / Answer: A
A. Titan embeddings can output 256, 512, or 1024 dimensions, trading storage/speed against fidelity.
B. These are not the supported Titan embedding sizes.
C. Embeddings are high-dimensional vectors, not scalars.
D. Titan supports specific sizes (256/512/1024), not arbitrary values.
A. Titan 嵌入可以输出 256、512 或 1024 维,以存储/速度和保真度为代价。
B. 这些不是支持的 Titan 嵌入大小。
C. 嵌入是高维向量,而不是标量。
D. Titan 支持特定大小 (256/512/1024),而不是任意值。
What is a practical trade-off of choosing a smaller embedding dimension (e.g., 256 vs 1024) with Titan?
使用 Titan 选择较小的嵌入尺寸(例如 256 与 1024)的实际权衡是什么?
- A
Smaller dimensions reduce storage and can speed search, potentially at some loss of representational fidelity
较小的尺寸可以减少存储并可以加快搜索速度,但可能会损失一定的表征保真度
- B
Smaller dimensions always improve accuracy
较小的尺寸总是会提高精度
- C
Dimension size has no effect on storage
尺寸大小对存储没有影响
- D
Smaller dimensions require Provisioned Throughput
较小的尺寸需要预置吞吐量
解析 / Explanation
正确答案 / Answer: A
A. Lower-dimensional vectors cost less to store/search but may capture slightly less nuance.
B. Fewer dimensions can reduce fidelity; it is a trade-off, not a pure win.
C. Dimension directly affects vector storage size.
D. Dimension choice is unrelated to capacity reservation.
A. 较低维度的向量存储/搜索的成本较低,但捕获的细微差别可能会稍微少一些。
B. 较少的维度会降低保真度;这是一种权衡,而不是纯粹的胜利。
C. 维度直接影响向量存储大小。
D. 维度选择与容量预留无关。
A search experience must match both exact keywords (like product codes) and semantic meaning. Which retrieval approach combines both?
搜索体验必须与确切的关键字(如产品代码)和语义相匹配。哪种检索方法结合了两者?
- A
Hybrid search — combining keyword (BM25) and semantic vector search
混合搜索——结合关键词(BM25)和语义向量搜索
- B
Pure keyword search only
仅纯关键字搜索
- C
Pure vector search only
仅纯矢量搜索
- D
Random retrieval
随机检索
解析 / Explanation
正确答案 / Answer: A
A. Hybrid search blends lexical BM25 matching with vector semantic matching for better coverage.
B. Keyword-only misses semantic matches and paraphrases.
C. Vector-only can miss exact tokens like codes/IDs.
D. Randomness is not a retrieval strategy.
A. 混合搜索将词汇 BM25 匹配与向量语义匹配相结合,以获得更好的覆盖范围。
B. 仅关键字会错过语义匹配和释义。
C. 仅向量可能会丢失代码/ID 等精确标记。
D. 随机性不是一种检索策略。
In hybrid search, what does the "BM25" component contribute?
在混合搜索中,“BM25”组件有什么贡献?
- A
Lexical/keyword relevance scoring based on term frequency
基于词频的词汇/关键词相关性评分
- B
Neural embedding generation
神经嵌入生成
- C
Vector dimensionality reduction
向量降维
- D
Document chunking
文档分块
解析 / Explanation
正确答案 / Answer: A
A. BM25 is a classic keyword ranking function; it provides the lexical half of hybrid search.
B. Embeddings power the vector half, not BM25.
C. BM25 is a ranking function, not a dimensionality technique.
D. BM25 ranks, it does not chunk.
A. BM25是经典的关键词排名功能;它提供了混合搜索的词汇部分。
B. 嵌入为向量的一半提供动力,而不是 BM25。
C. BM25 是一种排序函数,而不是一种维度技术。
D. BM25排名,它不分块。
After initial retrieval, the most relevant passages sometimes rank below less relevant ones. The team wants a managed way to reorder results by query relevance with minimal custom code. Which Bedrock capability fits?
初次检索后,最相关的段落有时会排在不太相关的段落之后。该团队希望采用一种托管方式,通过查询相关性使用最少的自定义代码对结果进行重新排序。哪种基岩功能适合?
- A
The Amazon Bedrock Rerank API with a reranker model
具有重新排序模型的 Amazon Bedrock Rerank API
- B
Increasing model temperature
提高模型温度
- C
Sharding the vector index
对向量索引进行分片
- D
Switching to no chunking
切换到无分块
解析 / Explanation
正确答案 / Answer: A
A. The Rerank API uses a cross-encoder reranker to reorder retrieved passages by relevance to the query.
B. Temperature affects generation randomness, not retrieval ordering.
C. Sharding affects capacity/latency, not relevance ranking.
D. Chunking strategy does not reorder results by relevance.
A. Rerank API 使用跨编码器重新排序器根据与查询的相关性对检索到的段落进行重新排序。
B. 温度影响生成随机性,而不影响检索顺序。
C. 分片影响容量/延迟,而不是相关性排名。
D. 分块策略不会按相关性对结果重新排序。
Which TWO reranker models are available through the Amazon Bedrock Rerank API?
通过 Amazon Bedrock Rerank API 可以使用哪两种重新排序模型?
- A
Amazon Rerank 1.0
亚马逊重新排名 1.0
- B
Cohere Rerank 3.5
Cohere 重新排名 3.5
- C
Titan Image Generator
泰坦图像生成器
- D
Stable Diffusion
稳定扩散
- E
Claude Rerank 2
克劳德·瑞兰克 2
解析 / Explanation
正确答案 / Answer: A,B
A. Amazon Rerank 1.0 is offered through the Rerank API.
B. Cohere Rerank 3.5 is available via the Bedrock Rerank API.
C. That is an image model, not a reranker.
D. Stable Diffusion is an image model, not a reranker.
E. No such Bedrock reranker model; rerankers are Amazon Rerank and Cohere Rerank.
A. Amazon Rerank 1.0 通过 Rerank API 提供。
B. Cohere Rerank 3.5 可通过 Bedrock Rerank API 获取。
C. 这是一个图像模型,而不是重新排序器。
D. 稳定扩散是一个图像模型,而不是重新排序器。
E. 没有这样的基岩重排序模型;重新排名工具包括 Amazon Rerank 和 Cohere Rerank。
How can reranking be applied to a Bedrock Knowledge Base query?
如何将重新排名应用于基岩知识库查询?
