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

Question 1

Which of the following is a known limitation of large language models that Retrieval-Augmented Generation (RAG) aims to address?

Accepted Answer

Hallucination of facts. RAG directly addresses the hallucination problem by grounding LLM outputs in retrieved, factual documents from an external knowledge base. Instead of relying solely on the model's parametric memory, RAG fetches relevant context at inference time, reducing the likelihood of generating plausible-sounding but incorrect facts.

Answer

High cost of training

Answer

Slow inference speed

Answer

Lack of multilingual support

Question 2

Which TWO are common causes of poor answer quality in a RAG system built on OCI Generative AI? (Choose two.)

Accepted Answer

Mismatch between the embedding model's training data and the domain of the documents.. Option A is correct because if the embedding model was trained on data that differs from the domain of the indexed documents, the resulting vectors will not capture the domain-specific semantics, so retrieval will surface chunks that are semantically distant from the user's query and the generator will produce poor answers. Option D is correct because RAG quality depends on the retriever returning enough relevant context; if the corpus lacks a sufficient number of relevant chunks for the query, the generation model has no grounding evidence and will hallucinate or give incomplete answers. Option B is not a common cause of poor RAG answer quality, since a larger generation model generally improves quality rather than degrading it, and model size is not the retrieval bottleneck. Option C is not correct because a low temperature makes outputs more deterministic, which does not by itself cause poor answer quality in RAG. Option E is not correct because vector search alone is a valid and common retrieval approach; adding keyword fallback is an optional hybrid enhancement, not a standard cause of poor answer quality.

Answer

Using a generation model that is too large for the task.

Answer

Setting the temperature parameter too low, causing overly deterministic outputs.

Answer

Using only vector search without keyword-based fallback.

Question 3

A company is using OCI Generative AI service with a dedicated AI cluster for text generation. They notice that the latency is higher than expected. The cluster is in the Ashburn region, and users are distributed globally. What is the most effective way to reduce latency?

Accepted Answer

Deploy dedicated AI clusters in regions closer to the users. Latency for globally distributed users is primarily driven by network distance and the speed of light. Deploying dedicated AI clusters in regions closer to the users reduces the physical distance data must travel, directly minimizing network round-trip time (RTT). This is the most effective architectural change because OCI's Generative AI service processes each request on the dedicated cluster and cannot bypass geographic latency through software optimizations alone.

Answer

Enable the OCI Generative AI inference optimizer

Answer

Increase the number of nodes in the dedicated AI cluster

Answer

Use a content delivery network (CDN) to cache responses

Question 4

A developer uses the OCI Generative AI Chat API to build a multi-turn conversational agent. They notice the model starts to lose context after several exchanges. What is the MOST likely cause?

Accepted Answer

The developer is not sending the conversation history with each request. The OCI Generative AI Chat API is stateless — each request is processed independently, and the model has no memory of prior turns unless the caller resends the full conversation history in the messages array. If the developer only sends the latest user message, the model loses all prior context, which is exactly the symptom described. The fix is to maintain a client-side conversation buffer and include all prior user/assistant turns in every request.

Answer

The temperature parameter is set too high, causing the model to forget

Answer

The model's context window is too small for the conversation length

Answer

The fine-tuning dataset did not include multi-turn examples

Question 5

A healthcare startup is building a chatbot to answer patient inquiries using a large language model (LLM) deployed on OCI Data Science AI Quick Actions. The chatbot must comply with HIPAA regulations, so all patient data must remain within the OCI tenancy and never be sent to third-party APIs. The team has fine-tuned a Llama 2 7B model on de-identified medical records using OCI Data Science notebooks. The model is deployed as a managed endpoint via AI Quick Actions. Early testing shows that the chatbot sometimes generates responses containing specific patient names or dates of birth that were present in the fine-tuning dataset. Moreover, the model occasionally hallucinates medication dosages that are not medically accurate. Which course of action should the team take to address both issues while maintaining HIPAA compliance?

Accepted Answer

Re-fine-tune the model using differential privacy to limit memorization of training data, and implement retrieval-augmented generation (RAG) with a curated medical knowledge base to ground medication-related responses.. It addresses both memorization of PII and hallucination of medication dosages while maintaining HIPAA compliance. Differential privacy during fine-tuning limits the model's ability to memorize specific patient data, and retrieval-augmented generation (RAG) grounds responses in a curated medical knowledge base, reducing hallucinations without sending data outside the OCI tenancy.

Answer

Deploy a rule-based post-processing script that checks each response against a list of known patient names and medication dosages, and rejects any response containing them.

