1Z0-1127-25 Practice Exam 52 — 20 Questions

Question 1

A company is using OCI Generative AI Service with a dedicated AI cluster. They notice that during peak hours, inference requests are queuing and latency increases significantly. They want to improve throughput without changing the model or the application code. Which action should they take?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Increase the number of model units in the dedicated AI cluster.. For a dedicated AI cluster, throughput is directly tied to the number of provisioned model units. Increasing model units adds more serving replicas, allowing the cluster to process more requests concurrently. This is the correct scaling action and does not require application changes or model modifications, making it the ideal solution for peak-hour latency issues.

Answer

Switch to an on-demand inference endpoint for the same model.

Answer

Enable auto-scaling on the dedicated AI cluster.

Answer

Reduce the max_tokens parameter in the inference requests.

Question 2

A machine learning engineer has deployed a custom model on an OCI Generative AI dedicated AI cluster. The model is used for real-time chatbot interactions. The engineer notices that response times are inconsistent, with occasional long delays. Which OCI service should the engineer use to monitor the latency of inference requests and identify bottlenecks?

Accepted Answer

OCI Monitoring. OCI Monitoring is designed to collect and analyze metrics from OCI resources, including dedicated AI clusters. It provides latency, request count, and error rate metrics that can be used to diagnose performance issues. With alarms and dashboards, the engineer can track response times and pinpoint bottlenecks, making it the correct choice.

Answer

OCI Notifications

Answer

OCI Audit

Answer

OCI Logging

Question 3

A data scientist is preparing a dataset to fine-tune a T-Few model in OCI Generative AI and wants the training to be effective. The task is to classify support tickets into categories. Which dataset format best matches the recommended approach for this tuning method?

Accepted Answer

Provide a modest set of high-quality labeled examples that pair each ticket with its target category, formatted as prompt-and-completion records.. T-Few fine-tuning is built to learn from a small, well-curated set of supervised examples, so prompt-and-completion pairs linking each support ticket to its category give the job the signal it needs. Unlabeled data, label lists alone, or raw weight files do not supply the input-output relationships required for effective classification tuning.

Answer

Provide only the category names as a comma-separated list so the model learns the label space without example tickets.

Answer

Provide raw model weights from a previous training run so the fine-tuning job can continue from that checkpoint.

Answer

Supply hundreds of thousands of unlabeled support tickets and let the fine-tuning job infer the categories during training.

Question 4

A team fine-tuned a model with OCI Generative AI and now needs every inference call to be attributable to a specific application for chargeback reporting, while keeping the model private to their tenancy. Which combination of actions satisfies this?

Accepted Answer

Deploy the model to a dedicated AI cluster endpoint and use OCI tagging plus the Generative AI usage and cost reports to attribute consumption per application.. Placing the fine-tuned model on a dedicated AI cluster endpoint keeps it on reserved, tenancy-isolated capacity, and because the endpoint is an OCI resource it can carry tags. Combining those tags with the Generative AI usage and cost reports produces accurate per-application attribution. The other approaches either break model privacy or substitute estimates for actual service accounting.

Answer

Publish the fine-tuned model to a public endpoint and use an API gateway usage plan to meter each application's calls.

Answer

Use the OCI Logging service to capture inference request payloads and derive per-application cost from the token counts in the logs.

Answer

Deploy the model to an on-demand endpoint and tag each request with a custom HTTP header that the service bills against.

Question 5

An administrator wants to confirm which OCI Generative AI models are available for a new project in a specific region before writing any code. Which action gives the authoritative, current list?

Accepted Answer

Call the Generative AI service's list models operation through the OCI CLI or SDK in that region.. The service API is the source of truth for regional model availability, so the list models operation through the CLI or SDK returns exactly what can be used in that region, including deployment mode differences. Documentation, pricing pages, and IAM policies describe other things and can be out of date or region-agnostic.

Answer

Read the general OCI Generative AI documentation page listing all supported foundation models worldwide.

Answer

Inspect the tenancy's IAM policies to see which model names appear in the policy statements.

Answer

Check the OCI pricing page for Generative AI and assume every listed model can be deployed.

