1Z0-1127-25 Practice Test 43 — 15 Questions

Question 1

An enterprise wants to deploy a large language model for processing sensitive internal documents. They must ensure that data does not leave their OCI tenancy. Which OCI GenAI deployment option meets this requirement?

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

Accepted Answer

Dedicated AI Cluster with on-demand model hosting. A Dedicated AI Cluster with on-demand model hosting provides a fully isolated, single-tenant infrastructure within your OCI tenancy. This ensures that all data processing and model inference occur on dedicated hardware that never leaves your tenancy, meeting strict data residency and security requirements for sensitive internal documents.

Answer

Using a third-party model via OCI Marketplace

Answer

Accessing models through OCI Console only

Answer

Using the OCI GenAI API with the default endpoint

Question 2

An organization needs to deploy a model that can both understand and generate text, such as for a translation task where the input is in English and output is in French. Which model architecture is most suitable?

Accepted Answer

Encoder-decoder (e.g., T5). Encoder-decoder architectures like T5 are designed for sequence-to-sequence tasks. The encoder processes the input sequence, and the decoder generates the output sequence, making it ideal for translation.

Answer

Decoder-only (e.g., GPT)

Answer

Encoder-only (e.g., BERT)

Answer

Mixture of Experts (MoE)

Question 3

An OCI customer is deploying a chatbot using a pre-trained LLM. They are concerned about the model generating biased or harmful content. Which TWO strategies should they implement as part of their responsible AI approach? (Choose two.)

Accepted Answer

Set up a human-in-the-loop review for sensitive queries. Option C is correct because a human-in-the-loop review process routes sensitive or high-risk queries to human reviewers who can validate, correct, or escalate the model's responses, which directly mitigates the risk of biased or harmful content reaching users. Option E is correct because a content filtering layer (for example, using OCI Language moderation or a guardrails service) inspects prompts and completions and blocks or redacts harmful outputs before they are returned, providing a deterministic safety control independent of the model's behavior. Option A is not appropriate here because the scenario uses a pre-trained LLM and retraining from scratch is costly, slow, and unnecessary for a responsible-AI guardrail. Option B is incorrect because enlarging the context window only gives the model more input tokens and does not by itself reduce bias or harmful generation. Option D is incorrect because top-k sampling with k=1 is greedy decoding, which affects determinism and diversity but provides no safety or bias mitigation.

Answer

Train the model from scratch on a curated dataset

Answer

Increase the context window to include more examples

Answer

Use top-k sampling with k=1

Question 4

A healthcare company is using OCI Generative AI to analyze patient records and generate clinical summaries. The company must comply with HIPAA regulations, which require that all protected health information (PHI) be encrypted at rest and in transit, and that access be logged and audited. The current architecture uses an OCI Data Science model deployment with a public endpoint. The model is stored in an OCI Object Storage bucket that is publicly accessible for testing. The company is now moving to production. The compliance officer has flagged the following issues: (1) The model endpoint is publicly accessible. (2) The bucket containing the model is public. (3) No audit logs are enabled. The company wants to remediate these issues while maintaining the ability to invoke the model from on-premises applications via a secure connection. Which set of actions should the architect take?

Accepted Answer

Switch the model endpoint to a private subnet with a service gateway, update the bucket policy to block all public access, enable OCI Audit service, and set up a VPN or FastConnect for on-premises access.. It addresses all three compliance issues: moving the model endpoint to a private subnet with a service gateway removes public exposure, making the bucket private with a policy that blocks all public access secures the model artifacts, and enabling OCI Audit provides the required logging. Additionally, setting up a VPN or FastConnect allows secure on-premises access without exposing the endpoint to the public internet, fully satisfying HIPAA encryption and audit requirements.

Answer

Switch the model endpoint to a private subnet with a service gateway, change the bucket to be accessible only via pre-authenticated requests, and enable OCI Logging for the model deployment.

Answer

Keep the public endpoint but restrict access using IAM policies and source IP addresses, make the bucket private, and enable OCI Audit.

Answer

Use a public load balancer with SSL termination, restrict bucket access to the load balancer's OCID, and enable OCI Audit.

Question 5

Your team is deploying a generative AI model for a clinical decision support system. The model must meet HIPAA compliance requirements. You have trained a model using OCI Data Science and now need to deploy it so that patient data is protected. The application requires real-time inference. Which set of actions should you take to ensure compliance while maintaining low latency?

Accepted Answer

Deploy in a private VCN subnet, use a service gateway, store keys in OCI Vault, and enable OCI Logging and OCI Audit. Deploying the model in a private VCN subnet ensures the inference endpoint is not exposed to the internet, meeting HIPAA's requirement for network isolation. Using a service gateway allows private connectivity to OCI services without traversing the internet, while storing encryption keys in OCI Vault enables customer-managed key control for data at rest. Enabling OCI Logging and OCI Audit provides the necessary audit trail for compliance, and the private subnet with service gateway keeps latency low by avoiding internet hops.

