1Z0-1127-25 Practice Test 26 — 15 Questions

Question 1

Which LangChain document loader should be used to load text from a web page given its URL?

Accepted Answer

WebBaseLoader. WebBaseLoader is the LangChain document loader designed to load text from a web page given its URL. It fetches the HTML content and extracts the text, making it suitable for loading web documents into a LangChain pipeline.

Answer

CSVLoader

Answer

PDFLoader

Answer

TextLoader

Question 2

Which of the following best describes the difference between pre-training and fine-tuning?

Accepted Answer

Pre-training trains a model from scratch on a general corpus; fine-tuning adapts the model to a specific task. Pre-training involves training a large language model from scratch on a vast, general corpus of unlabeled data to learn language patterns, grammar, and world knowledge. Fine-tuning then takes this pre-trained model and further trains it on a smaller, labeled dataset specific to a downstream task, such as sentiment analysis or question answering, adapting the model's weights for that particular objective. Option C correctly captures this fundamental distinction in the LLM workflow.

Answer

Pre-training uses labeled data; fine-tuning uses unlabeled data

Answer

Pre-training requires a GPU; fine-tuning can be done on CPU

Answer

Pre-training is only for encoder models; fine-tuning is only for decoder models

Question 3

An organization wants to use OCI Generative AI to summarize long legal documents. They need to ensure the summary is concise and retains key information. Which model parameter should they set to control the length of the summary?

Accepted Answer

max_tokens. The `max_tokens` parameter directly controls the maximum number of tokens (words or subwords) the model can generate in its response. By setting a lower value for `max_tokens`, the organization can enforce a concise summary, ensuring the output length is capped while retaining key information. Other parameters like `temperature` or `top_p` affect randomness or diversity, not output length.

Answer

frequency_penalty

Answer

top_p

Answer

temperature

Question 4

Which TWO are benefits of using dedicated AI clusters for OCI Generative AI?

Accepted Answer

Guaranteed throughput. Option B (Guaranteed throughput) is correct because dedicated AI clusters on OCI reserve compute capacity exclusively for a tenant, so the provisioned throughput is not shared with or throttled by other on-demand users. Option E (Predictable inference latency) is correct because that same reserved, isolated capacity avoids the noisy-neighbor contention typical of shared on-demand endpoints, yielding consistent response times. The unmarked options do not belong: automatic model updates (A) are not a property of dedicated clusters, since model versioning is controlled by the customer; lower cost than on-demand for all workloads (C) is false because dedicated capacity is only cheaper for sustained, high-utilization workloads and more expensive for intermittent ones; and no need to manage scaling (D) is incorrect because dedicated clusters still require capacity planning and scaling decisions by the customer.

Answer

Automatic model updates

Answer

Lower cost than on-demand for all workloads

Answer

No need to manage scaling

Question 5

Your company uses OCI Data Science for model development and deployment. You have a generative AI model that requires dynamic batching for efficient inference. You deployed the model using the OCI Model Deployment service with a custom inference script in a Docker container. However, you notice that the batch size is fixed at 1, leading to low throughput. The model can process multiple requests together efficiently. You want to implement dynamic batching to increase throughput without significantly increasing latency for individual requests. What is the best approach?

Accepted Answer

Implement a queuing mechanism in the inference script that collects incoming requests and processes them in batches. Dynamic batching must be implemented at the application level within the custom inference script when using OCI Model Deployment. The service does not provide built-in request batching; instead, you need to collect incoming requests in a queue and process them together in a single forward pass, which maximizes GPU utilization while controlling latency via a timeout or max batch size.

Answer

Modify the model deployment to use a larger GPU shape to handle larger batches

Answer

Enable the model deployment's built-in request batching feature

Answer

Use OCI Streaming service to buffer requests and then invoke the model in batches from a consumer

Question 6

A company uses an LLM to generate product descriptions. The outputs are consistently too verbose and include irrelevant details. The prompt includes a simple instruction: 'Describe the product.' Which adjustment to the prompt is most likely to yield concise, relevant descriptions?

