Question bank
Oracle Cloud Infrastructure 2026 Generative AI Professional
40 Oracle Cloud Infrastructure 2026 Generative AI Professional questions with the correct answer and a written explanation, organised across 4 exam domains.
- 40 questions
- 4 domains
- Foundation level
certilumQuestion index
Jump straight to any question. Answers and explanations are inline.
Fundamentals of Large Language Models
8 questions in this domain
- Question 1Single Choice
What is a large language model (LLM)?
- AA neural network trained on massive text data to understand and generate languageCorrect answer
- BA database that stores prewritten responses
- CA list of hand-coded grammatical rules
- DA program that only recognizes images
Why this is the answer
LLMs are neural networks trained on massive text corpora. They learn statistical patterns in language and can generate coherent text. Databases, rule systems, and image classifiers are not LLMs.
- Question 2Multiple Select
Which of the following are core characteristics of large language models? Select all that apply.
- AThey are pretrained on large collections of text.Correct answer
- BThey are often based on transformer architectures.Correct answer
- CThey generate text by predicting likely next tokens.Correct answer
- DThey require a labeled dataset for every new task.
Why this is the answer
LLMs are pretrained on large text corpora, usually use transformer architectures, and generate text by predicting next tokens. They support transfer learning and do not require labeled data for every new task.
- Question 3Interactive
Review the Python code in the code block. What is the primary task of the loaded pipeline in this script?
- AGenerate a continuation of the prompt textCorrect answer
- BClassify the sentiment of the prompt
- CTranslate the prompt into another language
- DSummarize the prompt in one sentence
Why this is the answer
The pipeline is created with the text-generation task. It receives the prompt and returns a model-generated continuation. This is not classification, translation, or summarization.
- Question 4Single Choice
What does the self-attention mechanism allow a transformer model to do?
- ADetermine how relevant each token is to other tokensCorrect answer
- BStore all text in a key-value database
- CApply punctuation rules to every sentence
- DConvert text into a fixed-size image
Why this is the answer
Self-attention computes relationships between all tokens in a sequence. It tells the model how much each token should influence the representation of another token.
- Question 5Multiple Select
Which statements about tokenization in LLMs are correct? Select all that apply.
- AA tokenizer splits text into words or subword units.Correct answer
- BThe tokenizer maps tokens to numeric IDs before model input.Correct answer
- CLLMs have a fixed vocabulary size defined by the tokenizer.Correct answer
- DEach token always represents exactly one character.
Why this is the answer
Tokenizers convert text into word or subword tokens and then to numeric IDs. Each model has a fixed vocabulary. A token is not limited to one character.
- Question 6Interactive
When the code block runs, which statements are true about the text generation logic? Select all that apply.
- AThe generation loop repeats for max_tokens iterations.Correct answer
- BAt each step, argmax selects the token with the highest predicted score.Correct answer
- CThis strategy is equivalent to random top-k sampling.
- DThe decoded result contains the original prompt and the generated tokens.Correct answer
Why this is the answer
The loop executes max_tokens times, uses argmax over the final logits to choose the highest probability token, and decodes the concatenated prompt and generated tokens. It is deterministic, not random sampling.
- Question 7Single Choice
What is fine-tuning in the context of LLMs?
- AAdapting a pretrained model to a task using labeled examplesCorrect answer
- BSelecting the best prompt without updating model weights
- CTraining a new model from randomly initialized weights
- DRemoving unnecessary transformer layers
Why this is the answer
Fine-tuning takes a pretrained model and continues training on labeled, task-specific data. It updates the model weights to improve performance on the target task.
- Question 8Multiple Select
Which statements about transformer-based LLMs are correct? Select all that apply.
- ASelf-attention allows modeling relationships between tokens in parallel.Correct answer
- BPositional encodings provide information about token order.Correct answer
- CTransformers do not require recurrent sequential processing.Correct answer
- DTransformers always process tokens one by one in strict sequence.
Why this is the answer
Transformers process entire sequences in parallel using self-attention. Positional encodings preserve word order, and the architecture avoids recurrent sequential processing.
