1Z0-1127-25 Practice Exam 54 — 20 Questions

Question 1

A developer is building a customer support assistant on OCI Generative AI using a Meta Llama model. They want the assistant to always answer in the persona of a calm, empathetic agent and to never discuss competitor products. They notice that adding these rules to the user message works inconsistently across sessions. What should they do to make the behavior reliable?

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

Accepted Answer

Move the persona and competitor-exclusion rules into the system prompt so they apply to every turn of the conversation.. Persistent behavioral rules such as persona, tone, and topic exclusions belong in the system prompt, which the model treats as standing instructions across all turns. User messages are task-level input and can be overridden or diluted. Sampling parameters and stop sequences affect randomness and truncation, not policy adherence.

Answer

Repeat the rules at the end of every user message to reinforce them.

Answer

Add the rules to a stop sequence list so the model halts before mentioning competitors.

Answer

Increase the temperature so the model has more freedom to express empathy.

Question 2

A team is planning to use OCI Generative AI to build a conversational agent that must maintain context over multiple turns. They want to ensure the agent remembers previous user inputs. Which two strategies are most effective for maintaining conversation context? (Choose two.)

Accepted Answer

Store conversation history in an external database and retrieve relevant past messages to include in the prompt.. To maintain conversation context, the model must receive the relevant history in each prompt. The two primary methods are including the full history (within token limits) or using an external store to retrieve and inject relevant past messages. Both ensure the model has the necessary information to generate context-aware responses. Other parameters like temperature or max_tokens do not provide memory.

Answer

Set a higher max_tokens limit to allow the model to generate longer responses that reference past turns.

Answer

Use a fine-tuned model that has been trained on similar conversations.

Answer

Increase the temperature parameter to encourage the model to recall previous messages.

Question 3

A prompt engineer is building a chain-of-thought prompt for a multi-step arithmetic word problem on OCI Generative AI. The model sometimes skips steps and produces a wrong final number. The engineer wants the model to reason step by step and then state a final answer. Which prompt design is MOST effective?

Accepted Answer

Instruct the model to reason step by step, then clearly separate the reasoning from the final answer with a delimiter.. Chain-of-thought prompting works by eliciting intermediate reasoning steps, which raises accuracy on multi-step arithmetic. Separating that reasoning from the final answer with a delimiter keeps the output parseable and ensures the numeric result is explicit. Suppressing steps, raising randomness, or truncating early all remove the mechanisms that make step-by-step reasoning effective.

Answer

Ask the model to output only the final numeric answer with no intermediate steps.

Answer

Add a stop sequence that halts generation as soon as the model emits the first digit.

Answer

Set temperature to the maximum so the model can try many arithmetic paths and pick the best one.

Question 4

A developer is prototyping an OCI Generative AI chatbot that must always respond in valid JSON with keys 'intent' and 'confidence'. The model sometimes wraps the JSON in markdown code fences or adds a conversational preamble. Which prompt engineering technique MOST directly prevents this formatting drift?

Accepted Answer

Provide a strict output schema instruction, include a one-shot example of the exact JSON with no preamble or code fences, and set temperature low.. A strict schema instruction paired with a one-shot example of the exact desired output and a low temperature setting is the most direct way to eliminate markdown fences and preamble text. The example demonstrates the format, the instruction forbids deviations, and low temperature reduces the randomness that produces formatting variance.

Answer

Set the stop sequence to '```' so the model stops before emitting a code fence.

Answer

Increase the model's context window so the full schema description fits in the prompt.

Answer

Provide a single clear instruction such as 'Please return JSON.'

Question 5

A developer is preparing a custom fine-tuning job in OCI Generative AI. They want the resulting model to be served from a dedicated, private endpoint inside their tenancy, and they want to control the endpoint's shape and scaling behavior. Which OCI Generative AI capability should they use to host the fine-tuned model?

Accepted Answer

Create a dedicated AI cluster in OCI Generative AI and host the fine-tuned model on a dedicated endpoint. In OCI Generative AI, dedicated AI clusters provide tenancy-scoped compute for hosting models on dedicated endpoints, including fine-tuned models. Because the developer needs a private endpoint with control over hosting shape and scaling, hosting the fine-tuned model on a dedicated endpoint within a dedicated AI cluster is the appropriate capability. Other options describe storage or experimentation tools that do not deliver dedicated model serving.