- A
Enable the reranker as a setting in the Retrieve / RetrieveAndGenerate request
启用重新排名器作为 Retrieve / RetrieveAndGenerate 请求中的设置
- B
It can only be used by training a custom model
它只能通过训练自定义模型来使用
- C
Reranking requires Provisioned Throughput on the embeddings model
重新排名需要嵌入模型的预配置吞吐量
- D
Reranking is only available outside AWS
重新排名仅在 AWS 外部可用
解析 / Explanation
正确答案 / Answer: A
A. KB users turn on reranking through the Retrieve and RetrieveAndGenerate APIs.
B. Reranking is a managed feature, no custom training needed.
C. Reranking is invoked via the API; it does not mandate provisioned embeddings capacity.
D. It is available in Amazon Bedrock.
A. KB 用户通过 Retrieve 和 RetrieveAndGenerate API 启用重新排名。
B. 重新排名是一项托管功能,无需自定义培训。
C. 重新排名是通过 API 调用的;它不强制规定嵌入能力。
D. 它可在亚马逊基岩中使用。
Why does a reranker (cross-encoder) often improve RAG relevance compared to vector similarity alone?
为什么与单独的向量相似度相比,重排序器(交叉编码器)通常可以提高 RAG 相关性?
- A
It jointly examines the query and each candidate passage to score true relevance, refining the first-stage results
它联合检查查询和每个候选段落以评分真实的相关性,从而完善第一阶段的结果
- B
It increases the embedding dimension automatically
自动增加嵌入维度
- C
It removes the need for any retrieval step
它消除了任何检索步骤的需要
- D
It guarantees deterministic generation
它保证了确定性生成
解析 / Explanation
正确答案 / Answer: A
A. Cross-encoders consider query+passage together, capturing relevance signals a single-vector similarity may miss.
B. Reranking does not change embedding dimensions.
C. Reranking refines retrieved candidates; retrieval still happens first.
D. Reranking orders passages; it does not control generation determinism.
A. 交叉编码器将查询+段落一起考虑,捕获单向量相似性可能错过的相关信号。
B. 重新排序不会改变嵌入维度。
C. 重新排序细化检索到的候选者;检索仍然首先发生。
D. 对段落重新排序;它不控制世代决定论。
A RAG app starts returning "no relevant information found" after a deploy. The FM invokes fine, OpenSearch is healthy, latency is normal, and there are no errors. What is the most likely cause?
部署后,RAG 应用程序开始返回“未找到相关信息”。 FM 调用正常,OpenSearch 运行正常,延迟正常,并且没有错误。最可能的原因是什么?
- A
The query embeddings are now generated by a different embedding model/version than the one used to index the documents
查询嵌入现在由与用于索引文档的嵌入模型/版本不同的嵌入模型/版本生成
- B
The temperature parameter was increased
温度参数增加
- C
The IAM role lost bedrock:InvokeModel
IAM 角色失去了基石:InvokeModel
- D
The vector index was sharded
向量索引被分片
解析 / Explanation
正确答案 / Answer: A
A. Embedding drift creates a vector-space mismatch between queries and stored vectors, silently breaking retrieval.
B. Temperature affects generation, not retrieval matching.
C. The FM invokes successfully, so the permission is present.
D. Sharding affects capacity/latency, not vector-space alignment.
A. 嵌入漂移会在查询和存储的向量之间造成向量空间不匹配,从而默默地破坏检索。
B. 温度影响生成,而不影响检索匹配。
C. FM调用成功,权限已存在。
D. 分片影响容量/延迟,而不是向量空间对齐。
Which rule must hold for vector retrieval to work correctly?
向量检索必须遵守哪条规则才能正确工作?
- A
Query embeddings must be produced by the same embedding model used to index the documents
查询嵌入必须由用于索引文档的相同嵌入模型生成
- B
Queries must always use a larger embedding model than indexing
查询必须始终使用比索引更大的嵌入模型
- C
Any embedding model can be mixed freely
任何嵌入模型都可以自由混合
- D
Embeddings must be regenerated for every query at index time
必须在索引时为每个查询重新生成嵌入
解析 / Explanation
正确答案 / Answer: A
A. Matching the embedding model/version keeps queries and documents in the same vector space.
B. They must match, not differ by size.
C. Mixing models causes a vector-space mismatch and poor retrieval.
D. Document embeddings are created at ingestion; queries are embedded at query time with the same model.
A. 匹配嵌入模型/版本可将查询和文档保留在同一向量空间中。
B. 它们必须匹配,而不是大小不同。
C. 混合模型会导致向量空间不匹配和检索效果不佳。
D. 文档嵌入是在摄取时创建的;查询在查询时嵌入相同的模型。
A team is deciding chunk size. What is the typical trade-off between very large and very small chunks?
团队正在决定块大小。非常大的块和非常小的块之间的典型权衡是什么?
- A
Large chunks preserve context but can dilute relevance and exceed limits; small chunks are precise but may lose surrounding context
大块内容可以保留上下文,但会削弱相关性并超出限制;小块是精确的,但可能会丢失周围的上下文
- B
Chunk size has no effect on retrieval quality
块大小对检索质量没有影响
- C
Smaller chunks always retrieve better in every case
在任何情况下,较小的块总是能更好地检索
- D
Larger chunks always reduce cost
更大的块总是会降低成本
解析 / Explanation
正确答案 / Answer: A
A. Chunk size trades context retention against retrieval precision and token limits.
B. Chunk size materially affects relevance and context.
C. Too-small chunks can lose needed context.
D. Larger chunks can increase tokens per retrieval and cost.
A. 块大小将上下文保留与检索精度和令牌限制进行交换。
B. 块大小会对相关性和上下文产生重大影响。
C. 太小的块可能会丢失所需的上下文。
D. 较大的块可以增加每次检索的令牌和成本。
A team wants chunk boundaries that follow the meaning of the text (paragraph/topic shifts) rather than arbitrary token counts. Which strategy should they pick?
团队希望块边界遵循文本的含义(段落/主题转换),而不是任意的标记计数。他们应该选择哪种策略?
- A
Semantic chunking
语义分块
- B
Fixed-size chunking
固定大小分块
- C
No chunking
无分块
- D
Sharding
分片
解析 / Explanation
正确答案 / Answer: A
A. Semantic chunking places boundaries at meaning shifts to keep coherent ideas intact.
B. Fixed-size splits by counts regardless of meaning.
C. No chunking keeps the whole document as one unit.
D. Sharding is storage distribution, not semantic splitting.
A. 语义分块在意义转换处设置了界限,以保持连贯的想法完整。
B. 无论含义如何,均按计数固定大小分割。
C. 没有分块将整个文档视为一个单元。
D. 分片是存储分配,而不是语义分割。
A developer is exposing tools to an LLM-based application using the Model Context Protocol (MCP). For a lightweight, event-driven MCP server, which AWS compute pattern is recommended?
开发人员使用模型上下文协议 (MCP) 将工具公开给基于 LLM 的应用程序。对于轻量级、事件驱动的 MCP 服务器,建议使用哪种 AWS 计算模式?
