1Z0-1127-25 Practice Test 49 — 15 Questions

Question 1

An engineer is evaluating prompt variants for a customer-facing assistant on OCI Generative AI. Variant X performs well on average but occasionally produces terse, unhelpful answers when the user question is ambiguous. The team wants to add a step that makes the model ask a clarifying question instead of guessing. Which prompt engineering pattern BEST addresses this?

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

Accepted Answer

Add a conditional instruction: if the user's request lacks required details, ask one focused clarifying question before answering; otherwise answer directly.. The best fix is a conditional instruction that defines when to ask a clarifying question and when to answer directly. This targets the ambiguous-input case while preserving normal behavior. Temperature changes, unrelated few-shot examples, and repetition penalties do not encode a decision rule, so they cannot produce the desired clarification behavior.

Answer

Add a few-shot example showing a helpful, detailed answer to a well-specified question.

Answer

Increase the frequency penalty so the model avoids repeating the user's wording and must generate new content.

Answer

Lower the temperature to 0 so the model always picks the most probable continuation.

Question 2

A solutions architect needs to programmatically invoke a Cohere model hosted in OCI Generative AI from an application running on an OCI Compute instance. The application must authenticate without storing long-lived user credentials on the instance. Which authentication approach should the architect configure?

Accepted Answer

Use an instance principal with a dynamic group and a policy granting access to the generative AI resource. Instance principals allow an OCI Compute instance to obtain short-lived credentials automatically through the instance metadata service. Combined with a dynamic group and an IAM policy granting access to OCI Generative AI, the application can call the model without embedding user API keys. This is the recommended pattern for workloads running on OCI compute resources that need service access without long-lived secrets.

Answer

Embed a pre-authenticated request (PAR) URL for the generative AI endpoint in the application code

Answer

Use a customer secret key tied to the instance's compartment to sign requests to the model

Answer

Create a local OCI user, generate an API key, and store the private key on the instance

Question 3

A developer is preparing prompts for an OCI Generative AI chat model. They notice that adding a short instruction such as 'Answer in one sentence' at the start of the prompt changes the model's output style, even though the model weights are not updated. Which concept explains this behavior?

Accepted Answer

In-context learning, because the model conditions on the instruction within the prompt to shape its response.. The model weights are unchanged, so the behavior must come from the prompt itself. In-context learning describes exactly this: the decoder conditions on instruction tokens and shifts its output distribution accordingly. Fine-tuning, RAG, and RLHF all involve external training or retrieval steps that are absent in this scenario.

Answer

Reinforcement learning from human feedback, because the model has been aligned to follow formatting instructions.

Answer

Fine-tuning, because the model adapts its parameters to the new instruction.

Answer

Retrieval-augmented generation, because the model retrieves relevant instructions from a vector store.

Question 4

A developer is assembling an LCEL chain that must send a fixed system instruction, the user's question, and the retrieved context to an OCI Generative AI chat model in a single call. Which LangChain component is responsible for combining these pieces into the final prompt sent to the model?

Accepted Answer

A ChatPromptTemplate with system, context, and human message placeholders.. Prompt construction for a chat model belongs to a chat prompt template, which declares message roles and placeholders for the system instruction, retrieved context, and user question. Splitters prepare documents before retrieval, output parsers handle the response afterward, and passthrough utilities only move data through the chain without building messages.

Answer

A RunnablePassthrough that concatenates all inbound dictionary values into one string.

Answer

An OutputParser that reformats the model response into the system instruction format.

Answer

A RecursiveCharacterTextSplitter configured with the system instruction as a separator.

Question 5

A data scientist is using OCI Generative AI to generate summaries of legal documents. The documents are long, often exceeding the model's context window. The data scientist wants to ensure that the entire document is considered while minimizing information loss. Which strategy should they use?

Accepted Answer

Split the document into chunks, generate summaries for each chunk, and then combine the summaries into a final summary.. For long documents exceeding the context window, the map-reduce summarization pattern is the most reliable: chunk the document, summarize each chunk, then summarize the summaries. Truncation loses information, temperature does not extend context, and even large context models have limits and cost implications. This approach ensures the entire document is considered while staying within model constraints.

Answer

Truncate the document to the first N tokens that fit the context window and generate a summary from that.

Answer

Use a larger model that supports a longer context window, such as a 128k token model, and pass the entire document in one request.

Answer

Increase the temperature parameter to encourage the model to generate more comprehensive summaries.