Answer

Switch to a larger model (e.g., Llama 2 70B) to improve accuracy and reduce hallucinations, and apply output filtering to remove any detected PII from responses.

Answer

Increase the fine-tuning dataset size with more varied de-identified records to reduce overfitting, and apply a temperature setting of 0 to make outputs deterministic.

Question 6

A RAG system returns irrelevant chunks even though the embedding model and vector index are correctly configured. After reviewing, the chunks are too large and contain extraneous information. Which combination of adjustments should be made to improve relevance?

Accepted Answer

Reduce chunk size, increase overlap, and adjust topK.. Reducing chunk size removes extraneous information, increasing overlap ensures context continuity across smaller chunks, and adjusting topK limits the number of retrieved chunks to the most relevant ones. This combination directly addresses the problem of large chunks containing irrelevant data while maintaining retrieval precision.

Answer

Increase chunk overlap only.

Answer

Decrease chunk size and increase chunk overlap.

Answer

Use semantic chunking and adjust topK.

Question 7

A developer sends the above request to the OCI Generative AI API. The response returns an error: 'InvalidParameter: The parameter 'topP' is not supported for this model.' What is the most likely reason?

Accepted Answer

The 'cohere.command-r-plus-v1:0' model does not accept the topP parameter.. The error 'InvalidParameter: The parameter 'topP' is not supported for this model' indicates that the model specified in the request, 'cohere.command-r-plus-v1:0', does not accept the 'topP' parameter. This is because the Cohere Command-R+ model uses a different sampling strategy (e.g., temperature and top-k) and does not expose a 'topP' parameter in the OCI Generative AI API. The API validates parameters against the model's capabilities, and unsupported parameters trigger this specific error.

Answer

The model ID is deprecated.

Answer

The topP parameter value is out of range.

Answer

The JSON request format is incorrect.

Question 8

Which THREE of the following are known limitations of large language models that practitioners must consider?

Accepted Answer

Hallucination of facts not present in the input.. Option A is correct because large language models are probabilistic next-token predictors and can generate fluent but fabricated content (hallucinations) that is not grounded in the provided input or any verified source. Option B is correct because models trained on unfiltered web-scale corpora can reproduce toxic, hateful, or harmful language, which is why content moderation and safety alignment are required in production systems. Option D is correct because LLMs learn statistical patterns from their training data, so societal biases present in that data can be amplified or reproduced in outputs, requiring bias evaluation and mitigation. Option C is incorrect because modern LLMs are typically multilingual and can process and generate text in many languages, sometimes even within a single prompt. Option E is incorrect because although context windows are finite, current models support thousands to hundreds of thousands of tokens, so the claim that they cannot process more than a few hundred tokens is technically inaccurate.

Answer

Limited to processing only one language at a time.

Answer

Inability to process inputs longer than a few hundred tokens.

Question 9

Which THREE are valid considerations when designing a RAG pipeline that uses OCI Generative AI and OCI OpenSearch? (Choose three.)

Accepted Answer

The quality of the text extraction from OCI Document Understanding directly impacts retrieval accuracy.. Option C is correct because OCI Document Understanding performs the text extraction (OCR/layout parsing) that feeds the embedding step, so errors, missing text, or garbled tables directly degrade the embeddings and therefore retrieval accuracy. Option D is correct because the retrieved chunks are concatenated into the prompt, and the generation model's context window (token limit) caps how many chunks plus the user query can be included, forcing trade-offs in top-k and chunk size. Option E is correct because chunk size and overlap determine semantic granularity and context continuity; they must be tuned to the document structure (e.g., dense technical docs vs. short FAQs) and to query patterns to balance recall and precision. Option A is incorrect because OCI OpenSearch supports multiple vector similarity metrics such as cosine similarity, dot product, and Euclidean distance, not only Euclidean. Option B is incorrect because documents are typically split into multiple chunks, each embedded as its own vector, to improve retrieval precision rather than being reduced to a single vector per document.

Answer

OCI OpenSearch only supports Euclidean distance for vector similarity.

Answer

Each document must be converted to a single vector for efficient retrieval.

Question 10

A DevOps engineer is setting up monitoring and logging for a generative AI inference endpoint. Which three resources should they enable? (Select THREE.)

Accepted Answer

OCI Logging for inference requests and responses. Option B is correct because OCI Logging captures the inference request and response payloads, which is essential for debugging model behavior, auditing prompts/completions, and troubleshooting generative AI endpoint issues. Option C is correct because OCI Monitoring provides time-series metrics such as endpoint latency and error rates, enabling alarms, dashboards, and SLA tracking for the inference service. Option D is correct because OCI Application Performance Monitoring (APM) provides distributed tracing of inference requests across the application stack, helping pinpoint latency bottlenecks and failures in the request path. Option A is not correct because VCN flow logs only capture network-level metadata (source/destination IP, ports, protocol, accept/reject) and do not provide application-level inference visibility. Option E is not correct because OCI Audit logs record control-plane API calls (e.g., create/update/delete of resources) rather than the data-plane inference requests and responses the scenario requires.

