trained C. When there is a need to add learnable parameters to a Large Language Model (LLM) without task- specific training D. When the model requires continued pretraining on unlabeled data Explanation: Comprehensive and Detailed In -Depth Explanation= Soft prompting adds trainable parameters (soft prompts) to adapt an LLM without retraining its core weights, ideal for low-resource customization without task-specific data. This makes Option C correct. Option A suits fine -tuning. Option B may require more than soft prompting (e.g., domain fine -tuning). Option D describes pretraining, not soft prompting. Soft prompting is efficient for specific adaptations. : OCI 2025 Generative AI documentation likely discusses soft prompting under PEFT methods. Which is a characteristic of T -Few fine -tuning for Large Language Models (LLMs)?
Explanation: Comprehensive and Detailed In -Depth Explanation= The RAG (Retrieval -Augmented Generation) Sequence model retrieves a set of relevant documents for a query from an external knowledge base (e.g., via a vector database) and uses them collectively with the LLM to generate a cohesive, informed response. This leverages multiple sources for better context, making Option B correct. Option A describes a simpler approach (e.g., RAG Token), not Sequence. Option C is incorrect —RAG considers the full query. Option D is false —query modification isn’t standard in RAG Sequence. This method enhances response quality with diverse inputs. : OCI 2025 Generative AI documentation likely details RAG Sequence under retrieval -augmented techniques. How are documents usually evaluated in the simplest form of keyword -based search?
Explanation: Comprehensive and Detailed In -Depth Explanation= Vector databases store embeddings that preserve semantic relationships (e.g., similarity between "dog" and "puppy") via their positions in high -dimensional space. This accuracy enables LLMs to retrieve contextually relevant data, improving understanding and generation, making Option B correct. Option A (linear) is too vague and unrelated. Option C (hierarchical) applies more to relational databases. Option D (temporal) isn’t the focus —semantics drives LLM performance. Semantic accuracy is vital for meaningful outputs. : OCI 2025 Generative AI documentation likely discusses vector database accuracy under embeddings and RAG. What is the purpose of Retrievers in LangChain?
Comprehensive and Detailed In -Depth Explanation= Greedy decoding selects the word with the highest probability at each step, aiming for locally optimal choices without considering future tokens. This makes Option C correct. Option A (random selection) describes sampling, not greedy decoding. Option B (position -based) isn’t how greedy decoding works —it’s probability -driven. Option D (weighted random) aligns with top -k or top -p sampling, not greedy. Greedy decoding is fast but can lack diversity. : OCI 2025 Generative AI documentation likely explains greedy decoding under decoding strategies. What do prompt templates use for templating in language model applications?
exaggerates —top words still have impact, just less dominance. Option B is backwards — decreasing temperature sharpens, not broadens. Option D is false —temperature directly alters distribution, not speed. This controls output creativity. : OCI 2025 Generative AI documentation likely reiterates temperature effects under decoding parameters. How does the structure of vector databases differ from traditional relational databases?
➢ TOTAL QUESTIONS: 168 What is the role of temperature in the decoding process of a Large Language Model (LLM)?