Question 6

You are designing a RAG pipeline on OCI Generative AI where user queries often contain abbreviations and internal jargon not present in the embedding model's training data. Retrieval recall for these queries is poor. Which technique should you apply before embedding the query to improve retrieval?

Accepted Answer

Query expansion or rewriting that maps abbreviations and jargon to their canonical forms. When user queries use abbreviations or internal jargon absent from the embedding model's training data, the query vector does not align with the document vectors. Query expansion or rewriting translates those terms into canonical forms before embedding, closing the vocabulary gap. This is a targeted, low-cost intervention that improves recall without retraining or re-indexing the corpus.

Answer

Switch to a larger embedding model with more parameters and re-embed the entire corpus

Answer

Increase the chunk size of indexed documents so more context is captured per vector

Answer

Lower the top-k retrieval parameter to reduce noise in the candidate set

Question 7

A developer is building a multilingual customer feedback summarizer on OCI Generative AI. They want one embedding model to index feedback in English, Spanish, and Japanese so that a single semantic search can retrieve relevant items across all three languages. Which model characteristic should they prioritize?

Accepted Answer

A multilingual embedding model that maps different languages into a shared vector space. One semantic search across English, Spanish, and Japanese requires embeddings where equivalent meanings from different languages are close together. A multilingual embedding model provides that shared vector space, so a query in any supported language retrieves relevant items regardless of the source language. Context window, generation output limits, and English-only tuning do not deliver cross-lingual alignment.

Answer

A model fine-tuned only on English customer support transcripts

Answer

A model with the highest possible output token limit for generation

Answer

A model with the largest maximum token context window for embeddings

Question 8

A developer needs to call an OCI Generative AI chat model from a Python script running on a local workstation. The developer has an API key configured in the OCI CLI profile and wants to authenticate the SDK calls using that same profile. Which authentication method should the developer configure for the Generative AI client?

Accepted Answer

API key-based authentication using the profile in the OCI config file. API key authentication uses the credentials stored in the OCI config file profile to sign SDK requests. Because the developer already created an API key for the CLI, referencing that profile in the Generative AI client is the correct and simplest approach. Instance, resource, and security token methods are designed for different execution environments or token lifetimes.

Answer

Security token authentication with a session token from the OCI Console

Answer

Instance principals

Answer

Resource principals

Question 9

A team is building a RAG application on OCI Generative AI and wants to reduce hallucinated answers when the retrieved context does not contain the answer. Which TWO techniques should they apply? (Choose two.)

Accepted Answer

Instruct the LLM in the prompt to answer only from the provided context and to say it does not know when the context is insufficient. Grounding instructions and post-generation verification are complementary controls. The prompt instructs the model to answer only from context and to abstain otherwise, while verification independently checks that claims are supported by retrieved passages. Together they reduce fabrication when retrieval fails to provide the needed evidence, without relying on chance or on increasing context volume.

Answer

Increase the LLM temperature to encourage more creative and complete answers

Answer

Remove the retrieved context from the prompt to avoid confusing the model with irrelevant passages

Answer

Increase the number of retrieved chunks to the maximum supported by the context window

Question 10

A team has a fine-tuned model in OCI Generative AI that must serve production inference. Compliance requires the model to be hosted on hardware isolated from other tenants and to support a predictable throughput baseline. Which deployment option should the team choose?

Accepted Answer

A dedicated AI cluster hosting the fine-tuned model with reserved hosting units. Dedicated AI clusters allocate isolated compute in hosting units to a single customer and support hosting fine-tuned models. This provides both the tenant isolation demanded by compliance and a predictable throughput baseline from the reserved units. Shared on-demand inference, notebook sessions, and serverless wrappers do not deliver those guarantees.

Answer

A custom model imported into OCI Data Science and served from a notebook session

Answer

On-demand model inference using the shared endpoint for the base model

Answer

An OCI Functions endpoint that invokes the model on each request

Question 11

A developer is building a RAG chatbot on OCI Generative AI. They need to store document embeddings and perform similarity search. Which OCI service is purpose-built for storing and querying vector embeddings at scale?

Accepted Answer

OCI OpenSearch. OCI OpenSearch is the managed service that supports vector fields and k-nearest-neighbor search, making it the purpose-built choice for storing and querying embeddings in a RAG pipeline on OCI. Object Storage, Data Science, and Streaming serve different roles: storage of raw files, model development, and data transport, respectively.