Answer

Use OCI Functions with API Gateway and allow anonymous access

Answer

Deploy in a public subnet with HTTPS and enable OCI Audit

Answer

Use OCI Data Flow for batch inference and store results in Object Storage with SSE

Question 6

A developer is using the OCI Generative AI Chat API to build a multi-turn conversational agent. They want the model to remember previous exchanges within the same session. How should they manage conversation history?

Accepted Answer

Use the Chat API's message list parameter, appending each user and assistant message. The OCI Generative AI Chat API is designed for multi-turn conversations and accepts a `messages` array (or equivalent message list parameter) where each entry has a role (user, assistant, system) and content. To maintain context, the developer appends each new user message and the model's assistant response to this list and sends the full list on every subsequent call. This preserves the conversation state across turns without relying on any server-side session memory, which the API does not provide.

Answer

Store history client-side and concatenate all previous outputs into the user prompt each time

Answer

Use the Generate API with a system prompt that includes the entire history

Answer

Rely on the model's built-in memory mechanism

Question 7

A company is deploying a LangChain application using OCI Generative AI. They need to comply with a policy that requires all prompts sent to the LLM to be logged for audit, and they must also handle rate limits gracefully. Which TWO strategies should they implement?

Accepted Answer

Implement a custom LangChain callback that logs the prompt before sending it to the model. Option B is correct because a custom LangChain callback handler (e.g., implementing on_llm_start or on_llm_end) is the standard mechanism to intercept and log the exact prompt payload before it is dispatched to the OCI Generative AI endpoint, satisfying the audit requirement. Option D is correct because OCI Generative AI returns HTTP 429 (TooManyRequests) when throttling limits are exceeded, and wrapping the invocation in a retry with exponential backoff (e.g., via LangChain's max_retries or a retry decorator) lets the application recover gracefully instead of failing. Option A is not correct because choosing a faster model addresses latency, not audit logging or rate-limit handling. Option C is not correct because batching more requests together increases the chance of hitting rate limits and does nothing for prompt logging. Option E is not correct because stuffing conversation history into the system message neither logs prompts for audit nor mitigates rate-limit errors.

Answer

Use a faster LLM to reduce response time

Answer

Increase the batch size of requests to reduce the number of API calls

Answer

Store the full conversation history in the prompt's system message

Question 8

Which TWO are benefits of using few-shot prompting compared to zero-shot prompting?

Accepted Answer

It helps the model understand the desired pattern, especially for uncommon tasks. Option D is correct because few-shot prompting supplies several input-output examples that demonstrate the target pattern, which is especially valuable for uncommon or niche tasks where the model has little relevant pretraining signal and would otherwise guess in zero-shot mode. Option E is correct because those same examples can encode a precise output format (e.g., JSON keys, label sets, or field ordering), letting the model imitate the demonstrated structure and thereby improving format compliance on tasks with strict output requirements. Options A, B, and C are not benefits of few-shot prompting: it does not always remove the need for parameter tuning, it does not eliminate the need for a system prompt, and adding examples typically increases rather than reduces the number of tokens in the prompt and does not guarantee shorter outputs.

Answer

It always reduces the need for parameter tuning

Answer

It eliminates the need for a system prompt

Answer

It reduces the number of tokens in the output

Question 9

A data scientist is designing a RAG pipeline using LangChain and Oracle AI Vector Search. They want to ensure that the retrieved documents are diverse and not overly similar to each other. Which TWO approaches can achieve this?

Accepted Answer

Set fetch_k larger than k in the retriever's search_kwargs and use MMR. Option A is correct because setting fetch_k larger than k in search_kwargs gives MMR (Maximal Marginal Relevance) a larger candidate pool to draw from before it selects the final k results, and MMR explicitly balances relevance against redundancy so the returned documents are diverse rather than near-duplicates. Option D is correct because calling as_retriever with search_type='mmr' switches the retriever from default similarity search to Maximal Marginal Relevance, which is the LangChain mechanism that penalizes similarity among selected documents and thus promotes diversity in the retrieved set. Option B is not correct because smaller chunk_size only changes granularity of chunks and does not itself enforce diversity among retrieved results. Option C is not correct because higher chunk_overlap deliberately increases redundancy between chunks, which works against the goal of diversity. Option E is not correct because similarity_search with a high k simply returns more of the most similar documents, which tends to increase redundancy rather than reduce it.

Answer

Use a smaller chunk_size to create more granular chunks

Answer

Use a higher chunk_overlap to increase redundancy

Answer

Use similarity_search with a high k value

Question 10

A financial services company must deploy a fine-tuned model for transaction categorization. The model must be isolated from other tenants and provide predictable low-latency inference. The compliance team also requires that training data never leaves the OCI tenancy. Which THREE steps should the team take? (Choose three.)

