1Z0-1127-25 Practice Test 48 — 15 Questions

Question 1

A developer is iterating on a prompt for an OCI Generative AI model used to classify customer emails into five categories. The prompt currently includes a long narrative describing the company's history. The accuracy is inconsistent. Which change is the MOST direct improvement to prompt clarity?

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

Accepted Answer

Replace the narrative with a concise task instruction, the five category definitions, and one labelled example per category.. Classification quality hinges on unambiguous task instructions and discriminative category definitions backed by representative examples. Irrelevant narrative competes for attention and blurs the decision boundaries. Removing filler and replacing it with a crisp instruction, five definitions, and one labelled example per class gives the model the signal it needs to assign labels consistently.

Answer

Append the company history again at the end of the prompt so the model sees it twice.

Answer

Increase the top p value so the model samples from a broader set of category labels.

Answer

Add a persona statement that the model is a helpful assistant before the narrative.

Question 2

A developer is prototyping a text-generation feature and wants to compare several prompt phrasings and parameter settings quickly, without writing any code or provisioning infrastructure. Which OCI Generative AI capability should the developer use?

Accepted Answer

The OCI Generative AI Playground, which provides an interactive console for testing prompts and parameters against available models.. The OCI Generative AI Playground is the console-based, no-code environment for interactively testing prompts and parameters against available models. It requires no infrastructure provisioning and no scripting, so a developer can rapidly compare prompt phrasings and settings. Dedicated clusters, direct API scripting, and notebook sessions all add setup or coding overhead that the scenario explicitly rules out.

Answer

An OCI Data Science notebook session with the OCI Python SDK installed, used to call the model.

Answer

A dedicated AI cluster hosting a fine-tuned model, which the developer can query from the console.

Answer

The Generative AI inference REST API called from a local script using a config file with the user's API key.

Question 3

A solutions architect is building a customer-facing chatbot on OCI Generative AI. Legal requires that no prompt or completion content ever be stored or used to train any model, and the architect must prove this to auditors. Which OCI Generative AI capability should the architect enable and document?

Accepted Answer

The 'Do not store model data' option on the OCI Generative AI service. The service setting 'Do not store model data' is the only control that directly addresses retention and training use of prompt and completion content. Cluster isolation, private networking, and content filtering all harden the deployment but leave the data-handling policy unchanged. Because the requirement is explicitly about non-persistence and non-training, enabling and documenting that setting is the correct action.

Answer

Guardrails configured with a personally identifiable information (PII) filter

Answer

A private endpoint backed by an OCI private subnet and a network security group

Answer

Dedicated AI clusters with a customer-managed encryption key stored in OCI Vault

Question 4

A solutions architect in the eu-frankfurt-1 region is building an application that calls the OCI Generative AI chat endpoint for a Cohere Command R model. The application must use IAM credentials that are automatically rotated and are never stored as long-lived secret keys on the compute instances. Which authentication mechanism should the application use?

Accepted Answer

Instance principals. Instance principals let a compute instance obtain short-lived, platform-managed certificates from the instance metadata service, so the application can call the OCI Generative AI chat endpoint without storing or rotating static credentials. This meets both requirements: automatic rotation and no long-lived secret keys on the instance.

Answer

User API keys stored in the instance metadata

Answer

Resource principal with a dynamic group for the model endpoint

Answer

OCI Vault secret with a bearer token in the request header

Question 5

A team is evaluating an OCI Generative AI model for a document summarization task where the input documents frequently exceed the model's context window. They plan to use a retrieval-augmented generation (RAG) pipeline with an OCI Generative AI embedding model and a vector store. Which two statements correctly describe how chunking and embedding choices affect the quality of the generated summaries? (Choose two.)

Accepted Answer

Using smaller chunks with overlap helps preserve context across chunk boundaries and can improve retrieval relevance for summarization.. Effective RAG summarization depends on coherent chunking with overlap and consistent embedding models for indexing and querying. Overlap preserves cross-boundary context, and using the same embedding model ensures similarity scores are meaningful. These two choices directly improve the relevance of retrieved context, which in turn improves summary quality.

Answer

Increasing the embedding dimension always improves retrieval accuracy regardless of the domain or corpus size.