Answer

Publish the fine-tuned model to OCI Object Storage and call it through a pre-authenticated request

Answer

Use the OCI Generative AI Playground to interact with the fine-tuned model

Answer

Import the fine-tuned model into OCI Data Science and deploy it as a model deployment

Question 6

A machine learning engineer is configuring inference for a text generation model on OCI Generative AI. They need the model to consistently select the most probable next token at each step to produce deterministic outputs for a testing pipeline. Which decoding strategy should they use?

Accepted Answer

Greedy decoding. Greedy decoding always picks the highest-probability token, yielding deterministic outputs. For a testing pipeline that requires reproducibility, this is the correct choice. Sampling methods like top-k and top-p introduce randomness, and beam search, while more deterministic than sampling, can still vary due to tie-breaking and is not guaranteed to be identical across runs.

Answer

Top-k sampling with k=50

Answer

Beam search with beam width 5

Answer

Top-p sampling with p=0.9

Question 7

You are designing a RAG pipeline on OCI Generative AI where documents are updated frequently. Users must see the latest content immediately after an update. The current batch re-indexing runs nightly, causing stale answers during the day. Which design change best meets the freshness requirement?

Accepted Answer

Implement incremental indexing that embeds and upserts changed documents in near real time.. Real-time freshness requires an event-driven, incremental indexing pipeline that embeds and upserts documents as they change. Batch re-indexing, chunk resizing, and scheduled cache invalidation all introduce delays or fail to update the vector store promptly. Incremental indexing minimizes the window between an update and its availability in retrieval, directly satisfying the requirement.

Answer

Reduce the chunk size so that updates affect fewer chunks and propagate faster.

Answer

Add a cache layer that stores LLM responses and invalidates it on a schedule.

Answer

Increase the batch re-indexing frequency to every hour.

Question 8

A support team using OCI Generative AI with a Meta Llama chat model finds that short user questions like 'How do I reset my password?' produce vague answers that skip the actual reset steps. They want to improve answer completeness without changing the model or the retrieval documents. Which prompt engineering change is BEST?

Accepted Answer

Instruct the model in the system prompt to always answer with numbered step-by-step procedures when the user asks a procedural question.. The most reliable fix is to change the prompt so the model is explicitly told how to answer procedural questions. System-prompt instructions shape output structure and content selection far more than sampling parameters or input trimming. Because the team cannot change the model or documents, prompt-level guidance is the only lever that directly targets the missing step-by-step detail.

Answer

Shorten the user prompt to a single keyword like 'password reset' before sending it to the model.

Answer

Increase the max_tokens parameter so the model has more room to write a longer answer.

Answer

Lower the temperature to 0 so the model produces deterministic answers.

Question 9

A company is deploying a generative AI application on OCI using a dedicated AI cluster. They need to ensure that the endpoint can handle sudden spikes in traffic while maintaining cost efficiency. Which two actions should they take? (Choose two.)

Accepted Answer

Enable autoscaling on the dedicated AI cluster with a minimum of one replica and a maximum based on peak load.. To handle traffic spikes cost-effectively, the dedicated AI cluster should use autoscaling with a minimum replica for baseline availability and a maximum to cap resources. Additionally, a metric-based autoscaling policy ensures scaling decisions are driven by actual load indicators like GPU utilization or queue length. Together, these actions provide dynamic capacity and cost control.

Answer

Configure a fixed number of replicas equal to the expected average load.

Answer

Deploy multiple dedicated AI clusters and use a load balancer to distribute traffic.

Answer

Use a burst buffer to queue requests during spikes and process them later.

Question 10

A developer is building a LangChain agent using OCI Generative AI (Cohere Command R) as the reasoning LLM. The agent must decide between three tools: a SQL query tool that returns structured rows, a web search tool, and a calculator. After several test runs, the agent consistently calls the SQL tool even for simple arithmetic questions. The developer confirms the tool descriptions are generic. What is the MOST likely cause and the correct fix?

Accepted Answer

The tool descriptions are too generic, so the LLM cannot distinguish them; the developer should rewrite each tool's description to state its exact purpose, inputs, and when to use it.. LangChain agents select tools by reasoning over each tool's name and description included in the prompt. When descriptions are generic, the LLM lacks the signal needed to map a question to the right tool, so it falls back to a familiar one. Rewriting descriptions to state purpose, inputs, and when to use each tool restores the discriminative information and corrects routing.

Answer

The agent's prompt is too vague; the developer should switch to a zero-shot ReAct agent type, which removes the need for tool descriptions.

Answer

The agent's memory is retaining the first tool call and repeating it; the developer should disable memory on the agent executor.

Answer

The OCI Generative AI chat model does not support function calling, so the developer must replace it with a model that does before any agent can work.