Accepted Answer

Fine-tune the model using T-Few technique within OCI. Option A is correct because T-Few (few-shot fine-tuning) is performed inside OCI Generative AI, so the training data stays within the tenancy and never leaves it, satisfying the compliance requirement. Option D is correct because a dedicated AI cluster with model units provides the isolated, tenant-specific compute capacity needed for the fine-tuned model. Option C is correct because deploying the fine-tuned model on that dedicated cluster after fine-tuning delivers the isolation and predictable low-latency inference the company requires. Option B is not appropriate because on-demand inference uses shared, multi-tenant endpoints rather than the dedicated isolated deployment required here. Option E is incorrect because a shared AI cluster does not provide tenant isolation and offers less predictable latency.

Answer

Use OCI GenAI on-demand inference for the fine-tuned model

Answer

Host the model on a shared AI cluster to reduce cost

Question 11

A prompt engineer is troubleshooting a chatbot that consistently fails to follow instructions when the user includes adversarial input. Which two strategies can mitigate prompt injection attacks? (Choose two.)

Accepted Answer

Use instruction shielding: clearly separate system instructions from user input. Option B is correct because instruction shielding (e.g., delimiting system prompts with tags like ... or using role separation) helps the model distinguish trusted developer instructions from untrusted user input, making it harder for injected text to override the original directives. Option E is correct because input validation and sanitization can detect and strip or neutralize known adversarial patterns (e.g., phrases like "ignore previous instructions" or encoded payloads) before they reach the model, reducing the attack surface. Option A is incorrect because raising temperature increases randomness and does not prevent injection; it can actually make behavior less reliable. Option C is incorrect because using a smaller model does not inherently mitigate prompt injection and may reduce instruction-following ability. Option D is incorrect because adding few-shot examples with safe outputs does not reliably block adversarial inputs and may even be exploited if examples are not carefully isolated.

Answer

Increase temperature to make model less predictable

Answer

Use a smaller model to reduce capability

Answer

Add more few-shot examples with safe outputs

Question 12

A prompt engineer is iteratively refining a prompt for a summarization task. Which THREE practices should she adopt to improve prompt performance systematically?

Accepted Answer

Test prompt with a diverse set of input documents. Option A is correct because testing the prompt against a diverse set of input documents exposes it to varied lengths, domains, and edge cases, revealing weaknesses that a narrow sample would hide and enabling systematic generalization. Option B is correct because A/B testing prompt variants on a held-out set provides an unbiased, quantitative comparison of candidate prompts, so improvements are measured rather than assumed. Option E is correct because establishing clear evaluation criteria such as accuracy and conciseness gives an objective, repeatable rubric against which each iteration can be scored, which is the foundation of systematic refinement. Option C is not appropriate because reusing the same input guarantees consistency but prevents discovering failures on other inputs, defeating iterative improvement. Option D is not appropriate because raising max tokens to the maximum merely allows longer outputs; it does not improve summarization quality and can increase cost, latency, and verbosity without any evaluation-driven benefit.

Answer

Always use the same input to ensure consistency

Answer

Increase max tokens to the maximum allowed

Question 13

A team deploys a generative AI model endpoint and notices intermittent 429 Too Many Requests errors. The endpoint is configured with auto-scaling using a dedicated AI cluster. What is the most likely cause?

Accepted Answer

The auto-scaling policy is not aggressive enough. The 429 Too Many Requests error indicates that the endpoint is receiving more requests than it can handle. With auto-scaling enabled on a dedicated AI cluster, the most likely cause is that the auto-scaling policy is not aggressive enough to keep up with the request rate, meaning it scales up too slowly or has insufficient maximum instance limits to handle the traffic spike.

Answer

The model's context window exceeded

Answer

Insufficient storage on the cluster

Answer

Rate limiting at the OCI API Gateway

Question 14

A data scientist is building a RAG pipeline on OCI. Which TWO components are essential for the retrieval step?

Accepted Answer

Embedding model to convert chunks into vectors. Option B (Embedding model to convert chunks into vectors) is correct because the retrieval step of a RAG pipeline requires encoding both the document chunks and the user query into dense vector representations so that semantic similarity search can be performed against a vector store. Option C (Document chunking) is correct because source documents must be split into manageable chunks before they can be embedded and indexed; chunking determines the granularity of the retrievable units and directly affects retrieval quality. Option A (Fine-tuned generation model) is not essential for retrieval since the generation model operates after retrieval to synthesize the answer, and fine-tuning is not required for the retrieval step. Option D (Beam search decoder) is a decoding strategy used during text generation, not during retrieval. Option E (Human feedback loop) is an optional evaluation/improvement mechanism and is not a required component of the retrieval step.

Answer

Fine-tuned generation model

Answer

Beam search decoder

Answer

Human feedback loop

Question 15

Which OCI Generative AI service component is designed to convert text into vector representations for use in semantic search?

Accepted Answer

Embedding API. The Embedding API generates vector embeddings from text, which can be used for semantic search, clustering, etc.

Answer

Chat API

Answer

Summarisation API

Answer

Generate API