Question 6

A solutions architect is comparing hosting options for a generative AI workload. They need the model weights to remain entirely under their control and want predictable throughput even when the shared service is busy. Which OCI Generative AI option should they choose?

Accepted Answer

A dedicated AI cluster with the model hosted on it. Reserved capacity and customer-controlled hosting point to a dedicated AI cluster, where the customer provisions GPU capacity and hosts the model themselves. That removes exposure to shared-service rate limits and keeps model weights inside the customer's environment. Shared on-demand inference, fine-tuned models on the shared endpoint, and notebook-based serving all leave throughput subject to shared capacity or are not designed for production inference.

Answer

A fine-tuned custom model served on the shared endpoint

Answer

On-demand inference against a shared pretrained model

Answer

Importing the model into OCI Data Science and calling it with a notebook session

Question 7

A team wants to evaluate several prompts against a hosted model in OCI Generative AI without writing application code. They need an interactive interface where they can adjust parameters and compare outputs quickly. Which OCI Generative AI feature should they use?

Accepted Answer

The OCI Generative AI Playground. The OCI Generative AI Playground is the console-based, no-code environment for selecting models, entering prompts, adjusting parameters, and reviewing outputs interactively. It supports rapid experimentation and comparison of prompt variants against hosted models. The other options involve hosting endpoints, fine-tuning workflows, or custom code, none of which provide the requested interactive, code-free prompt evaluation experience.

Answer

The OCI Generative AI fine-tuning workflow in the console

Answer

A dedicated AI cluster endpoint invoked through the OCI CLI

Answer

An OCI Functions application that calls the model and logs responses

Question 8

An ML engineer is preparing a fine-tuning dataset for an OCI Generative AI custom model that must answer questions about a proprietary 40-page employee handbook. The engineer has 300 question-answer pairs but no long-form documents in the training file. The evaluation shows the model answers simple factual questions correctly but fails on questions requiring synthesis across multiple sections. Which dataset design change best addresses this limitation?

Accepted Answer

Add training examples where each answer requires combining information from two or more distinct handbook sections, with the source sections cited.. The evaluation shows the fine-tuned model can retrieve isolated facts but cannot integrate evidence across sections, which points to a training-signal structure problem rather than a volume problem. Adding multi-section synthesis examples, especially with citations, teaches the model the reasoning pattern the evaluation is testing. Duplicating data, isolating answers to single chunks, or swapping the corpus does not create that multi-hop signal. Dataset design, not dataset size, is the lever that closes this gap.

Answer

Replace the handbook with a public HR policy corpus so the model has more diverse training material.

Answer

Duplicate the 300 question-answer pairs three times to increase the effective dataset size to 900 examples.

Answer

Convert the handbook into chunked passages and pair each chunk with a question whose answer appears only in that chunk.

Question 9

A solutions architect is building a RAG pipeline on OCI Generative AI. The team stores document chunks in an OCI Search with OpenSearch index and wants the vector dimension of the index to match the embeddings produced by the cohere.embed-english-v3.0 model. Which action should the architect take to ensure the index mapping is compatible with the embedding output?

Accepted Answer

Create the index with a knn_vector field whose dimension matches the embedding model's output dimension (for example, 1024 for cohere.embed-english-v3.0).. The k-NN vector field dimension must equal the embedding length produced by the model. Cohere Embed English v3.0 returns 1024-dimensional vectors by default, so declaring a knn_vector with dimension 1024 lets documents index and lets queries run approximate nearest neighbor search. Shard counts and dynamic mapping cannot substitute for matching the declared dimension.

Answer

Set the index's number_of_shards to match the embedding dimension so each shard stores one vector component.

Answer

Enable dynamic mapping and let OpenSearch infer the vector dimension from the first document indexed.

Answer

Store embeddings as base64-encoded strings in a text field and convert them at query time.

Question 10

A team is evaluating a decoder-only LLM for a summarization task and wants to understand the role of tokenization in model behavior. Which two statements about tokenization in LLMs are correct? (Choose two.)

Accepted Answer

Subword tokenization allows rare words to be represented as combinations of more frequent subword units.. Tokenization maps text to integer IDs consumable by embeddings and, through subword schemes, represents rare words compositionally. It occurs before the transformer and its granularity varies, affecting context usage and cost. Claims of one-token-per-word or tokenization inside attention contradict how the inference pipeline actually works.

Answer

Tokenization guarantees that every word in any language maps to exactly one token.

Answer

Because token counts are fixed per word, tokenization has no effect on context window usage.