Answer

OCI Data Science

Answer

OCI Object Storage

Answer

OCI Streaming

Question 12

A retail company is building a product-question answering assistant with OCI Generative AI. Product descriptions live in an Oracle Database 23ai table. The team wants the assistant to answer questions such as 'Which waterproof jackets are under 200 dollars?' by retrieving the most relevant product rows before generation. Which approach correctly uses the database for this retrieval step?

Accepted Answer

Generate an embedding for the user question with the same OCI embedding model used for the product rows, then query the VECTOR column for the nearest neighbors.. Vector retrieval compares a query embedding against stored document embeddings, so the query must be embedded with the same model that produced the stored vectors. That yields the nearest product rows by semantic similarity, which are then supplied to the LLM. The other approaches either avoid embeddings, recompute them inefficiently, or rely on lexical matching that misses paraphrases.

Answer

Send the raw user question to the LLM and let the model generate SQL that selects product rows by keyword matching on the description column.

Answer

Store the product descriptions as a JSON document in the table and use full-text search on the description field to return matching rows.

Answer

Call the OCI Generative AI embedding endpoint for every product row at query time and compare the vectors in the application memory.

Question 13

A company wants to fine-tune a model in OCI Generative AI using a dataset stored in an OCI Object Storage bucket. The training job fails immediately with a permissions error even though the data scientist can download the file manually from the console. What is the most likely cause?

Accepted Answer

The fine-tuning job lacks an IAM policy allowing the service to read objects from the bucket. Fine-tuning jobs access Object Storage using a service principal, so the job's permissions are separate from the user's. A policy must explicitly allow the Generative AI service to read the bucket or objects. The data scientist's manual download works because it uses the user's own credentials, which masks the missing service-level authorization.

Answer

The fine-tuning job was created with a public endpoint instead of a private endpoint

Answer

The Object Storage bucket is in a different region than the Generative AI fine-tuning job

Answer

The dataset file is larger than the maximum size supported for fine-tuning

Question 14

A healthcare analytics team is deploying a RAG service on OCI Generative AI with a vector store in Oracle Database 23ai. Clinicians report that answers sometimes cite outdated guideline versions even though the latest guidelines were ingested. The team confirms embeddings and index are current. Which TWO actions best improve retrieval precision for the latest guideline content? (Choose two.)

Accepted Answer

Apply a reranking step that scores retrieved chunks against the query and favors chunks with higher lexical and semantic relevance before building the prompt.. Precision improves when the retrieval stage excludes superseded content and ranks the remaining candidates more accurately. Metadata filtering removes outdated guideline versions from the eligible set, while reranking promotes the chunks that best match the query. Together they ensure the prompt contains current guidance, which the generation stage then summarizes.

Answer

Reduce the embedding dimension to speed up similarity computation so more candidate chunks can be evaluated per query.

Answer

Increase the generation model's temperature so the LLM explores alternative phrasings and avoids copying outdated text.

Answer

Lower the top-k value so fewer chunks are retrieved, on the assumption that the most recent guideline will always rank first.

Question 15

A developer is deploying a custom model to an OCI Generative AI dedicated AI cluster. The model artifacts are stored in an OCI Object Storage bucket. When creating the model, the deployment fails with an error indicating that the service cannot access the model artifacts. What is the most likely cause?

Accepted Answer

The service principal for OCI Generative AI does not have permission to read from the Object Storage bucket.. The most likely cause is that the OCI Generative AI service principal lacks the required IAM policy to read the model artifacts from Object Storage. When deploying a model, the service assumes a service principal identity and requires explicit permission to access the bucket. You must create a policy that allows the service to read objects in the specified compartment or bucket. Without this, the deployment fails with an access error.

Answer

The model artifacts are in a private bucket and must be made public for the service to access them.

Answer

The Object Storage bucket is in a different region than the dedicated AI cluster.

Answer

The dedicated AI cluster does not have a service gateway configured to reach Object Storage.

Question 16

An application needs to generate embeddings for text in multiple languages (English, Spanish, French). Which OCI Generative AI embedding model should be used?