Answer

Storing chunk embeddings in a vector store eliminates the need to pass retrieved text to the LLM for summarization.

Answer

Chunk size should always be set to the model's maximum context window to minimize the number of chunks stored.

Question 6

A team is building a retrieval-augmented generation (RAG) assistant on OCI Generative AI. Users report that answers sometimes cite facts not present in the retrieved context, a behavior known as hallucination. Which TWO prompt engineering changes are MOST likely to reduce unsupported claims? (Choose two.)

Accepted Answer

Add a system instruction stating that the model must answer only from the provided context and must reply 'I do not know' when the context is insufficient.. Reducing hallucination in RAG requires two things: an explicit grounding instruction with a safe fallback, and context presented with delimiters plus a citation requirement. Together they constrain the model to the retrieved evidence and make gaps visible. Sampling changes, longer outputs, and removing context all increase the risk of unsupported claims.

Answer

Raise the temperature to 0.9 so the model explores more of the context before answering.

Answer

Remove the retrieved context and rely on the model's pretrained knowledge to answer the question.

Answer

Increase the model's max_tokens so it has more space to explain its reasoning before answering.

Question 7

A developer is calling the OCI Generative AI inference endpoint for a Cohere Command model and wants the model to produce deterministic, repeatable output for a regression test suite. Which parameter should they set to 0?

Accepted Answer

temperature. Deterministic output requires removing the stochastic element from token selection, and temperature is the parameter that scales the logits before sampling. At temperature 0 the distribution collapses toward the single highest-probability token, giving greedy decoding and near-identical responses across identical requests. Length caps, candidate-pool limits, and repetition penalties all modify the generation in other ways but leave the random draw in place.

Answer

topK

Answer

frequencyPenalty

Answer

maxTokens

Question 8

A developer is using the OCI Generative AI Playground to compare how a model behaves when the prompt is phrased in different ways. They want each request to be evaluated independently so that earlier prompts do not influence later ones. Which Playground behavior should they rely on?

Accepted Answer

Each submission starts a new conversation with no retained context. The Playground treats each prompt submission as a new, independent request unless the user deliberately continues a chat session. That default behavior gives the developer clean, unbiased comparisons across different phrasings without needing any special configuration.

Answer

The Playground automatically summarizes previous prompts before each request

Answer

The model caches the first response and reuses it for similar prompts

Answer

A 'Reset context' toggle must be disabled to prevent history from being sent

Question 9

An enterprise wants its OCI Generative AI endpoint to remain inside the Oracle Cloud network and never traverse the public internet, while still being reachable from applications running in the same VCN. Which endpoint type should they configure?

Accepted Answer

A private endpoint associated with the VCN. Keeping generative AI traffic off the public internet means the endpoint itself must be private and bound to the VCN. A private endpoint gives the service a private IP inside the customer's network, where security lists, network security groups, and IAM control access. Public endpoints, dedicated capacity clusters, and service gateways address other concerns such as reachability, performance, and routing to Oracle services.

Answer

A public inference endpoint with an API key

Answer

A service gateway attached to the on-premises network

Answer

A dedicated AI cluster endpoint

Question 10

A solutions architect is explaining the difference between fine-tuning and prompt engineering on OCI Generative AI to a customer. The customer wants to adapt a base model to consistently follow a proprietary output format for internal reports without changing the model weights. Which approach should the architect recommend?

Accepted Answer

Use prompt engineering with few-shot examples that demonstrate the desired output format.. When the goal is to enforce an output format without modifying model weights, prompt engineering with few-shot examples is the appropriate technique. It leverages in-context learning to demonstrate the desired structure, is quick to iterate, and avoids the cost and complexity of fine-tuning or retraining.

Answer

Deploy a custom model endpoint with a higher temperature to encourage the model to adopt the format.

Answer

Retrain the model from scratch using the customer's report corpus to ensure the format is learned.

Answer

Fine-tune the base model with a dataset of formatted reports to bake the format into the weights.