- A
Run the MCP server on AWS Lambda
在 AWS Lambda 上运行 MCP 服务器
- B
Run it on a dedicated mainframe
在专用主机上运行
- C
Run it only on a developer laptop in production
仅在生产中的开发人员笔记本电脑上运行它
- D
Store it as an S3 object
将其存储为 S3 对象
解析 / Explanation
正确答案 / Answer: A
A. A common AWS pattern is hosting a lightweight MCP server on Lambda for serverless, event-driven tool exposure.
B. Mainframes are not an AWS MCP hosting pattern.
C. Local laptops are not a production hosting pattern.
D. An MCP server is a running service, not a static object.
A. 一种常见的 AWS 模式是在 Lambda 上托管轻量级 MCP 服务器,以实现无服务器、事件驱动的工具公开。
B. 大型机不是 AWS MCP 托管模式。
C. 本地笔记本电脑不是生产托管模式。
D. MCP 服务器是一个正在运行的服务,而不是一个静态对象。
A complex, long-running MCP server needs more control over runtime, networking, and resources than a function provides. Which AWS pattern fits best?
复杂、长时间运行的 MCP 服务器需要比函数提供的更多的运行时、网络和资源控制。哪种 AWS 模式最适合?
- A
Run the MCP server on Amazon ECS (containers)
在 Amazon ECS(容器)上运行 MCP 服务器
- B
Convert it to a CloudFront function
将其转换为 CloudFront 函数
- C
Host it as a static website on S3
将其作为静态网站托管在 S3 上
- D
Use Amazon SES
使用亚马逊SES
解析 / Explanation
正确答案 / Answer: A
A. For complex/long-running MCP servers, containers on ECS give more control than Lambda.
B. CloudFront functions are tiny edge scripts, unsuitable for a complex MCP server.
C. Static hosting cannot run a stateful MCP server.
D. SES is email, not compute for an MCP server.
A. 对于复杂/长时间运行的 MCP 服务器,ECS 上的容器比 Lambda 提供更多控制。
B. CloudFront 函数是微小的边缘脚本,不适合复杂的 MCP 服务器。
C. 静态托管无法运行有状态的 MCP 服务器。
D. SES 是电子邮件,而不是 MCP 服务器的计算。
What does the Model Context Protocol (MCP) standardize?
模型上下文协议 (MCP) 标准化了什么?
- A
How AI applications/agents connect to external tools and data sources through a common protocol
AI应用程序/代理如何通过通用协议连接到外部工具和数据源
- B
How vectors are quantized for storage
如何量化向量以进行存储
- C
How IAM roles are rotated
IAM 角色如何轮换
- D
How S3 buckets replicate
S3 存储桶如何复制
解析 / Explanation
正确答案 / Answer: A
A. MCP is an open protocol for connecting models/agents to tools and context sources uniformly.
B. That is a vector-storage concern, not MCP.
C. Credential rotation is unrelated to MCP.
D. Bucket replication is unrelated to MCP.
A. MCP 是一种开放协议,用于将模型/代理统一连接到工具和上下文源。
B. 这是矢量存储问题,而不是 MCP。
C. 凭证轮换与 MCP 无关。
D. 桶复制与MCP无关。
Which TWO statements about combining hybrid search and reranking in a RAG pipeline are correct?
关于在 RAG 管道中组合混合搜索和重新排名的哪两个陈述是正确的?
- A
Hybrid search broadens recall by mixing keyword and semantic matches
混合搜索通过混合关键字和语义匹配来扩大召回范围
- B
Reranking improves precision by reordering the retrieved candidates by relevance
重新排序通过按相关性对检索到的候选者重新排序来提高精度
- C
Reranking replaces the need to retrieve any candidates
重新排名取代了检索任何候选者的需要
- D
Hybrid search only works without embeddings
混合搜索仅适用于没有嵌入的情况
- E
Both require fine-tuning the base model
两者都需要对基础模型进行微调
解析 / Explanation
正确答案 / Answer: A,B
A. Hybrid search improves the set of candidate passages retrieved.
B. A reranker refines ordering so the best passages surface first.
C. Reranking operates on already-retrieved candidates.
D. Hybrid search specifically uses embeddings plus keywords.
E. Neither requires fine-tuning the generation model.
A. 混合搜索改进了检索到的候选段落集。
B. 重新排序器会优化排序,以便最好的段落首先出现。
C. 重新排名对已检索的候选者进行。
D. 混合搜索专门使用嵌入加关键字。
E. 两者都不需要微调生成模型。
A team's retrieval returns many candidates but the final answer quality is limited by irrelevant top results. With minimal overhead, what is the best next step?
团队的检索返回了许多候选者,但最终答案的质量受到不相关的顶级结果的限制。在开销最小的情况下,下一步最好的步骤是什么?
- A
Add a Bedrock reranker (Rerank API) to reorder candidates before generation
添加 Bedrock 重新排序器(Rerank API)以在生成之前重新排序候选者
- B
Switch the vector store to DynamoDB
将矢量存储切换到 DynamoDB
- C
Increase max tokens on the FM
增加 FM 上的最大令牌
- D
Disable chunking entirely
完全禁用分块
解析 / Explanation
正确答案 / Answer: A
A. Reranking is a low-overhead, managed way to push the most relevant passages to the top.
B. DynamoDB lacks native vector search and would not fix ranking.
C. More tokens do not improve which passages are most relevant.
D. Removing chunking typically hurts retrieval precision.
A. 重新排名是一种低开销、可管理的方法,可将最相关的段落推到顶部。
B. DynamoDB 缺乏本机矢量搜索,并且无法修复排名。
C. 更多标记并不能改善哪些段落最相关。
D. 删除分块通常会损害检索精度。
A team wants a managed, versioned library to create, parameterize, test, and approve prompt templates that multiple applications reuse. Which Amazon Bedrock feature provides this?
团队需要一个托管的、版本化的库来创建、参数化、测试和批准多个应用程序重复使用的提示模板。哪个 Amazon Bedrock 功能可以提供此功能?
- A
Amazon Bedrock Prompt Management
亚马逊基岩提示管理
- B
Amazon Bedrock Guardrails
亚马逊基岩护栏
- C
Amazon Bedrock Knowledge Bases
亚马逊基岩知识库
- D
AWS CodeCommit
AWS 代码提交
解析 / Explanation
正确答案 / Answer: A
A. Prompt Management stores versioned, parameterized prompt templates with collaboration and approval support.
B. Guardrails enforce content policies; they do not version prompts.
C. Knowledge Bases handle RAG, not prompt versioning.
D. CodeCommit is a Git repo, not a Bedrock prompt template service.
A. 提示管理存储版本化、参数化的提示模板,并提供协作和批准支持。
B. 护栏执行内容政策;他们没有版本提示。
C. 知识库处理 RAG,而不是提示版本控制。
D. CodeCommit 是一个 Git 存储库,而不是 Bedrock 提示模板服务。
A company must govern prompt templates with multi-stage approval workflows across business units, with minimal operational overhead. Which solution fits best?