Implement RAG using OCI Generative AI Service
8 questions in this domain
- Question 9Single Choice
When a large language model trains on text data, what core statistical relationship does it primarily learn?
- ASQL schema for storing relational tables
- BPhysical data center geographic coordinates
- CProbability of the next token given previous tokensCorrect answer
- DNetwork packet routing table settings
Why this is the answer
An LLM learns statistical relationships between tokens, estimating the most probable next token from prior tokens. This next-token prediction underlies its text generation.
- Question 10Multiple Select
Which capabilities of a large language model directly contribute to answering a query using retrieved context in RAG?
- AGenerating coherent text conditioned on contextCorrect answer
- BPerforming low-level disk storage compression
- CLearning semantic patterns in natural languageCorrect answer
- DExecuting embedding index scans automatically
Why this is the answer
An LLM contributes language generation and semantic understanding, while vector index scans and storage compression are handled by supporting infrastructure, not the model.
- Question 11Interactive
Review the code sample and identify the role of the object created before any text generation request is made.
- ACreates a virtual cloud network
- BManages API calls to OCI Generative AICorrect answer
- CConfigures an object storage bucket
- DDeletes model fine-tuning jobs
Why this is the answer
The code imports GenerativeAiInferenceClient and constructs a client from OCI config, enabling calls to the OCI Generative AI inference service.
- Question 12Single Choice
When text is provided to a large language model, what is the term used to describe a single text unit such as a subword?
- AAuthentication key for the inference API
- BContainer instance in a compute cluster
- CSmallest text unit processed by the modelCorrect answer
- DSQL statement used to query the database
Why this is the answer
A token is a discrete text unit, usually a word, subword, or character segment, created by a tokenizer before the model computes token probabilities.
- Question 13Multiple Select
Which statements correctly describe how the temperature parameter influences text output from a large language model?
- AControls how random the generated text isCorrect answer
- BAlways increases generated response length
- CSets the maximum input context size
- DLower values generally produce more deterministic outputsCorrect answer
Why this is the answer
Temperature affects the token probability distribution used during sampling. Lower values make outputs more focused and deterministic, while higher values increase randomness; it does not control length.
- Question 14Interactive
Which of the following best describes the main goal of the code sample in preparing content for RAG retrieval?
- ACreates a backup of the source files
- BEmploys embeddings to enable semantic searchCorrect answer
- CEncrypts documents with a symmetric key
- DCompiles code into a deployable artifact
Why this is the answer
Embedding models convert documents into vector representations. Storing these vectors in an index enables the retriever to find semantically relevant passages for a user query.
- Question 15Single Choice
In a retrieval-augmented generation system that uses LLM embeddings, what type of store is most commonly used to search relevant passages quickly?
- AObject storage bucket
- BDomain name server zone
- CNetwork address translation table
- DVector database or vector indexCorrect answer
Why this is the answer
A vector database or vector index stores embeddings and performs similarity search, allowing the RAG retrieval step to return passages most relevant to the query.
- Question 16Multiple Select
Why can retrieval-augmented generation improve responses when the model has no knowledge of very recent documents?
- AExpands model parameter count for each request
- BProvide relevant external context during inferenceCorrect answer
- CEliminates the need for prompt design
- DGround the answer in retrieved reference passagesCorrect answer
Why this is the answer
Retrieval adds external, current knowledge as context. This grounds the generated answer and helps reduce hallucinations, but it does not change model parameter counts or remove prompt design.
Using OCI Generative AI RAG Agents Service
8 questions in this domain
- Question 17Single Choice
What is the fundamental role of a large language model (LLM) inside an OCI Generative AI RAG agent?
- AIt generates the final answer in natural language based on retrieved context.Correct answer
- BIt creates vector indexes for all source documents.
- CIt routes API traffic between the agent and OCI services.
- DIt enforces IAM policies on the knowledge base.
Why this is the answer
The LLM is the generator in a RAG pipeline: it receives the user prompt plus retrieved documents and produces a natural-language answer. Indexing, routing, and security are separate concerns.
- Question 18Multiple Select
Which elements are necessary for the retrieval-augmented generation (RAG) pattern to work with OCI Generative AI Agents?