Accepted Answer

Add constraints like 'Max 30 words. Focus on key features.'. Adding explicit constraints like 'Max 30 words. Focus on key features.' directly instructs the LLM to limit verbosity and prioritize relevant details. This technique, known as prompt engineering with constraints, is the most effective way to control output length and content without altering model parameters or relying on examples that may not generalize.

Answer

Set temperature to 0.

Answer

Increase max_tokens to 500.

Answer

Include a few examples of desired short descriptions.

Question 7

Which THREE of the following are best practices when deploying a generative AI model on OCI?

Accepted Answer

Set up autoscaling for the endpoint.. Option B is correct because configuring autoscaling on an OCI generative AI endpoint (via the endpoint's scaling policy) lets the service dynamically add or remove dedicated AI cluster capacity as inference traffic fluctuates, maintaining performance while controlling cost. Option D is correct because production endpoints should run on a dedicated AI cluster, which provides isolated, reserved GPU capacity and predictable performance rather than sharing a multi-tenant host pool. Option E is correct because enabling content filtering (guardrails/moderation on the endpoint) helps block harmful, unsafe, or policy-violating prompts and responses, which is a security and responsible-AI best practice for generative AI deployments. Option A is wrong because API keys are secrets and must be stored in OCI Vault (or equivalent secret management), never embedded in endpoint configuration. Option C is wrong because disabling logging removes auditability and observability; logging should be enabled and routed to OCI Logging rather than turned off to save costs.

Answer

Store API keys in the model endpoint configuration.

Answer

Disable logging to save costs.

Question 8

Which technique allows an LLM to be adapted to a new task with only a few examples?

Accepted Answer

Few-shot learning. Few-shot learning (A) is the correct technique because it enables an LLM to adapt to a new task by providing a small number of input-output examples directly in the prompt, without updating the model's weights. This leverages the model's in-context learning ability, where it generalizes from the provided examples to perform the task. It is distinct from fine-tuning, which requires retraining on a labeled dataset.

Answer

Fine-tuning

Answer

Pre-training

Answer

Prompt engineering

Question 9

An organization is planning to use OCI Generative AI for sensitive customer data. Which three OCI services or features should they consider for data governance and security?

Accepted Answer

OCI Vault for managing API keys. OCI Vault (A) is correct because it securely stores and manages the API keys and secrets used to authenticate to OCI Generative AI, keeping credentials out of code and rotating them centrally. OCI IAM (C) is correct because it enforces least-privilege access control through policies, groups, and dynamic groups, governing who can invoke Generative AI endpoints and access sensitive data. OCI Audit (E) is correct because it automatically records API calls and control-plane activity, providing the logging and traceability needed for governance and compliance investigations. OCI Data Safe (B) is not the right fit here because it targets database security and masking for Oracle databases, not Generative AI model or prompt data. OCI Data Labeling (D) is not a governance or security control; it is a data-annotation service used to prepare training datasets.

Answer

OCI Data Safe for data masking and encryption

Answer

OCI Data Labeling for annotating data

Question 10

Which THREE of the following are known limitations of LLMs that practitioners must account for?

Accepted Answer

Knowledge cutoff: the model only knows information up to its training data date. Option A is correct because LLMs are trained on a fixed corpus with a knowledge cutoff date, so they cannot know events, facts, or updates that occurred after that point unless given external context (e.g., via RAG or tools). Option B is correct because LLMs learn statistical patterns from training data that may encode societal, historical, or representational biases, which the model can reproduce or amplify in its outputs. Option C is correct because LLMs generate tokens probabilistically to maximize plausibility rather than truth, so they can produce fluent, confident, but factually incorrect statements (hallucinations). Option D is not a known limitation, since LLMs are widely used for creative text generation such as stories, poems, and brainstorming. Option E is also incorrect, as LLMs demonstrably model syntax and grammar statistically, enabling coherent and grammatically structured language production.