公司必须通过跨业务部门的多阶段审批工作流程来管理提示模板,并以最小的运营开销。哪种解决方案最适合?
- A
Amazon Bedrock Prompt Management with approval workflows and IAM-based access
具有审批工作流程和基于 IAM 访问权限的 Amazon Bedrock 提示管理
- B
Store prompts in DynamoDB and build a custom approval app
在 DynamoDB 中存储提示并构建自定义审批应用程序
- C
Email prompt diffs for manual sign-off
手动签核的电子邮件提示差异
- D
Keep prompts only in application source code
仅在应用程序源代码中保留提示
解析 / Explanation
正确答案 / Answer: A
A. Prompt Management provides versioning, multi-stage approvals, and access control natively.
B. Custom approval logic adds significant development and maintenance overhead.
C. Manual email review is not scalable or auditable.
D. Embedding prompts in code lacks governed approval workflows and reuse.
A. 提示管理本身提供版本控制、多阶段批准和访问控制。
B. 自定义审批逻辑增加了大量的开发和维护开销。
C. 手动电子邮件审核不可扩展或不可审核。
D. 在代码中嵌入提示缺乏受监管的审批工作流程和重用。
A no-code builder is needed to visually chain prompts, models, and logic into a generative AI workflow without writing orchestration code. Which Bedrock feature is this?
需要一个无代码构建器来将提示、模型和逻辑直观地链接到生成式 AI 工作流程中,而无需编写编排代码。这是哪个基岩功能?
- A
Amazon Bedrock Prompt Flows
亚马逊基岩提示流程
- B
Amazon Bedrock Prompt Management
亚马逊基岩提示管理
- C
AWS Step Functions Workflow Studio
AWS Step Functions 工作流程工作室
- D
Amazon Bedrock Guardrails
亚马逊基岩护栏
解析 / Explanation
正确答案 / Answer: A
A. Prompt Flows is a visual, low/no-code builder for orchestrating prompts, models, and logic.
B. Prompt Management versions prompts; Prompt Flows orchestrates them into workflows.
C. Step Functions is general orchestration; the Bedrock-native visual prompt workflow builder is Prompt Flows.
D. Guardrails are safety policies, not a workflow builder.
A. Prompt Flows 是一个可视化、低代码/无代码构建器,用于编排提示、模型和逻辑。
B. 提示管理版本提示; Prompt Flows 将它们编排到工作流程中。
C. Step Functions 是通用编排; Bedrock 原生视觉提示工作流构建器是 Prompt Flows。
D. 护栏是安全策略,而不是工作流程构建器。
A developer wants the model to stop generating as soon as it produces a specific phrase or delimiter. Which inference parameter should they use?
开发人员希望模型在生成特定短语或分隔符后立即停止生成。他们应该使用哪个推理参数?
- A
stop_sequences
停止序列
- B
temperature
温度
- C
top_k
前k个
- D
max_tokens only
仅限 max_tokens
解析 / Explanation
正确答案 / Answer: A
A. Stop sequences deterministically halt generation when a specified string is produced.
B. Temperature controls randomness, not where generation stops.
C. top_k limits candidate tokens; it does not stop on a phrase.
D. max_tokens caps length but cannot stop precisely at a chosen phrase.
A. 当生成指定的字符串时,停止序列确定性地停止生成。
B. 温度控制随机性,而不是生成停止的地方。
C. top_k 限制候选令牌;它不会停留在一个短语上。
D. max_tokens 限制长度,但不能精确地停在所选短语处。
Which statement best contrasts stop_sequences, temperature, and top_k?
哪个语句最能对比 stop_sequences、温度和 top_k?
- A
stop_sequences control WHERE generation ends; temperature controls HOW random it is; top_k controls HOW MANY candidate tokens are considered
stop_sequences 控制生成在哪里结束;温度控制它的随机性; top_k 控制考虑多少个候选标记
- B
All three control where generation stops
所有三个控制发电停止的位置
- C
temperature controls stopping and top_k controls randomness
温度控制停止,top_k控制随机性
- D
They are all storage settings
它们都是存储设置
解析 / Explanation
正确答案 / Answer: A
A. Each parameter governs a different aspect: stopping, randomness, and candidate breadth.
B. Only stop_sequences controls stopping.
C. These roles are swapped/incorrect.
D. They are inference parameters, not storage settings.
A. 每个参数控制不同的方面:停止、随机性和候选宽度。
B. 只有 stop_sequences 控制停止。
C. 这些角色被交换/不正确。
D. 它们是推理参数,而不是存储设置。
A team wants more deterministic, focused output from a model for a compliance-sensitive summarization task. Which parameter change is most appropriate?
团队希望从模型中获得更具确定性、更集中的输出,以执行合规性敏感的摘要任务。更改哪个参数最合适?
- A
Lower the temperature toward 0
将温度降低到0
- B
Raise the temperature near maximum
将温度提高到接近最高温度
- C
Add more stop sequences to increase creativity
添加更多停止序列以提高创造力
- D
Increase max_tokens to reduce randomness
增加 max_tokens 以减少随机性
解析 / Explanation
正确答案 / Answer: A
A. Lower temperature reduces randomness, producing more deterministic, focused output.
B. High temperature increases randomness, the opposite of deterministic.
C. Stop sequences end generation; they do not control creativity.
D. max_tokens affects length, not determinism.
A. 较低的温度会减少随机性,从而产生更具确定性、更集中的输出。
B. 高温会增加随机性,与确定性相反。
C. 停止序列结束生成;他们无法控制创造力。
D. max_tokens 影响长度,而不影响确定性。
Which inference parameter limits sampling to the top K most probable next tokens?
哪个推理参数将采样限制为前 K 个最可能的下一个标记?
- A
top_k
前k个
- B
temperature
温度
- C
stop_sequences
停止序列
- D
presence_penalty
存在惩罚
解析 / Explanation
正确答案 / Answer: A
A. top_k restricts sampling to the K highest-probability tokens.
B. Temperature scales the probability distribution; it does not cap the candidate count.
C. Stop sequences end generation; they do not limit token candidates.
D. Presence penalty discourages repetition; it is not a top-K cap.
A. top_k 将采样限制为 K 个最高概率的标记。
B. 温度缩放概率分布;它不限制候选人数量。
C. 停止序列结束生成;他们不限制代币候选者。
D. 存在惩罚会阻止重复; it is not a top-K cap.
Which inference parameter uses cumulative probability (nucleus sampling) to choose the candidate token set?
哪个推理参数使用累积概率(核采样)来选择候选标记集?
