Free 1z0-1127-25: Oracle Cloud Infrastructure 2025 Generative AI Professional Exam Questions and Answers

A developer stores 5 million chunk embeddings in a table with a VECTOR column in Oracle Database 23ai. Exact similarity queries now take too long, and a small loss of accuracy is acceptable. What should the developer do to speed up the searches?

• AA. Convert the VECTOR column to a CLOB and create an Oracle Text index on it
• BB. Create a B-tree index on the VECTOR column that stores the embeddings
• CC. Partition the table by hash on the chunk identifier column
• DD. Create a vector index, such as an HNSW or IVF index, on the VECTOR column

Which OCI capability controls which users can create and manage dedicated AI clusters and endpoints in the OCI Generative AI service?

• AA. Cloud Guard detector recipes applied to the compartment
• BB. Security lists defined on the VCN that hosts the endpoint
• CC. Network security groups attached to the cluster subnet
• DD. IAM policies written for the Generative AI resource types

A team is choosing a model architecture for a chatbot that generates free-form replies. Which two characteristics of decoder-only transformer models support this choice? (Choose two.)

Select 2 answers.

• AA. They use masked self-attention, so each position attends only to earlier positions in the sequence
• BB. They generate output autoregressively, feeding each predicted token back in to predict the next one
• CC. They require a separate encoder stack to build a full representation of the input before any output is produced
• DD. They are pretrained by predicting randomly masked tokens using context from both directions

A data scientist stores a training dataset in an Object Storage bucket to fine-tune a base model in the OCI Generative AI service. Which file format must the dataset use?

• AA. JSONL file in which each line holds one prompt and completion pair
• BB. CSV file with one column for the input text and one for the expected output
• CC. JSON array file that contains a list of prompt and completion objects
• DD. Parquet file with a prompt column and a completion column

A team fine-tuned a base model on a fine-tuning dedicated AI cluster and now wants to serve the custom model with predictable throughput. What must the team do?

• AA. Change the type of the fine-tuning cluster to hosting and reuse it for inference
• BB. Delete the fine-tuning cluster and call the custom model in on-demand mode
• CC. Export the model weights to a bucket and deploy them on a GPU Compute instance
• DD. Create a hosting dedicated AI cluster and create an endpoint for the custom model on it

A startup wants to test prompts against a pretrained chat model for a few hours each week. It wants to pay only for usage and provision no infrastructure. Which approach meets these needs?

• AA. Provision a GPU Compute instance and deploy an open-source model on it
• BB. Create a hosting dedicated AI cluster and an endpoint for the pretrained model
• CC. Call the pretrained model in on-demand mode through the playground or API
• DD. Create a fine-tuning dedicated AI cluster and train a custom model first

Which two statements about few-shot prompting are true? (Choose two.)

Select 2 answers.

• AA. A labeled training dataset must be uploaded to Object Storage before the prompt can be submitted
• BB. The model adapts its response to the pattern shown in the examples without any change to its parameters
• CC. Examples of the desired inputs and outputs are included in the prompt text that is sent to the model
• DD. The model weights are updated on a dedicated AI cluster each time a new example is added to the prompt

A developer has split a set of policy documents into chunks and embedded them with OCI Generative AI. The developer now wants to use LangChain to persist the vectors in Oracle Database 23ai and run similarity searches against them. Which LangChain component should the developer use?

• AA. OracleVS
• BB. OCIGenAIEmbeddings
• CC. RecursiveCharacterTextSplitter
• DD. ChatOCIGenAI

What is the primary purpose of the playground in the OCI Generative AI service?

• AA. Provisioning dedicated AI clusters by selecting a GPU shape and a number of unit hours
• BB. Storing training datasets and keeping a version history of fine-tuned custom models
• CC. Testing models interactively in the Console while adjusting parameters such as temperature
• DD. Tracking token consumption and billing charges for each endpoint in a tenancy

An ingestion job for a knowledge base data source fails with an authorization error when it tries to read files from a private Object Storage bucket in the compartment DocsProd. What should the administrator do to resolve the error?

• AA. Write an IAM policy that lets the Generative AI Agents service read objects in the bucket's compartment
• BB. Add the administrator who created the knowledge base to a group that can manage all tenancy resources
• CC. Change the bucket visibility to public so that the ingestion job can read the files anonymously
• DD. Move the files to a bucket in the root compartment and recreate the data source from that bucket