Question 11

A company wants to deploy a generative AI model on OCI and needs to ensure that all inference requests and responses are logged for auditing purposes. They plan to use the OCI Generative AI service. Which OCI service should they integrate with to capture and store these logs?

Accepted Answer

OCI Logging. OCI Logging is the correct service for capturing and storing logs from OCI Generative AI. It allows you to enable logging for the Generative AI service, which then emits logs for inference requests and responses. These logs can be searched, analyzed, and retained for compliance. Integrating with OCI Logging ensures that all necessary data is captured in a centralized, secure location.

Answer

OCI Notifications

Answer

OCI Audit

Answer

OCI Monitoring

Question 12

You are building a RAG assistant on OCI Generative AI. After ingestion, you notice that semantically similar questions phrased differently (for example, 'How do I reset my password?' vs 'What is the procedure to change my login credentials?') return different top-k chunks. You want to improve semantic recall without changing the LLM. Which action should you take?

Accepted Answer

Replace the embedding model with a higher-quality OCI Generative AI embedding model and re-index the corpus.. Semantic recall for paraphrased queries is primarily a function of embedding quality. A stronger OCI embedding model maps semantically equivalent text to nearby vectors, so different phrasings retrieve the same relevant chunks. Chunk overlap, top-k, and temperature do not change the underlying vector representation, so they cannot fix a semantic mismatch between query and document embeddings.

Answer

Lower the temperature of the LLM to 0 so that paraphrased questions produce consistent answers.

Answer

Increase the top-k retrieval parameter from 5 to 50 and pass all chunks to the LLM.

Answer

Increase the chunk overlap from 10% to 20% while keeping the same embedding model and vector index.

Question 13

An insurance company runs a RAG application on OCI Generative AI. The vector store is Oracle Database 23ai with a VECTOR column and an IVF index. During a peak claims period, the same query returns different top-k document sets on consecutive calls, and some relevant claims are excluded. The ingestion job is confirmed to have committed all rows. Which cause is most likely?

Accepted Answer

The IVF index search uses approximate nearest neighbor probing with a limited number of probes, so the candidate set varies and some relevant vectors are missed.. IVF vector indexes are approximate: they cluster vectors and probe only a subset of clusters per query. When the probe count is low relative to the data distribution, the candidate clusters inspected can differ between calls, and true nearest neighbors in unprobed clusters are excluded. Increasing probes or switching to exact search stabilizes results and raises recall for the claims workload.

Answer

The VECTOR column was declared with a dimension smaller than the embedding model output, silently truncating stored vectors.

Answer

The LLM's context window is exceeded, so the application drops chunks before sending the prompt to the model.

Answer

The embedding model was changed between the two calls, producing vectors in a different semantic space than the stored vectors.

Question 14

A data scientist is fine-tuning a large language model on OCI Generative AI for a specialized medical diagnosis task. They have a small dataset of 500 labeled examples. Which fine-tuning approach is most appropriate to avoid overfitting?

Accepted Answer

Parameter-efficient fine-tuning (PEFT) with LoRA. PEFT with LoRA is designed for scenarios with limited data, as it only trains a small number of parameters, reducing overfitting risk. Full fine-tuning and training from scratch require large datasets. Zero-shot learning lacks domain adaptation. Therefore, PEFT with LoRA is the best choice.

Answer

Full fine-tuning of all model parameters

Answer

Training from scratch

Answer

Zero-shot learning without fine-tuning

Question 15

A developer is using the OCI Generative AI Service to build a text generation application. They want to ensure that the model's output is deterministic and reproducible for a given prompt, which is required for testing and debugging. Which parameter should they set to achieve this?

Accepted Answer

temperature. Temperature controls the randomness of the model's output. Setting it to 0 forces the model to always pick the most probable next token, eliminating sampling randomness and ensuring that the same prompt yields the same output every time. This is essential for testing and debugging scenarios where reproducibility is required.

Answer

top_k

Answer

frequency_penalty

Answer

max_tokens

Question 16

A platform team is designing a generative AI workload in OCI Generative AI that must be resilient and cost-aware. They plan to use dedicated AI clusters for hosting a fine-tuned model. Which TWO operational considerations should they account for when planning capacity and lifecycle? (Choose two.)

Accepted Answer

Dedicated AI clusters consume quota and incur cost while active, so clusters should be stopped or deleted when no longer needed. Dedicated AI clusters are billed, quota-consuming resources, and endpoints inside them share the cluster's compute. Capacity planning must therefore cover cluster lifecycle to avoid waste and endpoint sizing to meet throughput needs. The other statements misstate model persistence, elasticity, and access control, which are not accurate characteristics of dedicated AI clusters in OCI Generative AI.