- A
top_p
顶部_p
- B
top_k
前k个
- C
max_tokens
最大令牌数
- D
stop_sequences
停止序列
解析 / Explanation
正确答案 / Answer: A
A. top_p (nucleus sampling) selects the smallest set of tokens whose cumulative probability exceeds p.
B. top_k uses a fixed count, not cumulative probability.
C. max_tokens caps output length, not sampling.
D. Stop sequences end generation, not select candidates.
A. top_p(核心采样)选择累积概率超过 p 的最小 token 集合。
B. top_k 使用固定计数,而不是累积概率。
C. max_tokens 限制输出长度,而不是采样。
D. 停止序列结束生成,而不是选择候选序列。
A team needs an audit trail of who used which prompt and runtime logs for a governed prompt library. Which TWO AWS services support observability for Bedrock Prompt Management?
团队需要对谁使用了受控提示库的提示和运行时日志进行审计跟踪。哪两项 AWS 服务支持 Bedrock Prompt Management 的可观察性?
- A
AWS CloudTrail for API usage/auditing
用于 API 使用/审核的 AWS CloudTrail
- B
Amazon CloudWatch Logs for runtime logging
用于运行时日志记录的 Amazon CloudWatch Logs
- C
Amazon Macie
亚马逊麦西
- D
Amazon Polly
亚马逊波莉
- E
AWS Shield
AWS 盾
解析 / Explanation
正确答案 / Answer: A,B
A. CloudTrail records Bedrock API calls, providing an audit trail of prompt usage.
B. CloudWatch Logs captures runtime logs for monitoring prompt activity.
C. Macie discovers sensitive data in S3; it is not a Prompt Management observability tool.
D. Polly is text-to-speech, unrelated to observability.
E. Shield is DDoS protection, unrelated to prompt observability.
A. CloudTrail 记录 Bedrock API 调用,提供即时使用情况的审核跟踪。
B. CloudWatch Logs 捕获运行时日志以监控提示活动。
C. Macie 在 S3 中发现敏感数据;它不是一个提示管理可观察性工具。
D. Polly 是文本转语音,与可观察性无关。
E. Shield 是 DDoS 防护,与即时可观察性无关。
Which statement best distinguishes Bedrock Prompt Management from Bedrock Prompt Flows?
哪种说法最能区分基岩即时管理和基岩即时流程?
- A
Prompt Management is a versioned library of prompt templates; Prompt Flows is a visual builder for chaining prompts/models into workflows
提示管理是一个版本化的提示模板库; Prompt Flows 是一个可视化构建器,用于将提示/模型链接到工作流程中
- B
They are the same feature with different names
它们是相同的功能,但名称不同
- C
Prompt Flows stores versioned prompts and Prompt Management builds workflows
提示流存储版本化提示,提示管理构建工作流程
- D
Both are vector stores
两者都是矢量商店
解析 / Explanation
正确答案 / Answer: A
A. Management = store/version/approve prompts; Flows = orchestrate them visually.
B. They are distinct: one stores prompts, the other orchestrates workflows.
C. This reverses their roles.
D. Neither is a vector store.
A. 管理=存储/版本/审批提示;流程=以视觉方式编排它们。
B. 它们是不同的:一个存储提示,另一个协调工作流程。
C. 这颠倒了他们的角色。
D. 两者都不是矢量存储。
A developer notices the model's tool-call output is being cut off because generation continues past the expected end token. Which parameter helps cleanly terminate generation at the right boundary?
开发人员注意到模型的工具调用输出被切断,因为生成继续超过了预期的结束标记。哪个参数有助于在正确的边界处干净地终止生成?
- A
stop_sequences set to the expected end delimiter
stop_sequences 设置为预期的结束分隔符
- B
Increasing temperature
温度升高
- C
Increasing top_k
增加top_k
- D
Lowering max_tokens to 1
将 max_tokens 降低到 1
解析 / Explanation
正确答案 / Answer: A
A. Configuring stop sequences ends generation precisely at the delimiter, preventing overrun.
B. Higher temperature adds randomness; it does not control termination.
C. top_k changes candidate breadth, not stopping behavior.
D. Truncating to one token breaks the output rather than stopping at the right boundary.
A. 配置停止序列会在分隔符处精确结束生成,从而防止溢出。
B. 较高的温度会增加随机性;它不控制终止。
C. top_k 改变候选广度,而不是停止行为。
D. 截断为一个标记会破坏输出,而不是停在正确的边界处。
A team wants reusable prompts with typed input variables that applications fill at runtime (e.g., {{customer_name}}). Which capability supports parameterized prompt templates?
团队希望可重用的提示带有应用程序在运行时填充的键入输入变量(例如,{{customer_name}})。哪些功能支持参数化提示模板?
- A
Amazon Bedrock Prompt Management (parameterized prompt templates with versions)
Amazon Bedrock 提示管理(带有版本的参数化提示模板)
- B
Amazon Bedrock Guardrails
亚马逊基岩护栏
- C
Amazon SQS
亚马逊SQS
- D
AWS AppConfig feature flags
AWS AppConfig 功能标志
解析 / Explanation
正确答案 / Answer: A
A. Prompt Management supports variables/parameters in versioned templates for reuse.
B. Guardrails filter content; they do not template prompts.
C. SQS is messaging, not prompt templating.
D. AppConfig manages config, not parameterized prompt templates with versioning.
A. 提示管理支持版本化模板中的变量/参数以供重用。
B. 护栏过滤内容;他们没有模板提示。
C. SQS 是消息传递,而不是提示模板。
D. AppConfig 管理配置,而不是具有版本控制的参数化提示模板。
Which scenario is the best use case for Bedrock Prompt Flows?
哪种场景是基岩提示流的最佳用例?
- A
Visually orchestrating a multi-step GenAI process (e.g., classify → retrieve → summarize) without writing orchestration code
可视化编排多步骤 GenAI 流程(例如,分类 → 检索 → 总结),无需编写编排代码
- B
Storing 7 years of model invocation logs
存储7年的模型调用日志
- C
Detecting PII in documents
检测文档中的 PII
- D
Reserving model capacity
预留模型容量
解析 / Explanation
正确答案 / Answer: A
A. Prompt Flows chains steps and models visually for multi-step GenAI workflows.
B. That is invocation logging to S3, not Prompt Flows.
C. That is Comprehend, not Prompt Flows.
D. That is Provisioned Throughput.
A. Prompt Flows 以可视方式链接步骤和模型,以实现多步骤 GenAI 工作流程。
B. 这是对 S3 的调用日志记录,而不是提示流。
C. 那是理解,而不是即时流动。
D. 这就是预置吞吐量。
A team wants to compare two versions of a prompt template and roll back if the new one performs worse. Which feature supports this?
团队想要比较提示模板的两个版本,如果新版本表现较差,则回滚。哪个功能支持这个?
