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

Question 1

A data scientist is fine-tuning a Cohere Command model in OCI Generative AI on a dataset of 12,000 customer-support conversations. After training, the model produces fluent responses but frequently hallucinates product names that do not exist in the company's catalog. The dataset itself contains only real product names. Which training hyperparameter should the scientist adjust first to reduce this behavior?

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

Accepted Answer

Reduce the temperature used at inference time to make the output distribution less random.. Hallucinated entities in a fine-tuned model are usually a decoding-time symptom rather than a training-data defect, especially when the corpus contains only real product names. Lowering temperature sharpens the probability distribution so the model selects tokens it actually learned, reducing the chance of inventing plausible-sounding but nonexistent names. Retraining changes such as more epochs, a different learning rate, or longer sequences do not directly constrain token selection and may even worsen overfitting.

Answer

Lower the learning rate so weight updates during fine-tuning are smaller and more stable.

Answer

Increase the maximum sequence length of the training examples to capture more context.

Answer

Increase the number of training epochs so the model sees each conversation more times.

Question 2

A developer is building a chatbot using OCI Generative AI. They want to ensure the model produces coherent and contextually relevant responses while avoiding repetitive or nonsensical output. Which two parameters should they adjust to control the randomness and creativity of the generated text? (Choose two.)

Accepted Answer

Temperature. Temperature and top-p are the primary parameters for controlling randomness and creativity in LLM generation. Temperature adjusts the sharpness of the probability distribution, while top-p limits the token pool dynamically. Together, they allow fine-tuning of the balance between coherence and diversity. The other options affect length or repetition but not the core randomness.

Answer

Max tokens

Answer

Frequency penalty

Answer

Stop sequences

Question 3

A data scientist is using OCI Generative AI to generate product descriptions. They notice that the model sometimes produces outputs that are too short and lack detail. Which inference parameter should they adjust to encourage longer, more detailed responses?

Accepted Answer

Max tokens. The max tokens parameter caps the number of tokens generated. If it is too low, responses are cut short. Increasing it allows the model to continue generating, resulting in longer and more detailed product descriptions. Other parameters affect style or diversity, not length.

Answer

Top-p (nucleus sampling)

Answer

Temperature

Answer

Frequency penalty

Question 4

A company is deploying a large language model for a customer service chatbot. The model needs to understand industry-specific jargon and maintain low latency. Which approach best balances these requirements?

Accepted Answer

Fine-tune a small open-source LLM on domain-specific data. Fine-tuning a small open-source LLM on domain-specific data is the best approach because it adapts the model to understand industry-specific jargon while keeping the model small enough to maintain low latency. Unlike larger models, a fine-tuned small model can run efficiently on local hardware, reducing inference time and avoiding the overhead of external API calls or large model sizes.

Answer

Employ retrieval-augmented generation (RAG) with a general model

Answer

Rely solely on prompt engineering with a general model

Answer

Use a large general-purpose LLM with zero-shot prompting

Question 5

A data scientist observes that their fine-tuned LLM performs well on training data but generates repetitive and dull responses in production. What is the most likely cause and best solution?

Accepted Answer

The temperature is set too low; increase temperature during inference. The model's repetitive and dull responses indicate that the temperature parameter is too low, causing the model to always select the most probable tokens, leading to deterministic and monotonous outputs. Increasing temperature during inference introduces randomness into token sampling, allowing for more diverse and creative responses. This is a common issue in production LLMs where low temperature settings optimized for training metrics fail to produce engaging real-world outputs.

Answer

The model is overfitted; apply stronger regularization

Answer

The training data lacks diversity; add more varied examples

Answer

The model has too many layers; reduce model size

Question 6

An organization wants to use an LLM to summarize legal documents. Which consideration is most important for ensuring accurate summaries?

Accepted Answer

Fine-tune the model on a curated legal corpus. Fine-tuning on a curated legal corpus is the most important consideration because legal summarization demands domain-specific terminology, citation conventions, and reasoning patterns that general-purpose models often mishandle. Fine-tuning adapts the model's weights to legal language, improving factual fidelity and reducing hallucinations on statutes, case law, and contracts. While prompting and model size help, domain adaptation is the decisive factor for accuracy in a specialized field.