Answer

Dedicated AI clusters automatically scale their GPU count up and down based on request volume without any configuration

Answer

Because the cluster is dedicated, IAM policies and compartment boundaries no longer apply to endpoint access

Answer

Fine-tuned models must be re-trained from scratch whenever a cluster is restarted, because model weights are held only in cluster memory

Question 17

A developer is using LangChain with OCI Generative AI to create a conversational agent. The agent uses a custom tool that executes SQL queries against an Oracle Autonomous Database. During testing, the agent sometimes generates SQL that is syntactically correct but semantically wrong, leading to incorrect results. Which LangChain feature should the developer implement to improve the agent's ability to generate accurate SQL queries?

Accepted Answer

Few-shot prompting with examples of correct SQL queries. Few-shot prompting with correct SQL examples teaches the model the specific schema and query patterns, reducing semantic errors. Increasing temperature adds randomness, syntax validation only catches syntactic issues, and a larger model alone doesn't address schema-specific accuracy. Therefore, few-shot prompting is the most effective LangChain feature to implement here.

Answer

Adding a separate validation tool that checks SQL syntax

Answer

Increasing the temperature setting of the language model

Answer

Switching to a larger language model with more parameters

Question 18

An enterprise is rolling out an OCI Generative AI Agents-based assistant that answers questions from internal HR policy documents. Compliance requires that answers be traceable to source documents and that access to sensitive policies be limited by user role. Which TWO capabilities should the team rely on to meet these requirements? (Choose two.)

Accepted Answer

Use OCI IAM policies and compartment-level access controls to restrict which users can query knowledge bases containing sensitive policies.. Traceability and role-based access are addressed by two distinct mechanisms: citations that link generated answers to retrieved source passages, and OCI IAM plus compartment policies that limit who can query sensitive knowledge bases. Together they give auditors a verifiable chain from answer to source while enforcing least privilege. The other options either lack a security function or actively weaken access controls.

Answer

Store the source documents in an object storage bucket with public read access so the agent can retrieve them quickly.

Answer

Increase the model's temperature so answers vary between users and are harder to reverse-engineer.

Answer

Enable a guardrail that blocks any response longer than a fixed number of characters.

Question 19

A RAG application on OCI Generative AI ingests 40,000 technical manuals. Each manual is chunked, embedded with the Cohere Embed model, and stored in an OCI OpenSearch vector index. Users now complain that answers are slow and often cite chunks from unrelated documents. Logs show every query performs an exhaustive scan across all 40,000 vectors. What should you implement to improve both latency and relevance?

Accepted Answer

Create an OpenSearch index with an HNSW-based knn_vector field and query it with approximate nearest neighbor search using the query embedding.. The exhaustive scan across all vectors is the core bottleneck, and the unrelated citations indicate ranking is not using vector similarity effectively. Building an HNSW-backed knn_vector index in OCI OpenSearch and querying with approximate nearest neighbor search reduces comparisons and ranks semantically similar chunks first. The other choices either keep the full scan, weaken semantic precision, or only mask repeated traffic.

Answer

Increase the chunk size to 8,000 tokens so each vector represents a full manual section and fewer vectors are compared per query.

Answer

Enable OCI Generative AI response caching and set a longer time-to-live so repeated queries return faster.

Answer

Switch the embedding model to a larger Cohere model and re-embed all documents nightly to raise retrieval accuracy.

Question 20

A media company uses OCI Generative AI embeddings to index English-language articles into Oracle Database 23ai. They now want the same RAG assistant to answer questions in Spanish about the same English articles. The team plans to embed the Spanish questions with the same English embedding model. What is the most likely outcome of this approach?

Accepted Answer

Retrieval quality will likely degrade because the model was not trained for cross-lingual alignment, so Spanish query vectors will not reliably match English article vectors.. Embedding models capture semantics only for languages they were trained on. An English-only model produces unreliable vectors for Spanish input, so similarity against English articles is not meaningful and retrieval degrades. The remedy is a multilingual embedding model or a translation step so queries and documents share a language the model understands.

Answer

The LLM will translate the Spanish question into English before embedding, so the embedding model receives English input.

Answer

The retrieval will work correctly as long as the vector index is rebuilt with a higher probe count to compensate for the language difference.

Answer

Spanish query vectors will align with English document vectors because embedding models map all languages into a shared semantic space automatically.