- A
Prompt versioning in Bedrock Prompt Management
Bedrock Prompt Management 中的提示版本控制
- B
Cross-Region inference profiles
跨区域推理配置文件
- C
HNSW index tuning
HNSW 指数调整
- D
Stop sequences
停止序列
解析 / Explanation
正确答案 / Answer: A
A. Prompt Management versions templates so you can compare and revert.
B. Profiles route traffic; they do not version prompts.
C. That is vector index tuning, unrelated to prompts.
D. Stop sequences end generation; they do not version prompts.
A. 提示管理版本模板,以便您可以比较和恢复。
B. 配置文件路由流量;他们没有版本提示。
C. 那是向量索引调整,与提示无关。
D. 停止序列结束生成;他们没有版本提示。
What does raising the temperature parameter generally do to model output?
提高温度参数通常会对模型输出产生什么影响?
- A
Increases randomness/creativity and diversity of responses
增加响应的随机性/创造力和多样性
- B
Forces the model to stop sooner
强制模型更快停止
- C
Guarantees factual accuracy
保证事实准确性
- D
Reduces the vocabulary to one token
将词汇量减少到一个标记
解析 / Explanation
正确答案 / Answer: A
A. Higher temperature flattens the distribution, yielding more varied output.
B. Stopping is controlled by stop_sequences/max_tokens, not temperature.
C. Temperature does not control factual correctness.
D. It does not collapse the vocabulary to a single token.
A. 较高的温度使分布变得平坦,从而产生更加多样化的输出。
B. 停止由 stop_sequences/max_tokens 控制,而不是温度。
C. 温度不能控制事实的正确性。
D. 它不会将词汇表折叠成单个标记。
A regulated workload requires capturing full prompt content and model responses for 7 years with immutable retention and minimal overhead. Which combination is best?
受监管的工作负载需要在 7 年内捕获完整的提示内容和模型响应,并以不可变的保留和最小的开销。哪种组合最好?
- A
Bedrock model invocation logging to Amazon S3 with S3 Object Lock (compliance mode) for 7-year retention
使用 S3 对象锁定(合规模式)将 Bedrock 模型调用记录到 Amazon S3,保留 7 年
- B
CloudTrail data events only
仅 CloudTrail 数据事件
- C
Prompt Flows execution history only
仅提示流程执行历史记录
- D
Storing logs in an ephemeral cache
将日志存储在临时缓存中
解析 / Explanation
正确答案 / Answer: A
A. Invocation logging captures prompts/responses; S3 Object Lock enforces immutable retention.
B. CloudTrail logs API metadata, not full prompt/response content.
C. Flow history is not a compliant 7-year immutable record of all invocations.
D. Ephemeral storage cannot meet immutable long-term retention.
A. 调用日志记录捕获提示/响应; S3 对象锁强制执行不可变保留。
B. CloudTrail 记录 API 元数据,而不是完整的提示/响应内容。
C. 流历史记录不是所有调用的 7 年不可变记录。
D. 临时存储无法满足不可变的长期保留。
A developer relies on a prompt instruction ("please stop after the summary") to halt generation, but the model sometimes ignores it. What is the more reliable mechanism?
开发人员依靠提示指令(“请在摘要后停止”)来停止生成,但模型有时会忽略它。更可靠的机制是什么?
- A
Use the stop_sequences inference parameter
使用 stop_sequences 推理参数
- B
Repeat the instruction three times in the prompt
重复提示中的指令三遍
- C
Lower top_p to 0
将 top_p 降低到 0
- D
Increase temperature
提高温度
解析 / Explanation
正确答案 / Answer: A
A. Stop sequences are enforced by the API and reliably terminate generation, unlike prompt-only instructions.
B. Restating instructions does not reliably enforce stopping.
C. top_p shapes candidate selection, not stopping.
D. That adds randomness and makes behavior less controlled.
A. 与仅提示指令不同,停止序列由 API 强制执行并可靠地终止生成。
B. 重述指令并不能可靠地强制停止。
C. top_p 塑造候选选择,而不是停止。
D. 这增加了随机性并使行为更难以控制。
A team wants consistent, governed prompts injected into a real-time application that routes by request type, with the ability to update prompts without redeploying code. Which approach is most aligned?
团队希望将一致的、受管控的提示注入到按请求类型路由的实时应用程序中,并且能够在不重新部署代码的情况下更新提示。哪种方法最一致?
- A
Manage prompts centrally in Bedrock Prompt Management and reference versioned templates at runtime
在基岩提示管理中集中管理提示并在运行时参考版本化模板
- B
Hard-code each prompt variant in the Lambda source
对 Lambda 源中的每个提示变体进行硬编码
- C
Store prompts only in client-side JavaScript
仅在客户端 JavaScript 中存储提示
- D
Use Provisioned Throughput to manage prompts
使用预配置吞吐量来管理提示
解析 / Explanation
正确答案 / Answer: A
A. Central versioned templates let you update governed prompts without code redeploys.
B. Hard-coding requires redeploying to change prompts and lacks governance.
C. Client-side prompts are ungoverned and insecure.
D. Provisioned Throughput reserves capacity; it does not manage prompts.
A. 中央版本控制模板允许您更新受管理的提示,而无需重新部署代码。
B. 硬编码需要重新部署以更改提示,并且缺乏治理。
C. 客户端提示不受监管且不安全。
D. 预留吞吐量储备能力;它不管理提示。
A team wants a managed agent on Amazon Bedrock that can reason over a task, call APIs/tools, and consult a knowledge base, without building the orchestration loop themselves. Which feature fits?
一个团队希望在 Amazon Bedrock 上有一个托管代理,它可以推理任务、调用 API/工具并查阅知识库,而无需自行构建编排循环。适合哪个功能?
- A
Amazon Bedrock Agents
亚马逊基岩代理
- B
Amazon Bedrock Guardrails
亚马逊基岩护栏
- C
Amazon Bedrock Prompt Flows
亚马逊基岩提示流程
- D
Amazon SageMaker Processing
Amazon SageMaker 处理
解析 / Explanation
正确答案 / Answer: A
A. Bedrock Agents are managed agents that orchestrate reasoning, tool/API calls, and knowledge-base lookups.
B. Guardrails enforce content safety; they are not an agent runtime.
C. Prompt Flows orchestrate prompt steps but are not the managed agent feature.
D. Processing runs data jobs, not agents.
A. Bedrock 代理是托管代理,可编排推理、工具/API 调用和知识库查找。
B. 护栏加强内容安全;它们不是代理运行时。
C. 提示流编排提示步骤,但不是托管代理功能。
D. 处理运行数据作业,而不是代理。
How can tools (actions) be defined for an Amazon Bedrock Agent? (Select TWO.)