Answer

OCI VCN flow logs for network traffic

Answer

OCI Audit logs for all API calls

Question 11

A developer is building a RAG application using OCI Generative AI Agents. They want to ensure the agent only retrieves information from approved documents in a specific compartment. Which THREE steps are required?

Accepted Answer

Create a knowledge base that references the Object Storage bucket. Option B is correct because an OCI Generative AI Agents knowledge base is the component that ingests and indexes the approved documents, and it must be created with a data source pointing to the Object Storage bucket holding those documents. Option E is correct because the approved documents must first reside in an Object Storage bucket within the target compartment, which serves as the data source for the knowledge base. Option D is correct because the agent's service principal needs an IAM policy granting read access to the objects in that specific compartment; without it, ingestion and retrieval will fail with authorization errors. Option A is not required because a dedicated AI cluster is for hosting fine-tuned or dedicated model endpoints, not for building a RAG knowledge base. Option C is not required because OCI Generative AI Agents automatically handles embedding generation and vector storage when the knowledge base is created, so manual Embedding API calls are unnecessary.

Answer

Provision a dedicated AI cluster in the same compartment

Answer

Use the Embedding API to manually generate embeddings for each document

Question 12

What is the PRIMARY purpose of the 'stop sequences' parameter in text generation?

Accepted Answer

To define sequences that, when generated, cause the model to cease generation. Stop sequences tell the model when to stop generating further tokens, allowing control over output length or termination upon specific strings.

Answer

To prevent the model from generating offensive content

Answer

To specify tokens that the model should avoid using

Answer

To encourage the model to include specific phrases in the output

Question 13

Which TWO are best practices for chunking documents in a RAG pipeline? (Choose two.)

Accepted Answer

Use semantic chunking to preserve meaning.. Option D is correct because semantic chunking splits text along meaning-preserving boundaries (e.g., paragraphs, sections, or embedding-similarity breakpoints) so each retrieved chunk remains coherent and self-contained, which improves retrieval relevance and answer grounding in a RAG pipeline. Option E is correct because overlapping adjacent chunks (typically 10–20% of chunk size, e.g., 50–100 tokens of overlap) ensures that context spanning a boundary is not lost, so a fact split across two chunks still appears intact in at least one retrieved chunk. Option A is not a best practice because fixed-size chunks ignore content boundaries and can cut sentences, tables, or code mid-thought, degrading embedding quality and retrieval. Option B is not a best practice because 100-character chunks are far too small to carry meaningful context, producing fragmented embeddings and poor answers. Option C is not a best practice because embedding an entire document as one chunk dilutes relevance, exceeds typical context windows, and prevents precise retrieval of the specific passage needed.

Answer

Use fixed-size chunks regardless of content boundaries.

Answer

Always use small chunks (e.g., 100 characters).

Answer

Chunk entire documents as a single chunk.

Question 14

A team is building a code generation assistant using OCI Generative AI. They notice that the model occasionally produces code with subtle security vulnerabilities. Which approach would most effectively reduce this risk without compromising the assistant's usefulness?

Accepted Answer

Fine-tune the model on a dataset of secure code examples and security best practices. Fine-tuning the model on a dataset of secure code examples and security best practices directly teaches the model to generate more secure code by adjusting its weights to favor secure patterns. This approach reduces the likelihood of subtle vulnerabilities without sacrificing the assistant's ability to generate useful code, as it learns from curated examples. Other options like larger context window or more parameters do not specifically address security, and greedy decoding only reduces randomness, not inherent biases.

Answer

Use a larger context window to include all project files in every prompt

Answer

Switch to a model with more parameters

Answer

Use greedy decoding to reduce randomness in code generation

Question 15

A developer notices that an LLM's responses are too verbose. Which parameter adjustment would most effectively reduce verbosity?

Accepted Answer

Decrease max_tokens. Decreasing max_tokens directly limits the maximum length of the LLM's response, which is the most straightforward way to reduce verbosity. This parameter caps the number of tokens the model can generate, forcing it to produce shorter completions. Other parameters like frequency_penalty, top_p, and temperature influence the style, diversity, or randomness of the output but do not directly control response length.

Answer

Increase frequency_penalty

Answer

Increase top_p

Answer

Decrease temperature