Using the OCI Generative AI Service
Free Oracle Cloud Infrastructure 2025 Generative AI Professional practice — 6 questions on Using the OCI Generative AI Service, with explanations. No sign-up. Full 12-question mixed test →
A media company fine-tunes a custom text-generation model using a Large Cohere dedicated fine-tuning cluster in OCI Generative AI Service. The fine-tuning job finishes in 6 hours, but the company must serve inference traffic from the resulting custom model for the next 90 days. Which approach minimizes cost while meeting the requirement?
Fine-tuning clusters and hosting clusters are separate resource and billing types in OCI Generative AI Service. A fine-tuning cluster is meant only for training and should be terminated once the job completes; a hosting cluster is created to actually serve inference for the tuned model, and only the hosting cluster needs to run for the 90-day serving window.
According to OCI documentation, what is the minimum billing commitment period when you create a dedicated AI cluster (hosting) in OCI Generative AI Service?
OCI dedicated AI clusters (hosting) require a minimum commitment of 744 unit-hours, equivalent to running continuously for 31 days, even if the cluster is used for less time or deleted early.
You need to fine-tune a cohere.command model on a small custom dataset (fewer than 100 examples) in OCI Generative AI Service, while minimizing training time and compute cost. Which fine-tuning method should you select?
T-Few is OCI's parameter-efficient fine-tuning method, designed specifically for smaller custom datasets. It updates a small subset of model parameters, resulting in significantly faster and cheaper training compared to full fine-tuning, which is exactly suited to a sub-100-example dataset.
In the OCI Generative AI console playground, what happens when you enable the 'Content Moderation' toggle for a chat model?
The Content Moderation feature in OCI Generative AI Service inspects both the incoming prompt and the outgoing model response, allowing harmful or toxic content to be flagged/blocked on either side of the interaction, not just one.
A serverless OCI Function needs to call the OCI Generative AI chat endpoint without embedding user credentials or API keys anywhere in its code. Which IAM configuration correctly grants the required access?
OCI Functions can use resource principals: matching the function to a dynamic group and granting that dynamic group a policy to use the generative-ai-family resource type provides secure, credential-free access to the Generative AI service at runtime.
Your team is building a RAG search index over millions of documents. Vector storage cost is a primary concern, and the corpus includes documents in multiple languages. Which embedding model configuration BEST balances these requirements?
The multilingual light embedding model produces smaller 384-dimension vectors instead of 1024, substantially reducing vector storage footprint while still supporting multiple languages, which directly addresses both the storage-cost concern and the multilingual requirement.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →