1Z0-1127-25 Fundamentals of Large Language Models Practice Test 3 — 15 Questions

Question 1

Which TWO factors most significantly influence the computational cost of fine-tuning a large language model?

Accepted Answer

Number of model parameters. The number of model parameters directly determines the size of the weight matrices that must be updated during backpropagation. Fine-tuning requires storing gradients and optimizer states for each parameter, so the memory and compute scale linearly with parameter count. This is the primary driver of FLOPs (floating-point operations) per training step.

Answer

Batch size

Answer

Quantization bits

Answer

Dataset size

Question 2

Which TWO techniques are commonly used to reduce the memory footprint of LLM inference?

Accepted Answer

Quantization. Quantization reduces the memory footprint by lowering the precision of model weights and activations from FP32 to lower bit-widths like INT8 or FP16, which directly decreases the memory required to store and compute with the model. KV cache optimization reduces memory usage by efficiently managing the key-value cache during autoregressive decoding, often through techniques like shared memory, pruning, or compression, which is critical for long-context inference.

Answer

Increasing batch size

Answer

Gradient checkpointing

Answer

Using full precision (FP32)

Question 3

Which THREE are essential steps in the prompt engineering process for an LLM?

Accepted Answer

Test the prompt with a variety of input examples. Testing the prompt with a variety of input examples is essential to evaluate the LLM's generalization, robustness, and sensitivity to different phrasing or contexts. This step helps identify edge cases, biases, or inconsistencies in the model's responses before deployment.

Answer

Fine-tune the model on a domain corpus

Answer

Quantize the model to INT8

Question 4

A data scientist is using a large language model to summarize customer support tickets. The model occasionally generates summaries that include hallucinated details not present in the original ticket. Which technique would best reduce hallucinations while maintaining summary quality?

Accepted Answer

Implement retrieval-augmented generation (RAG) to ground the model in relevant documents.. Retrieval-Augmented Generation (RAG) reduces hallucinations by grounding the model's output in external, verifiable documents retrieved from a knowledge base. Instead of relying solely on the model's parametric memory, RAG fetches relevant context (e.g., the original ticket) at inference time, ensuring the summary is factually aligned with the source. This maintains summary quality because the model can still generate fluent text while being constrained to the retrieved evidence.

Answer

Use a longer system prompt instructing the model to be factual.

Answer

Fine-tune the model on a large corpus of general text to improve its knowledge.

Answer

Increase the temperature parameter to 0.9 to encourage more deterministic outputs.

Question 5

An enterprise is deploying a chat application using a large language model. Users report that the model sometimes generates toxic or biased responses. Which best practice should be applied to mitigate this issue?

Accepted Answer

Implement a content filtering layer using a safety classifier to detect and block toxic outputs.. Implementing a content filtering layer using a safety classifier is a proven best practice to detect and block toxic or biased outputs in real-time. This approach acts as a guardrail, intercepting harmful responses before they reach users, and is independent of the model's internal parameters or training data.

Answer

Use few-shot prompting with examples of toxic responses so the model learns to avoid them.

Answer

Increase the max_tokens parameter to allow the model more context to correct itself.

Answer

Disable the temperature parameter to make outputs deterministic.

Question 6

A team is fine-tuning a large language model for a domain-specific Q&A application. After fine-tuning, they observe that the model performs well on the training distribution but struggles with out-of-distribution (OOD) questions. Which approach would best improve OOD robustness?

Accepted Answer

Include a diverse set of examples from related domains in the fine-tuning dataset.. Including diverse examples from related domains (Option A) exposes the model to varied patterns, reducing overfitting to the training distribution and improving generalization to out-of-distribution inputs. Option B (early stopping) helps prevent overfitting on the validation set but does not specifically address OOD robustness. Option C (reducing model size) limits capacity, which can harm performance on both in-distribution and OOD data. Option D (increasing learning rate) risks catastrophic forgetting and training instability, not OOD robustness.