- AA source of knowledge, such as PDFs or web pages, indexed as a knowledge base.Correct answer
- BA retriever that searches the indexed content for relevant textual chunks.Correct answer
- CA generative LLM that composes an answer from the retrieved chunks.Correct answer
- DA physical database server deployed in the customer's data center.
Why this is the answer
RAG must combine an indexed knowledge base, a retrieval step, and a generator LLM. No on-premises physical server is required because OCI provides managed services.
- Question 19Interactive
Review the inference parameters in the code block. How should you adjust the configuration to obtain more consistent, less random answers from the OCI Generative AI model?
- ALower the temperature from 0.7 to 0.1.Correct answer
- BRaise the temperature from 0.7 to 1.5.
- CChange max_tokens from 200 to 50.
- DSet top_p to 1.0.
Why this is the answer
Sampling parameters, especially temperature, control randomness. Lowering temperature sharpens the model's token distribution and makes responses more consistent.
- Question 20Single Choice
In an OCI RAG knowledge base, what representation is used to search for semantically similar text?
- ANumerical vector embeddings of text chunks.Correct answer
- BRaw PDF files as originally uploaded.
- CSQL queries from user sessions.
- DTraining weights from the foundation model.
Why this is the answer
Documents are chunked, and each chunk is converted into an embedding vector. The retriever measures embedding similarity to find semantically related passages.
- Question 21Multiple Select
Which statements about the knowledge base in an OCI Generative AI RAG agent are correct?
- ADocuments are normally broken into smaller chunks before indexing.Correct answer
- BEmbedding models generate vector representations for the text.Correct answer
- CSearch results are ranked based on semantic similarity.Correct answer
- DUser queries must contain the exact keywords from the original document.
Why this is the answer
The RAG knowledge base pipeline uses chunking, embeddings, and similarity search. Retrieval is semantic, so exact keyword matching is unnecessary.
- Question 22Interactive
You run the OCI Generative AI Agents CLI command shown in the code block. Which conditions must hold for the agent to use RAG against your enterprise knowledge base?
- AThe agent endpoint must be associated with a configured knowledge base.Correct answer
- BThe session ID must correspond to an existing session that is valid for the agent endpoint.Correct answer
- CThe user's message must match the exact title of a stored document.
- DThe agent endpoint must be publicly accessible without authentication.
Why this is the answer
RAG relies on the endpoint's linked knowledge base and session context. It does not require exact title matching, and the endpoint must not be unauthenticated.
- Question 23Single Choice
Which mechanism allows an OCI Generative AI RAG agent to show the supporting sources for generated answers?
- ASource citations or reference metadata.Correct answer
- BObject storage pre-authenticated URLs.
- COCI Audit log events.
- DCustom model training metrics.
Why this is the answer
When answers are grounded, the response can include citations that identify the source document or location and allow users to verify the information.
- Question 24Multiple Select
Which are advantages of using RAG in OCI Generative AI Agents instead of using a general LLM alone?
- AYou can update the agent's knowledge by refreshing the underlying documents.Correct answer
- BAnswers can be grounded in private company data.Correct answer
- CThe model will never produce hallucinations.
- DThe response can include references to the source documents.Correct answer
Why this is the answer
RAG supports document refresh, grounding in private data, and optional citations. It reduces hallucinations but does not remove all possible generation errors.
Using OCI Generative AI Service
16 questions in this domain
- Question 25Single Choice
What is the main function of a large language model in generative AI?
- AText predictor trained on massive dataCorrect answer
- BApplication for entering numerical data
- CRelational database management system
- DRouting protocol for cloud networks
Why this is the answer
A large language model learns language patterns from massive text corpora to generate coherent and contextually relevant text.
- Question 26Multiple Select
Which two components are considered foundational to how large language models process text?
- ATokenizationCorrect answer
- BAttention mechanismCorrect answer
- CIndexing
- DCompression
Why this is the answer
LLMs process text into tokens and use attention mechanisms to understand the relationships among those tokens during generation.
- Question 27Interactive
What is the action performed by the code block when the client method is called?
- ACalling a model to generate textCorrect answer
- BCreating an object storage bucket
- CDeleting a compute instance
- DProvisioning a load balancer
Why this is the answer
The code sends a prompt to the model and receives generated text as the response.