Question 11

A developer is building a summarization feature on OCI Generative AI that ingests long support tickets. They want the model to produce a fixed three-section summary: 'Issue', 'Root Cause', and 'Resolution'. The first attempts yield inconsistent headings and sometimes merge sections. Which prompting approach is BEST?

Accepted Answer

Provide a few-shot example showing the exact three headings with a short sample summary, then instruct the model to follow that template for the new ticket.. Enforcing a fixed three-section template is most reliably done with few-shot examples that show the exact headings and section content. The model imitates demonstrated structure, keeping headings consistent and preventing merged sections. Vague instructions, sampling tweaks, and unspecified JSON requests all leave the format underdetermined.

Answer

Append the instruction 'Return JSON' to the prompt and parse the sections from the JSON keys.

Answer

Ask the model to 'summarize the ticket clearly and organize the output nicely' without specifying headings.

Answer

Increase top_k so the model has more candidate tokens for each heading word.

Question 12

A data scientist is preparing a 10,000-document corpus to fine-tune a base LLM hosted in OCI Generative AI. They want the model to reliably answer questions about these proprietary documents without the model memorizing the entire corpus verbatim. Which training approach should they use?

Accepted Answer

Supervised fine-tuning (SFT) on question-answer pairs derived from the corpus. Supervised fine-tuning on curated question-answer pairs derived from the corpus is the correct training approach because it directly teaches the model to produce the desired answers from domain content without memorizing documents verbatim. Continued pre-training on raw text does not teach question-answering behavior, RAG does not modify weights, and RLHF requires preference data the scenario does not provide.

Answer

Continued pre-training on the raw, unlabeled corpus without any instruction formatting

Answer

Reinforcement learning from human feedback (RLHF) using only the document corpus as reward signal

Answer

Retrieval-augmented generation (RAG) with an embedding model and a vector store

Question 13

A prompt engineer is building a long-context question-answering assistant on OCI Generative AI using a model with a large context window. Users paste entire policy documents and ask specific questions. The engineer wants to maximize the chance that the model finds and uses the relevant passage. Which TWO prompt engineering practices are MOST effective? (Choose two.)

Accepted Answer

Instruct the model to quote the exact sentence from the document that supports its answer before giving the final response.. Long-context question answering benefits most from two practices: positioning the question after the document so it is recent in the model's attention, and requiring the model to quote supporting text so its answer is anchored to the source. Together these reduce hallucination and make answers verifiable, which is critical when users rely on policy documents for decisions.

Answer

Repeat the entire document twice in the prompt to reinforce the content.

Answer

Set the stop sequence to the end-of-document marker so the model stops reading the document.

Answer

Increase the temperature to 0.9 so the model considers a wider range of possible answers.

Question 14

A solutions architect is deploying a customer-facing application that uses OCI Generative AI on-demand models. The application experiences highly variable traffic, with occasional bursts of concurrent requests. The architect needs to ensure that the application handles throttling gracefully and does not fail during peak load. Which approach should the architect take?

Accepted Answer

Implement exponential backoff and retry logic in the application code to handle 429 Too Many Requests responses.. The service returns HTTP 429 when rate limits are exceeded. Exponential backoff with jitter is the standard client-side pattern to handle throttling: it retries after increasing delays, reduces contention, and prevents synchronized retries. Fixed intervals can worsen throttling, max tokens affects response length not rate, and dedicated clusters still have limits.

Answer

Configure a fixed retry interval of 1 second for all failed requests to ensure rapid recovery.

Answer

Increase the model's max tokens parameter to allow more requests per minute.

Answer

Deploy the model on a dedicated AI cluster to eliminate rate limits entirely.

Question 15

A solution architect is choosing an open model on OCI Generative AI to power a long-document analysis pipeline where each input can exceed 100,000 tokens. Which two factors most directly determine whether a given model can handle these inputs? (Choose two.)

Accepted Answer

The model's maximum context length (context window). The maximum context length sets the hard token limit per request, and the positional encoding scheme determines how well the model generalizes to long sequences. Parameter count, vocabulary size, and temperature do not directly govern the ability to process 100,000-token inputs, so they are not the primary factors.

Answer

The number of parameters in the model

Answer

The default temperature setting exposed in the OCI console

Answer

The tokenizer's vocabulary size