如何为 Amazon Bedrock 代理定义工具(操作)? (选择两项。)
- A
With an OpenAPI schema describing the API actions
使用描述 API 操作的 OpenAPI 架构
- B
With function definitions (function details) for the actions
具有操作的函数定义(函数详细信息)
- C
By uploading a vector index
通过上传矢量索引
- D
By setting the temperature parameter
通过设置温度参数
- E
By creating an S3 lifecycle policy
通过创建 S3 生命周期策略
解析 / Explanation
正确答案 / Answer: A,B
A. Bedrock Agents accept an OpenAPI schema to define callable API actions.
B. Agents also support defining actions via function definitions.
C. A vector index is for retrieval, not tool definitions.
D. Temperature is an inference setting, not a tool definition.
E. Lifecycle policies manage object storage, not agent tools.
A. Bedrock Agent 接受 OpenAPI 架构来定义可调用的 API 操作。
B. 代理还支持通过函数定义来定义操作。
C. 矢量索引用于检索,而不是工具定义。
D. 温度是一个推断设置,而不是工具定义。
E. 生命周期策略管理对象存储,而不是代理工具。
A developer is debugging why a Bedrock Agent chose certain steps. Which capability lets them inspect the agent's reasoning/decision steps?
开发人员正在调试基岩代理选择某些步骤的原因。哪种功能可以让他们检查代理的推理/决策步骤?
- A
Agent tracing
代理追踪
- B
S3 access logs
S3访问日志
- C
VPC Flow Logs
VPC流日志
- D
AWS Config timeline
AWS 配置时间表
解析 / Explanation
正确答案 / Answer: A
A. Agent tracing exposes the agent's reasoning steps and tool calls for debugging.
B. S3 access logs track object requests, not agent reasoning.
C. Flow Logs capture network traffic, not agent decisions.
D. Config tracks resource changes, not agent reasoning.
A. 代理跟踪公开代理的推理步骤和工具调用以进行调试。
B. S3 访问日志跟踪对象请求,而不是代理推理。
C. 流日志捕获网络流量,而不是代理决策。
D. 配置跟踪资源更改,而不是代理推理。
A company needs to deploy and operate AI agents at scale in production, with managed serverless runtime, identity, tool gateway, memory, and observability — usable with multiple agent frameworks. Which AWS offering targets this?
公司需要在生产中大规模部署和操作人工智能代理,并具有托管的无服务器运行时、身份、工具网关、内存和可观察性——可与多个代理框架一起使用。哪些 AWS 产品针对此目标?
- A
Amazon Bedrock AgentCore
亚马逊基岩代理核心
- B
Amazon Bedrock Guardrails
亚马逊基岩护栏
- C
Amazon Bedrock Knowledge Bases
亚马逊基岩知识库
- D
Amazon SageMaker Feature Store
Amazon SageMaker 功能商店
解析 / Explanation
正确答案 / Answer: A
A. AgentCore provides production-grade, framework-agnostic services (Runtime, Identity, Gateway, Memory, Observability, etc.) to deploy and scale agents.
B. Guardrails are safety policies, not an agent operations platform.
C. Knowledge Bases provide RAG, not agent runtime/identity/observability.
D. Feature Store manages ML features, not agents.
A. AgentCore 提供生产级、与框架无关的服务(运行时、身份、网关、内存、可观察性等)来部署和扩展代理。
B. 护栏是安全策略,而不是代理操作平台。
C. 知识库提供 RAG,而不是代理运行时/身份/可观察性。
D. 特征存储管理 ML 特征,而不是代理。
Which AgentCore component provides a secure, serverless runtime that can host agents built with various frameworks (LangGraph, CrewAI, Strands, etc.) and protocols (MCP, A2A)?
哪个 AgentCore 组件提供安全的无服务器运行时,可以托管使用各种框架(LangGraph、CrewAI、Strands 等)和协议(MCP、A2A)构建的代理?
- A
AgentCore Runtime
AgentCore运行时
- B
AgentCore Memory
代理核心内存
- C
AgentCore Gateway
AgentCore网关
- D
AgentCore Identity
代理核心身份
解析 / Explanation
正确答案 / Answer: A
A. AgentCore Runtime is the secure serverless runtime for deploying/scaling agents across frameworks, protocols, and models.
B. Memory manages context retention, not the runtime host.
C. Gateway turns APIs/Lambda into tools; it is not the runtime.
D. Identity handles access control, not hosting.
A. AgentCore Runtime 是安全的无服务器运行时,用于跨框架、协议和模型部署/扩展代理。
B. 内存管理上下文保留,而不是运行时主机。
C. Gateway 将 API/Lambda 转变为工具;这不是运行时。
D. 身份处理访问控制,而不是托管。
Which AgentCore component turns existing REST APIs and AWS Lambda functions into agent-ready tools (including via MCP) with tool discovery?
哪个 AgentCore 组件通过工具发现将现有 REST API 和 AWS Lambda 函数转变为代理就绪工具(包括通过 MCP)?
- A
AgentCore Gateway
AgentCore网关
- B
AgentCore Browser
AgentCore浏览器
- C
AgentCore Code Interpreter
AgentCore 代码解释器
- D
AgentCore Observability
AgentCore 可观测性
解析 / Explanation
正确答案 / Answer: A
A. Gateway converts APIs and Lambda functions into agent tools and supports tool discovery over protocols like MCP.
B. Browser provides managed web browsing, not API-to-tool conversion.
C. Code Interpreter runs code; it does not expose APIs as tools.
D. Observability monitors agents; it does not create tools.
A. Gateway 将 API 和 Lambda 函数转换为代理工具,并支持通过 MCP 等协议进行工具发现。
B. 浏览器提供托管 Web 浏览,而不是 API 到工具的转换。
C. 代码解释器运行代码;它不将 API 作为工具公开。
D. 可观测性监控代理;它不创造工具。
Which AgentCore component manages both short-term (session) and long-term memory so agents can retain context and learn from past interactions?
哪个 AgentCore 组件管理短期(会话)和长期记忆,以便代理可以保留上下文并从过去的交互中学习?
- A
AgentCore Memory
代理核心内存
- B
AgentCore Identity
代理核心身份
- C
AgentCore Runtime
AgentCore运行时
- D
AgentCore Gateway
AgentCore网关
解析 / Explanation
正确答案 / Answer: A
A. Memory provides session and long-term memory for context retention across interactions.
B. Identity handles secure access, not memory.
C. Runtime hosts/scales agents; memory is a separate component.
D. Gateway exposes tools, not memory.
A. 记忆提供会话记忆和长期记忆,以便在交互过程中保留上下文。
B. 身份处理安全访问,而不是内存。
C. 运行时主机/缩放代理;内存是一个单独的组件。
D. 网关公开工具,而不是内存。
An agent must securely access third-party services like GitHub, Salesforce, and Slack on behalf of a user, with pre-authorized consent. Which AgentCore component handles this?