- Question 28Single Choice
What is the meaning of token when used by large language models during text processing?
- ASubword unit of textCorrect answer
- BCPU cache memory
- CRow in a SQL table
- DNetwork packet header
Why this is the answer
A token is a basic text unit, typically a word or subword, used as the model's smallest processing element.
- Question 29Multiple Select
Which two model parameters control the randomness of generated text during inference?
- ATemperatureCorrect answer
- BTop-pCorrect answer
- CThread count
- DBatch size
Why this is the answer
Temperature and top-p control sampling randomness, making generated text more diverse or more focused depending on their values.
- Question 30Interactive
What is the intended outcome of the function call shown in the code block?
- AList available language modelsCorrect answer
- BGenerate a text response
- CStore result in a database
- DConfigure an identity policy
Why this is the answer
The code calls the model listing operation to show accessible language models in the compartment.
- Question 31Single Choice
What does the context window of a large language model represent in a generation task?
- AMaximum tokens considered for generationCorrect answer
- BAmount of free memory on a GPU
- CNumber of layers in the neural network
- DTime required to train a model
Why this is the answer
The context window defines the maximum number of tokens the model can consider when producing a response.
- Question 32Multiple Select
Which two common tasks can a generative AI text model effectively perform?
- AText summarizationCorrect answer
- BLanguage translationCorrect answer
- CDisk partitioning
- DSSH key generation
Why this is the answer
LLMs are commonly used for summarization and translation, where they generate fluent and context-aware text.
- Question 33Interactive
In the code block, what is the result of setting the temperature parameter to zero?
- AMore deterministic responseCorrect answer
- BFully random response
- CGuaranteed longer output
- DNo model inference
Why this is the answer
A temperature of zero makes sampling more deterministic, leading to more predictable model output.
- Question 34Single Choice
What is meant by hallucination when a large language model produces an answer?
- APlausible but incorrect informationCorrect answer
- BText copied exactly from a database
- CIntentional security alert
- DNetwork timeout error
Why this is the answer
A hallucination is generated content that appears fluent and credible but is not grounded in facts.
- Question 35Multiple Select
Which two methods can reduce hallucination by providing the model with relevant external information?
- ARetrieval augmented generationCorrect answer
- BGrounding with source documentsCorrect answer
- CRaising temperature to maximum
- DRemoving all context
Why this is the answer
RAG and grounding give the model relevant source context, reducing unsupported or invented answers.
- Question 36Interactive
What is the main purpose of the embedding operation shown in the code block?
- AConvert text into vector embeddingsCorrect answer
- BGenerate new tokens from a prompt
- CBackup files to object storage
- DCreate an IAM group
Why this is the answer
Embedding transforms text into dense numeric vectors used for semantic similarity and retrieval.
- Question 37Single Choice
What is a prompt in the context of interacting with a generative AI model?
- AInput text given to a modelCorrect answer
- BOutput file produced by a model
- CNetwork connection to a model
- DTraining dataset for a model
Why this is the answer
A prompt is the input instruction or text given to a model to produce an output.
- Question 38Multiple Select
Which two service elements are managed by OCI Generative AI for model deployment and use?
- AModel catalogCorrect answer
- BDedicated AI clustersCorrect answer
- COn-premises mainframes
- DLegacy database servers
Why this is the answer
OCI Generative AI exposes model catalogs and dedicated AI clusters for deployment and inference.
- Question 39Interactive
What technique does the code block illustrate when the prompt includes retrieved context?
- ARetrieval augmented generationCorrect answer
- BModel quantization
- CToken pruning
- DBatch normalization
Why this is the answer
The code retrieves relevant context and includes it in the prompt, following RAG principles.
- Question 40Single Choice
What is an embedding in the context of natural language processing and LLMs?
- ADense numeric vector for textCorrect answer
- BEncrypted text file
- CRelational database row
- DVersion control commit
Why this is the answer
An embedding is a dense numeric vector that captures the semantic meaning of text.
Frequently asked questions
Everything about the Oracle Cloud Infrastructure 2026 Generative AI Professional question bank and how to use it.