Oracle Cloud Infrastructure 2025 Generative AI Professional Free Sample Questions

The temperature parameter controls the randomness of the output. A lower temperature (e.g., 0.2) makes the model more deterministic and focused, while a higher temperature (e.g., 0.9) increases randomness, leading to more creative and diverse outputs. This is the correct parameter to adjust for the desired outcome.

Chain-of-thought (CoT) prompting is specifically designed for problems requiring multi-step reasoning. It involves providing examples where the intermediate reasoning steps are explicitly written out. This guides the model to 'think step-by-step,' breaking down the complex problem into smaller, manageable parts, which significantly improves accuracy on arithmetic and logical reasoning tasks.

A Dedicated AI Cluster is a managed set of dedicated GPU resources. Its primary purpose is to provide the necessary computational power for the intensive tasks of fine-tuning a base model with custom data and for hosting custom model endpoints for dedicated, high-performance inference.

This statement is true. The self-attention mechanism in a transformer processes all tokens in parallel and has no inherent sense of their order. Positional encodings are vectors that are added to the token embeddings to give the model explicit information about the position of each token in the sequence.

The OCI Generative AI Agents service maintains dependencies between resources. A Knowledge Base cannot be deleted if one or more Agents are actively using it. This is a protective measure to prevent breaking deployed applications. The correct procedure is to first delete any Agents that rely on the Knowledge Base, or edit them to use a different one, before attempting to delete the Knowledge Base itself.

OCI Identity and Access Management (IAM) is the primary mechanism for controlling who can do what with OCI resources. Creating granular policies to allow or deny actions like creating fine-tuning jobs, deploying endpoints, or invoking models is fundamental to securing the service.

The OCI Audit service automatically records all API calls made to OCI services in a tenancy. This is crucial for monitoring who is using the Generative AI service, what actions they are performing, and when. It provides an immutable log for security analysis and compliance.

This is the correct sequence for the RAG ingestion pipeline. First, the raw documents are loaded (e.g., using a PDF loader). Second, because LLM embedding models have context limits, the loaded text is split into smaller, manageable chunks. Finally, each of these text chunks is passed to an embedding model to create its corresponding vector representation, which can then be stored.

This is the core advantage of RAG for this use case. With RAG, updating the system's knowledge is as simple as updating the documents in the vector store, a process that can be easily automated. In contrast, fine-tuning embeds knowledge into the model's weights, requiring a costly and time-consuming retraining process every time the documentation changes. For frequently updated knowledge, RAG is far more practical and scalable.