NCP-GENL Practice Questions: Prompt Engineering Domain

Correct Answer: B

Explanation: Option B is correct because including domain-specific keywords in prompts can help the model focus on relevant topics and generate more on-topic responses. Option A is incorrect because increasing the batch size does not directly affect prompt relevance. Option C is incorrect because lowering the learning rate is more about controlling the training process rather than prompt relevance. Option D is incorrect because reducing parameter count affects model capacity, not prompt specificity. Best practice involves using prompt engineering to guide model outputs effectively.

Correct Answer: C

Explanation: Option C is correct because a minimum length constraint ensures that the model generates outputs of a certain length, reducing the likelihood of incomplete sentences. Option A is incorrect because reducing the token limit may exacerbate the issue of incomplete sentences. Option B is incorrect because a lower temperature setting increases coherence but does not directly address sentence completion. Option D is incorrect because increasing hidden layer size affects model capacity and may slow down inference. Best practice involves using prompt constraints to guide output length effectively.

Correct Answer: B

Explanation: Selectively including relevant excerpts from retrieved documents ensures that the LLM has access to pertinent information without overwhelming it with unnecessary data, enhancing response accuracy. Option B is correct because it optimizes the use of retrieved information. Option A is incorrect as including all documents can lead to information overload. Option C is incorrect because it ignores the benefits of retrieval. Option D is incorrect because it doesn't utilize the RAG system's strengths. Best practice: Use selective retrieval to enhance LLM responses in RAG systems.

Correct Answer: B

Explanation: Few-shot prompting involves providing the model with examples of desired and undesired outputs, which can help guide the model towards generating responses that are more contextually appropriate. Increasing the temperature (A) would make responses more random, not necessarily relevant. Reducing the token limit (C) might truncate responses but won't ensure relevance. RLHF (D) is more complex and typically used after initial prompt engineering techniques like few-shot prompting.

Correct Answer: B

Explanation: Option B is correct because dynamic prompt templates allow the chatbot to adapt its responses based on user interactions, providing contextually appropriate and consistent replies. Option A is incorrect as it fails to adapt to different inquiries. Option C overlooks the benefits of prompt engineering for specific tasks. Option D could lead to inefficiencies and unnecessary computational overhead. Best practice is to tailor prompts to leverage context effectively.

Correct Answer: B

Explanation: Option B is correct because using a structured prompt that specifies the format and tone can guide the model to produce content that aligns with the desired output, reducing the likelihood of inappropriate content. Option A is incorrect as post-generation filtering does not prevent inappropriate content generation. Option C is incorrect because increasing temperature increases randomness, which could lead to more inappropriate content. Option D is incorrect as a larger model may not necessarily reduce inappropriate content without proper prompt engineering. Best practice: Use structured prompts to guide LLM outputs towards desired content characteristics.

Correct Answer: B

Explanation: Option B is correct as incorporating context-specific keywords in the prompt can help the model generate more relevant responses by focusing on the right context. Option A is incorrect as increasing hidden layer size doesn't directly affect prompt relevance. Option C is incorrect because dynamic batching improves performance, not relevance. Option D is incorrect as reducing transformer layers might degrade model understanding. Best practice: Tailor prompts with context-specific details to enhance response relevance.

Correct Answer: B

Explanation: Setting a max token limit in the prompt configuration directly controls the length of the model's output, effectively managing verbosity. Increasing temperature affects randomness, not length, more examples may lead to more verbosity, and switching models doesn't address prompt-based verbosity control. NVIDIA's best practice involves configuring token limits to manage output length.

Correct Answer: B

Explanation: Option B is correct because using a length penalty during decoding discourages overly long responses, promoting conciseness. Option A is incorrect as increasing the temperature leads to more random outputs, not necessarily shorter ones. Option C is incorrect because batch size affects throughput, not response length. Option D is incorrect as fine-tuning with a smaller dataset doesn't directly address verbosity. Best practice: Apply a length penalty to control response length effectively.

Correct Answer: A

Explanation: Option A is correct because including examples of complex questions and answers in the prompt can help the model understand how to handle such queries. Option B is incorrect as increasing batch size doesn't directly affect understanding. Option C is incorrect because RLHF is a fine-tuning technique, not a prompt strategy. Option D is incorrect as LoRA is a fine-tuning method, not directly related to prompt engineering. Best practice: Use example-based prompts to guide LLM behavior in complex scenarios.