代理必须在预先授权同意的情况下代表用户安全地访问 GitHub、Salesforce 和 Slack 等第三方服务。哪个 AgentCore 组件处理这个问题?
- A
AgentCore Identity
代理核心身份
- B
AgentCore Memory
代理核心内存
- C
AgentCore Code Interpreter
AgentCore 代码解释器
- D
AgentCore Browser
AgentCore浏览器
解析 / Explanation
正确答案 / Answer: A
A. Identity lets agents securely access AWS and third-party services on behalf of users or with pre-authorized consent.
B. Memory stores context; it does not broker third-party access.
C. Code Interpreter runs code, not identity brokering.
D. Browser automates web sessions; access/consent is Identity's role.
A. 身份允许代理代表用户或在预先授权同意的情况下安全地访问 AWS 和第三方服务。
B. 内存存储上下文;它不代理第三方访问。
C. 代码解释器运行代码,而不是身份代理。
D. 浏览器自动执行网络会话;访问/同意是身份的角色。
An agent needs to run code it generates in a secure, isolated sandbox (e.g., for data analysis). Which AgentCore component provides this?
代理需要在安全、隔离的沙箱中运行它生成的代码(例如,用于数据分析)。哪个 AgentCore 组件提供此功能?
- A
AgentCore Code Interpreter
AgentCore 代码解释器
- B
AgentCore Gateway
AgentCore网关
- C
AgentCore Identity
代理核心身份
- D
AgentCore Memory
代理核心内存
解析 / Explanation
正确答案 / Answer: A
A. Code Interpreter offers an isolated environment to execute agent-generated code.
B. Gateway exposes tools; it does not run arbitrary generated code in a sandbox.
C. Identity manages access, not code execution.
D. Memory stores context, not a code sandbox.
A. 代码解释器提供了一个隔离的环境来执行代理生成的代码。
B. 网关公开工具;它不会在沙箱中运行任意生成的代码。
C. 身份管理访问,而不是代码执行。
D. 内存存储上下文,而不是代码沙箱。
An agent must perform web automation (navigating sites, filling forms) at scale using managed browser instances. Which AgentCore component fits?
代理必须使用托管浏览器实例大规模执行 Web 自动化(导航站点、填写表单)。哪个 AgentCore 组件适合?
- A
AgentCore Browser
AgentCore浏览器
- B
AgentCore Observability
AgentCore 可观测性
- C
AgentCore Runtime
AgentCore运行时
- D
AgentCore Memory
代理核心内存
解析 / Explanation
正确答案 / Answer: A
A. Browser provides managed web browser instances for agent web-automation workflows.
B. Observability monitors agents; it does not browse the web.
C. Runtime hosts agents; managed browsing is the Browser component.
D. Memory is context retention, not browsing.
A. 浏览器为代理 Web 自动化工作流程提供托管 Web 浏览器实例。
B. 可观测性监控代理;它不浏览网页。
C. 运行时主机代理;托管浏览是浏览器组件。
D. 记忆是上下文保留,而不是浏览。
A team needs step-by-step visualization of an agent's execution, trajectory inspection, and troubleshooting filters in production. Which AgentCore component provides this?
团队需要对生产中的代理执行、轨迹检查和故障排除过滤器进行逐步可视化。哪个 AgentCore 组件提供此功能?
- A
AgentCore Observability
AgentCore 可观测性
- B
AgentCore Gateway
AgentCore网关
- C
AgentCore Identity
代理核心身份
- D
AgentCore Code Interpreter
AgentCore 代码解释器
解析 / Explanation
正确答案 / Answer: A
A. Observability gives step-by-step execution visibility, trajectory inspection, and debugging filters.
B. Gateway exposes tools, not observability.
C. Identity manages access, not monitoring.
D. Code Interpreter runs code; it is not the monitoring component.
A. 可观察性提供了逐步执行的可见性、轨迹检查和调试过滤器。
B. 网关公开工具,而不是可观察性。
C. 身份管理访问,而不是监视。
D. 代码解释器运行代码;它不是监控组件。
Which statement about how AgentCore's services are consumed is correct?
关于 AgentCore 的服务如何使用的哪种说法是正确的?
- A
Each service (Runtime, Gateway, Identity, Memory, Observability, Browser, Code Interpreter) can be used independently or together, paying only for what you use
每个服务(运行时、网关、身份、内存、可观察性、浏览器、代码解释器)都可以单独使用或一起使用,只需为您使用的服务付费
- B
You must adopt all AgentCore services as a single bundle
您必须将所有 AgentCore 服务作为一个捆绑包采用
- C
AgentCore only works with Amazon Nova models
AgentCore 仅适用于 Amazon Nova 型号
- D
AgentCore only supports one agent framework
AgentCore仅支持一种代理框架
解析 / Explanation
正确答案 / Answer: A
A. AgentCore is modular — adopt individual services or combine them, with pay-as-you-go pricing.
B. Services are modular and can be used independently.
C. AgentCore Runtime is model-agnostic (Nova, OpenAI, Gemini, etc.).
D. It supports many frameworks (LangGraph, CrewAI, Strands, etc.).
A. AgentCore 是模块化的——采用单独的服务或将它们组合起来,并采用即用即付的定价方式。
B. 服务是模块化的,可以独立使用。
C. AgentCore Runtime 与模型无关(Nova、OpenAI、Gemini 等)。
D. 它支持许多框架(LangGraph、CrewAI、Strands 等)。
How does AgentCore relate to Amazon Bedrock Agents?
AgentCore 与 Amazon Bedrock Agent 有何关系?
- A
AgentCore extends agent capabilities toward production scale (serverless runtime, identity, gateway, memory, observability) and works across frameworks, beyond a single managed-agent construct
AgentCore 将代理功能扩展到生产规模(无服务器运行时、身份、网关、内存、可观察性),并跨框架工作,超越单一托管代理构造
- B
AgentCore is an older, deprecated version of Bedrock Agents
AgentCore 是 Bedrock Agents 的较旧且已弃用的版本
- C
AgentCore only stores prompts
AgentCore仅存储提示
- D
AgentCore is a vector database
AgentCore是一个矢量数据库
解析 / Explanation
正确答案 / Answer: A
A. AgentCore is the newer, production-grade, framework-agnostic platform for operating agents at scale.
B. AgentCore is the newer offering, not deprecated.
C. That describes Prompt Management, not AgentCore.
D. AgentCore is an agent platform, not a vector store.
A. AgentCore 是用于大规模运营代理的新型、生产级、与框架无关的平台。
B. AgentCore 是较新的产品,并未弃用。
C. 这描述的是提示管理,而不是 AgentCore。
D. AgentCore 是一个代理平台,而不是矢量商店。