Answer

Use early stopping based on training loss to avoid overfitting.

Answer

Reduce the model size to prevent overfitting to the training data.

Answer

Increase the learning rate during fine-tuning to adapt faster to new patterns.

Question 7

A developer is using a large language model to generate code snippets. The model often produces code that is syntactically correct but functionally incorrect. What is the most effective way to improve the functional correctness of the generated code?

Accepted Answer

Provide few-shot examples of correct code in the prompt.. Providing few-shot examples of correct code in the prompt directly demonstrates the desired functional behavior to the model. This technique, known as few-shot prompting, grounds the model's output in concrete examples, significantly improving the likelihood that the generated code will be functionally correct by aligning the model's pattern completion with the intended logic, not just syntax.

Answer

Increase the temperature parameter to generate more creative solutions.

Answer

Ask the model to only output syntactically valid code.

Answer

Set max_tokens to a very high value to allow the model more room to think.

Question 8

A company is deploying a large language model in a customer-facing chatbot. The model's responses must be both accurate and safe. Which combination of techniques should be employed?

Accepted Answer

Use retrieval-augmented generation (RAG) for factual accuracy and a content safety filter for safe outputs.. RAG grounds the model's responses in a verified external knowledge base, reducing hallucinations and improving factual accuracy, while a content safety filter (e.g., a classifier or guardrail) actively blocks toxic or unsafe outputs before they reach the user. This combination addresses both accuracy and safety independently, unlike a single system prompt which is easily bypassed.

Answer

Use only a system prompt instructing the model to be accurate and safe.

Answer

Use a high temperature for creativity and a safety classifier for blocking toxic outputs.

Answer

Fine-tune the model on all historical chat logs and use a high temperature.

Question 9

A research team is experimenting with few-shot prompting to improve a model's performance on a complex reasoning task. They find that the model's performance degrades when the few-shot examples are too similar to each other. What is the likely cause and best remedy?

Accepted Answer

The examples lack diversity, causing the model to overfit to a narrow pattern. Use more diverse examples.. When few-shot examples are too similar, the model overfits to a narrow pattern, reducing its ability to generalize to the diverse reasoning paths required by the task. This is a known limitation of in-context learning: the model treats the examples as a template rather than as diverse demonstrations. Using more diverse examples exposes the model to a wider range of reasoning patterns, improving robustness.

Answer

The model has not seen enough examples. Increase the number of few-shot examples.

Answer

The examples are presented in a confusing order. Reorder them by difficulty.

Answer

The temperature is too low, making the model too deterministic. Increase temperature slightly.

Question 10

Which TWO of the following are common applications of large language models in enterprise settings?

Accepted Answer

Summarizing lengthy legal documents.. Large language models (LLMs) excel at abstractive summarization, which involves condensing lengthy legal documents into concise summaries while preserving key facts and legal reasoning. This is a common enterprise application for legal departments, as LLMs can process large volumes of text and generate coherent, context-aware summaries without requiring manual reading.

Answer

Performing real-time signal processing for audio streams.

Answer

Replacing relational databases for data storage.

Answer

Enhancing low-resolution images through super-resolution.

Question 11

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.. Large language models (LLMs) are prone to hallucination, where they generate plausible-sounding but factually incorrect information that was not present in the input. This occurs because LLMs are next-token predictors without a built-in fact-checking mechanism, and they can invent details, citations, or events to maintain coherence. Practitioners must implement retrieval-augmented generation (RAG) or external verification to mitigate this risk.

Answer

Limited to processing only one language at a time.

Answer

Inability to process inputs longer than a few hundred tokens.

Question 12

Refer to the exhibit. A developer ran the OCI CLI command shown and received the JSON output. What does the output indicate about the model's confidence and why?

Accepted Answer

The model is highly confident the text is positive, as indicated by the 0.98 score.. The JSON output shows a sentiment score of 0.98 for 'positive', which is very close to 1.0, indicating the model is highly confident that the text is positive. In sentiment analysis models, scores represent probabilities for each class, and a value near 1.0 for one class with much lower scores for others reflects strong confidence.