Accepted Answer

embed-multilingual-v3.0. The OCI Generative AI embedding model embed-multilingual-v3.0 is specifically designed to generate embeddings for text in multiple languages, including English, Spanish, and French. It supports over 100 languages and is optimized for cross-lingual tasks, making it the correct choice for multilingual embedding needs.

Answer

embed-english-v3.0

Answer

Cohere Command R

Answer

Meta Llama 3

Question 17

A developer wants to index a large corpus of HTML web pages for a RAG pipeline using LangChain. They need to load the content from URLs, split the text into chunks, and generate embeddings. Which combination of LangChain components should they use?

Accepted Answer

WebBaseLoader, RecursiveCharacterTextSplitter, OCIGenAIEmbeddings. WebBaseLoader is designed to load and parse HTML content from URLs, making it the correct loader for web pages. RecursiveCharacterTextSplitter is the standard splitter for chunking text while preserving semantic boundaries, and OCIGenAIEmbeddings generates embeddings using OCI's Generative AI service. This combination covers loading, splitting, and embedding for a RAG pipeline.

Answer

PDFLoader, RecursiveCharacterTextSplitter, OCIGenAIEmbeddings

Answer

CSVLoader, RecursiveCharacterTextSplitter, OCIGenAIEmbeddings

Answer

WebBaseLoader, TokenTextSplitter, OCIGenAI

Question 18

Which THREE OCI Generative AI service features help in controlling the cost of API calls? (Select three.)

Accepted Answer

Using a smaller model. Option A (Using a smaller model) is correct because OCI Generative AI prices inference per token, and smaller models (e.g., lower-parameter or lighter-tier models) have lower per-token rates, directly reducing the cost of each API call. Option C (Using stop sequences) is correct because stop sequences terminate generation as soon as a specified string is produced, cutting the number of output tokens billed and thus lowering cost. Option D (Setting max_tokens limit) is correct because it caps the number of tokens the model can generate, preventing unexpectedly long completions from inflating output-token charges. Option B (Increasing temperature) does not belong because temperature only affects randomness/creativity of sampling, not token counts or pricing. Option E (Enabling response streaming) does not belong because streaming only changes how tokens are delivered to the client, not how many tokens are generated or billed.

Answer

Increasing temperature

Answer

Enabling response streaming

Question 19

A developer is deploying a LangChain agent that uses an LLM from OCI Generative AI. The agent interacts with external APIs and must handle rate limits gracefully. Which TWO practices should the developer implement?

Accepted Answer

Implement retry logic with exponential backoff for API calls. Option C is correct because implementing retry logic with exponential backoff is the standard way to handle transient rate-limit errors (such as HTTP 429 responses) from external APIs and OCI Generative AI endpoints, spacing out retries so the agent recovers gracefully instead of failing immediately. Option E is correct because a rate limiter proactively controls the frequency of outgoing API requests, keeping the agent within the provider's requests-per-second or tokens-per-minute quotas and preventing rate-limit violations before they occur. Together, C and E address the scenario's requirement to handle rate limits gracefully by combining prevention (throttling) with recovery (backoff retries). Option A is not appropriate because disabling caching increases call volume and cost without helping with rate limits. Option B is not appropriate because raising max_tokens does not reduce the number of calls and can actually consume more of the token quota per request. Option D is not appropriate because setting temperature to 0 only affects response determinism, not rate-limit handling.

Answer

Disable caching to ensure fresh responses

Answer

Increase the max_tokens parameter to reduce the number of calls

Answer

Set the model temperature to 0

Question 20

An AI engineer is using OCI Generative AI to build a retrieval-augmented generation (RAG) system. They notice that the model sometimes ignores the retrieved context and generates answers from its parametric knowledge. Which technique would most directly improve the model's adherence to the retrieved context?

Accepted Answer

Fine-tune the model on examples where the correct answer is derived solely from the provided context.. When a model ignores retrieved context, it is often because it has not been trained to prioritize that context. Fine-tuning with examples that explicitly require using the provided context can realign the model's behavior. This approach is more effective than adjusting inference parameters or context size, which do not change the model's underlying tendency to favor its pretrained knowledge.

Answer

Use a larger context window model and include more retrieved documents.

Answer

Reduce the number of retrieved documents to only the most relevant one.

Answer

Increase the temperature parameter to encourage more creative responses.