Answer

Inability to produce creative text

Answer

Complete lack of understanding of language syntax

Question 11

An AI assistant needs to solve complex math word problems step by step. Which prompting technique is most suitable?

Accepted Answer

Chain-of-thought prompting with few-shot examples.. Chain-of-thought prompting with few-shot examples is most suitable because it guides the LLM to break down complex math word problems into intermediate reasoning steps, mimicking human problem-solving. Few-shot examples provide a template for the desired reasoning structure, which significantly improves accuracy on multi-step arithmetic tasks compared to direct answer generation.

Answer

Zero-shot prompting with the problem only.

Answer

Prompting with a high temperature setting.

Answer

Using a model with a larger context window.

Question 12

Which of the following best describes the difference between an embedding model and a generation model?

Accepted Answer

Embedding models output a single vector per input; generation models output a sequence of tokens. Embedding models transform input data (text, images, etc.) into a fixed-size vector representation (a single vector per input) that captures semantic meaning, while generation models (like GPT or LLaMA) produce variable-length sequences of tokens as output, predicting one token at a time. This fundamental architectural difference defines their distinct roles: embeddings for similarity search and retrieval, generation for producing coherent text.

Answer

Embedding models cannot be used with RAG; generation models can

Answer

Embedding models are used only for classification; generation models are used only for translation

Answer

Embedding models require fine-tuning; generation models do not

Question 13

A data scientist is evaluating an LLM for a summarization task. They have a set of human-written reference summaries. Which THREE metrics are commonly used to evaluate summarization quality? (Choose three.)

Accepted Answer

BLEU. ROUGE, BLEU, and BERTScore are all used for summarization evaluation. Perplexity measures model confidence, and cosine similarity is for embedding comparison.

Answer

Cosine similarity

Answer

Perplexity

Question 14

Which THREE techniques effectively reduce query latency in a RAG system?

Accepted Answer

Pre-compute embeddings for all documents. Option A (pre-compute embeddings for all documents) is correct because it removes the embedding-model inference cost from the query path: documents are embedded once at ingestion time, so at query time only the user query must be embedded, cutting per-request latency. Option B (use approximate nearest neighbor search) is correct because ANN indexes such as HNSW or IVF trade a small amount of recall for sublinear search, avoiding the exhaustive exact k-NN scan over millions of vectors and dramatically lowering retrieval latency. Option E (use a smaller embedding model) is correct because a lower-dimensional, lighter encoder produces query vectors faster and yields smaller vectors that reduce distance-computation and index-lookup time, directly lowering latency. Option C (use a larger generation model) does not belong because a bigger LLM increases token-generation time and thus raises, not reduces, end-to-end latency. Option D (increase the number of shards) does not belong because more shards add fan-out, coordination, and merge overhead across nodes, which typically increases query latency rather than reducing it.

Answer

Use a larger generation model

Answer

Increase the number of shards

Question 15

A prompt engineer is designing a system prompt for a customer support agent. The goal is to ensure the agent always responds in a polite, professional tone and never provides medical advice. Which approach best achieves this?

Accepted Answer

Use a system prompt that includes both the desired tone and the constraint about medical advice. A system prompt is the correct place to encode persistent behavioral rules — desired tone and hard constraints — because it is applied to every turn and takes precedence in steering model behavior. Including both the polite/professional tone and the 'never provide medical advice' constraint in the system prompt ensures the rules apply consistently across all user interactions. This is the standard prompt-engineering pattern for durable guardrails.

Answer

Set the temperature to 0 to ensure consistent tone and hope it avoids medical topics

Answer

Specify tone in the system prompt and add the medical advice restriction in each user message

Answer

Rely on the model's inherent training to avoid medical advice, so no explicit instruction needed