Answer

Use the largest available general-purpose model

Answer

Rely on zero-shot summarization with careful prompting

Answer

Pre-train a new model from scratch on legal texts

Question 7

A developer is building a code generation assistant. The model occasionally produces syntactically correct but semantically wrong code. Which technique directly addresses semantic correctness?

Accepted Answer

Apply RLHF using human-validated code examples. Reinforcement Learning from Human Feedback (RLHF) directly addresses semantic correctness by fine-tuning the model using human-validated code examples. This process teaches the model to prefer outputs that are not only syntactically valid but also logically correct and aligned with developer intent, reducing semantically wrong code generation.

Answer

Expand the token vocabulary

Answer

Lower the temperature to 0

Answer

Increase beam search width

Question 8

A company fine-tunes an LLM on internal support tickets. After deployment, the model hallucinates company-specific product names. What is the most effective mitigation?

Accepted Answer

Implement RAG with a verified product database. RAG (Retrieval-Augmented Generation) grounds the LLM's output in a verified product database, providing factual context that prevents hallucination of company-specific product names. Unlike fine-tuning, which only adjusts model weights and can still produce plausible but incorrect names, RAG retrieves exact records at inference time, ensuring accuracy for proprietary terminology.

Answer

Switch to a smaller model to reduce hallucination risk

Answer

Use prompt engineering to remind the model to be accurate

Answer

Fine-tune further with more ticket data

Question 9

A team wants to evaluate an LLM's performance on a text classification task. Which metric is most appropriate for a balanced dataset?

Accepted Answer

Accuracy. Accuracy is the most appropriate metric for evaluating an LLM on a text classification task with a balanced dataset because it directly measures the proportion of correctly predicted labels out of total predictions. For balanced classes, accuracy provides a reliable and intuitive performance indicator without the distortion caused by class imbalance.

Answer

BLEU score

Answer

Perplexity

Answer

ROUGE score

Question 10

An LLM-based application must comply with data privacy regulations by not memorizing personally identifiable information (PII). Which technique best reduces memorization of PII?

Accepted Answer

Train with differential privacy. Differential privacy (DP) is the correct technique because it directly limits the model's ability to memorize training data, including PII, by adding calibrated noise to the gradient updates during training. This ensures that the model's parameters do not encode specific individual records, providing a formal mathematical guarantee against memorization. Other options like model size, temperature, or training epochs do not address the root cause of memorization in the training process.

Answer

Use a larger model with more parameters

Answer

Decrease the temperature during inference

Answer

Increase the number of training epochs

Question 11

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

Accepted Answer

Quantization. Quantization (A) is correct because it stores model weights and/or activations in lower-precision formats such as INT8, INT4, or FP8 instead of FP16/FP32, directly shrinking the memory needed to hold the model and reducing bandwidth during inference. KV cache optimization (C) is correct because the key/value cache grows with sequence length and batch size and can dominate memory during long-context generation; techniques like PagedAttention, cache quantization, or sliding-window/eviction policies reduce that footprint. Increasing batch size (B) is wrong because it enlarges activation and KV-cache memory rather than reducing it. Gradient checkpointing (D) is a training-time memory-saving technique that trades extra compute for storing fewer activations, and it does not apply to standard inference. Using full precision FP32 (E) is wrong because it increases, not decreases, the memory footprint compared with FP16 or quantized formats.

Answer

Increasing batch size

Answer

Gradient checkpointing

Answer

Using full precision (FP32)

Question 12

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

Accepted Answer

Test the prompt with a variety of input examples. Option A is correct because testing the prompt with a variety of input examples is essential to validate that the prompt generalizes across different cases and to surface edge cases or ambiguities before deployment. Option C is correct because defining the desired output format and constraints (for example, JSON schema, length limits, tone, or required fields) gives the model explicit structure and makes responses consistent and machine-parseable. Option E is correct because prompt engineering is inherently iterative: you evaluate model responses, adjust wording, add examples or constraints, and re-test until the output reliably meets the goal. Options B and D are not part of prompt engineering: fine-tuning on a domain corpus changes model weights through training, and INT8 quantization is a model-compression/deployment optimization, neither of which is an essential step in designing and refining a prompt.

Answer

Fine-tune the model on a domain corpus

Answer

Quantize the model to INT8

Question 13

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 14

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 15

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.