Answer

The model is uncertain because all scores are roughly equal.

Answer

The model is neutral because the neutral score is lowest.

Answer

The model is unsure because the scores are probabilities that sum to 1.

Question 13

Refer to the exhibit. A developer runs the OCI CLI command and receives the output. However, the text "Hello, how are you?" is actually a mix of English and French words. Why does the model assign only 0.03 to French?

Accepted Answer

The text is overwhelmingly English, so the model assigns a low probability to French.. The model's output shows a probability distribution over languages, and the text is predominantly English with only a few French words. The model assigns a low probability (0.03) to French because the overwhelming majority of tokens are English, making the text far more likely to be classified as English. This reflects how language identification models evaluate the overall composition of the input.

Answer

The model is limited to identifying a single language per query.

Answer

The model cannot detect multiple languages in a single text.

Answer

The model's scores are normalized to sum to 1, so a high English score forces low others.

Question 14

You are a machine learning engineer at a large e-commerce company. You have been tasked with deploying a large language model to power a customer service chatbot that handles product returns and refunds. The model will answer customer queries based on a knowledge base of return policies and FAQs. The company has strict requirements: (1) responses must be factually accurate and grounded in the knowledge base, (2) the system must be cost-effective, and (3) latency should be under 2 seconds per response. You decide to use a pre-trained LLM from OCI Data Science and implement retrieval-augmented generation (RAG). You have two options for the retriever: a dense embedding-based retriever (e.g., using OCI AI Language embeddings) or a sparse keyword-based retriever (e.g., BM25). You also need to decide on the generation model size: a 7B parameter model or a 70B parameter model. You run a pilot test: with the dense retriever + 7B model, average latency is 1.8 seconds and accuracy is 85%. With the sparse retriever + 7B model, latency is 1.2 seconds but accuracy drops to 75%. With the 70B model (any retriever), latency exceeds 5 seconds. Which combination should you choose to meet all requirements?

Accepted Answer

Dense retriever + 7B model.. (dense retriever + 7B model) is correct because it meets all three requirements: factual accuracy (85% accuracy from dense retrieval grounding), latency under 2 seconds (1.8 seconds), and cost-effectiveness (7B model is cheaper to run than 70B). The dense retriever provides better semantic matching for nuanced return policy queries, while the 7B model keeps inference fast and affordable.

Answer

Sparse retriever + 70B model.

Answer

Dense retriever + 70B model.

Answer

Sparse retriever + 7B model.

Question 15

A healthcare startup is building an AI assistant to help doctors draft clinical notes from patient-physician conversations. They have a large language model that is fine-tuned on medical data. During testing, they notice the model occasionally generates plausible-sounding but incorrect medical recommendations. The startup wants to deploy the assistant to assist doctors, not replace them. They have the following options: (A) Deploy the model as-is and rely on doctors to catch errors, (B) Add a disclaimer that the model may make mistakes, (C) Implement a fact-checking pipeline that cross-references outputs with a trusted medical knowledge base before presenting to doctors, (D) Reduce the model's temperature to 0 to ensure deterministic outputs. Which option best balances safety and utility?

Accepted Answer

Implement a fact-checking pipeline that cross-references outputs with a trusted medical knowledge base.. Implementing a fact-checking pipeline that cross-references outputs with a trusted medical knowledge base directly mitigates the risk of hallucinated medical recommendations while preserving the assistant's utility. This approach leverages retrieval-augmented generation (RAG) principles to ground the model's outputs in verified facts, ensuring safety without sacrificing the flexibility needed for drafting clinical notes.

Answer

Add a disclaimer that the model may make mistakes.

Answer

Deploy the model as-is and rely on doctors to catch errors.

Answer

Reduce the model's temperature to 0 to ensure deterministic outputs.