Answer

Tokenization is performed by the attention mechanism inside each transformer layer at inference time.

Question 11

A developer is using the OCI Generative AI inference API with a Cohere model. They observe that responses vary significantly between identical requests and sometimes include irrelevant content. They want more deterministic, focused answers for a structured extraction task. Which parameter adjustment strategy should they apply?

Accepted Answer

Lower the temperature toward zero and reduce top-p to narrow the sampling pool. Response variability and irrelevant content in extraction tasks are typically addressed by reducing sampling randomness. Lowering temperature toward zero makes the model choose the most probable tokens, and reducing top-p narrows the candidate pool. Together these produce more deterministic, focused outputs. Increasing randomness, changing output length, or relying on a larger model with default sampling does not directly control the variability observed.

Answer

Increase the maximum output tokens so the model has more room to explain its reasoning

Answer

Switch to a larger model family and keep all sampling parameters at their defaults

Answer

Increase the temperature value and raise top-p to broaden sampling

Question 12

A data scientist is fine-tuning a base model on OCI Generative AI using a dataset of customer support conversations. The dataset contains many examples where the assistant's response is a single word like 'Yes' or 'No'. After fine-tuning, the model frequently produces overly terse answers even for complex queries. Which data characteristic is the most likely cause?

Accepted Answer

The dataset lacks sufficient examples of longer, more detailed responses, causing the model to learn a bias toward brevity.. Fine-tuning a model on a dataset where most assistant responses are very short teaches the model to mimic that brevity. The model adjusts its weights to minimize loss on the training data, so it learns that short answers are appropriate. To correct this, the dataset should include a balanced mix of short and detailed responses, or the training data should be filtered to remove trivial examples.

Answer

The learning rate used during fine-tuning was too high, causing the model to forget how to generate long sequences.

Answer

The base model's tokenizer is incompatible with the dataset's vocabulary, causing truncation of longer responses.

Answer

The training dataset was not shuffled before fine-tuning, so the model only saw short examples in the final epochs.

Question 13

An enterprise is deploying an LLM assistant that must answer questions grounded in a frequently updated internal knowledge base. The team wants to minimize hallucinations and avoid retraining the model whenever documents change. Which approach best meets these requirements?

Accepted Answer

Use retrieval-augmented generation to fetch relevant passages from a vector store and include them in the prompt.. Retrieval-augmented generation grounds responses in retrieved, up-to-date passages without changing model weights, so document updates only require re-indexing. Fine-tuning is costly and repeated, higher temperature worsens accuracy, and pasting the whole knowledge base is infeasible and dilutes relevance. RAG therefore best satisfies both grounding and freshness requirements.

Answer

Increase the model's temperature so it explores more possibilities and finds the correct answer.

Answer

Expand the context window and paste the entire knowledge base into every prompt.

Answer

Fine-tune the base model on the entire knowledge base each time documents are updated.

Question 14

A team is comparing two approaches for a customer support assistant. Approach 1 fine-tunes a base model on 5,000 historical support tickets. Approach 2 uses few-shot prompting with examples embedded in each request. The team wants the fastest path to a working prototype with minimal infrastructure. Which approach should they choose and why?

Accepted Answer

Approach 2, because few-shot prompting requires no training and can be iterated quickly. Few-shot prompting is the fastest path to a working prototype because it needs no training pipeline, labeled dataset for weight updates, or dedicated model hosting. Fine-tuning can improve accuracy later but adds latency, cost, and complexity that conflict with the stated goal of minimal infrastructure and quick iteration.

Answer

Approach 2, because few-shot prompting eliminates the need for any labeled data

Answer

Approach 1, because few-shot prompting cannot handle multi-turn conversations

Answer

Approach 1, because fine-tuning always produces higher accuracy than prompting

Question 15

A developer is using LangChain to build a retrieval-augmented generation (RAG) application with OCI Generative AI. They need to split a large PDF document into smaller chunks before embedding. Which LangChain component is specifically designed to split text into chunks based on character count?

Accepted Answer

TextSplitter. TextSplitter is the LangChain abstraction specifically built to break large texts into smaller chunks. It supports various splitting methods, including character-based splitting, which matches the requirement. DocumentLoader loads the PDF, Embeddings vectorize chunks, and VectorStore stores them—none of these split text. Thus, TextSplitter is the correct choice for chunking the document.

Answer

VectorStore

Answer

Embeddings

Answer

DocumentLoader