1Z0-1127-25 Fundamentals of Large Language Models — 40 Questions

Question 1

Which TWO statements about tokens in large language models are correct?

Accepted Answer

Common tokenization methods include word-based and subword-based.. Common tokenization methods in large language models include word-based tokenization (splitting text into whole words) and subword-based tokenization (like Byte-Pair Encoding or WordPiece), which handle out-of-vocabulary words and morphological variations more effectively. Subword tokenization is widely used in models like GPT and BERT to balance vocabulary size and coverage.

Answer

All tokens have the same embedding size.

Answer

Tokens are only used during training.

Answer

Tokens are always whole words.

Question 2

A company wants to build a retrieval-augmented generation (RAG) system using OCI Generative AI and a vector database. Which model type should they use to convert documents into vector embeddings?

Accepted Answer

Embedding model (e.g., cohere.embed). In a RAG system, the retrieval step requires converting documents into dense vector representations that capture semantic meaning. An embedding model like Cohere Embed is specifically designed for this task, producing fixed-length vectors that can be indexed and queried in a vector database for similarity search.

Answer

Instruct model (e.g., cohere.command)

Answer

Image generation model

Answer

Base model (e.g., cohere.base)

Question 3

A developer is reviewing the model card for an LLM on OCI Generative AI and notices it was trained on a dataset that is predominantly English. The application will serve users in multiple languages. What is the most likely limitation of using this model without additional steps?

Accepted Answer

The model may produce lower quality responses in non-English languages.. B is correct because a model trained predominantly on English data will have learned linguistic patterns, vocabulary, and cultural contexts specific to English. When applied to non-English languages, the model's internal representations and generation capabilities are less optimized, leading to lower quality responses such as grammatical errors, unnatural phrasing, or misinterpretation of nuances.

Answer

The embedding vectors will be less accurate for any language.

Answer

The model will hallucinate facts more frequently.

Answer

The context window size will be effectively reduced.

Question 4

Based on the exhibit, which model is best suited for a conversational chatbot that needs to handle multi-turn dialogues?

Accepted Answer

cohere.command. D is correct because cohere.command is specifically designed for text generation and conversational tasks, including multi-turn dialogues. It is optimized to maintain context across multiple exchanges, making it the best choice among the listed models for a chatbot that needs to handle ongoing conversations.

Answer

cohere.embed

Answer

A model with embeddings capability

Answer

cohere.base

Question 5

A developer in the GenAIDevelopers group tries to call the OCI Generative AI inference API but receives an unauthorized error. Which statement best explains the issue?

Accepted Answer

The developer lacks the required permission to 'use generative-ai-inference' to invoke the model.. The OCI Generative AI inference API requires the specific 'use generative-ai-inference' permission in an IAM policy. Without this permission, any API call to invoke a model will result in an unauthorized error, regardless of other permissions the developer might have. The error indicates the developer's group policy does not grant the necessary verb for the inference resource.

Answer

The developer does not have permission to use the generative-ai-model-family.

Answer

The policy does not allow any operations in the compartment.

Answer

The compartment name in the policy is incorrect.

Question 6

Based on the exhibit, what is the primary action the developer must take to successfully make the inference request?

Accepted Answer

Reduce max_new_tokens to 2000 to stay within the context length.. The error indicates that the combined length of the input prompt and the requested max_new_tokens exceeds the model's context length limit (e.g., 4096 tokens). Reducing max_new_tokens to 2000 ensures the total token count stays within the allowed context window, allowing the inference request to succeed. This is a fundamental constraint in transformer-based LLMs where the model cannot generate beyond its fixed context size.

Answer

Increase max_new_tokens to 5000 to get a longer response.

Answer

Ignore the error and retry the request.

Answer

Switch to a model in a different region.

Question 7

An OCI administrator wants to limit which users can invoke a specific LLM endpoint. Which resource type should be used?

Accepted Answer

IAM policies. IAM policies are the correct resource type because they define authorization rules that control which users, groups, or service principals can invoke specific OCI APIs, including the LLM inference endpoint. By writing a policy with a verb like 'inspect' or 'use' on the resource type 'ai-service-language-family' or a specific model family, the administrator can grant or deny access to the LLM endpoint. This is the standard OCI mechanism for fine-grained access control to AI services.

Answer

OCI Audit

Answer

OCI Vault

Answer

Network security groups

Question 8

A developer is using the OCI Generative AI Python SDK. They receive a 400 error 'InvalidParameter'. What is the most likely reason?

Accepted Answer

Invalid model name. The 400 'InvalidParameter' error in the OCI Generative AI Python SDK specifically indicates that the request contained a parameter that is not recognized or is malformed. The most common cause is providing an incorrect or unsupported model name, as the SDK validates model names against a predefined list of available models in the OCI Generative AI service. Other parameter errors, such as invalid temperature or top_p values, would also produce this error, but model name mismatches are the most frequent trigger.

Answer

Exceeded token limit

Answer

Network timeout

Answer

Missing API key

Question 9

A machine learning engineer is fine-tuning a model on OCI Data Science and notices that the training loss decreases but then suddenly increases. What is the most likely cause?

Accepted Answer

Reduce model size. The sudden increase in training loss after a period of decrease is often due to the model being too large for the dataset. An oversized model can create a complex loss landscape with erratic gradients, leading to divergence. Reducing the model size simplifies the architecture, resulting in more stable convergence and preventing the loss from spiking.

Answer

Add dropout

Answer

Increase batch size

Answer

Increase learning rate

Question 10

Which technique allows an LLM to be adapted to a new task with only a few examples?

Accepted Answer

Few-shot learning. Few-shot learning (A) is the correct technique because it enables an LLM to adapt to a new task by providing a small number of input-output examples directly in the prompt, without updating the model's weights. This leverages the model's in-context learning ability, where it generalizes from the provided examples to perform the task. It is distinct from fine-tuning, which requires retraining on a labeled dataset.

Answer

Fine-tuning

Answer

Pre-training

Answer

Prompt engineering

Question 11

Refer to the exhibit. A user in group GenAIGroup cannot see models in the Production compartment using OCI Generative AI. What is the most likely issue?

Accepted Answer

Resource type should be 'generative-ai-model'. The resource type in the policy statement must match the exact OCI resource type for Generative AI models, which is 'generative-ai-model'. Using a generic or incorrect resource type (e.g., 'generative-ai' or 'model') will cause the policy to not apply, preventing the user from seeing models even if they have the appropriate permissions.

Answer

Statement should be 'use' instead of 'read'

Answer

Missing 'inspect' permission

Answer

Policy syntax incorrect (missing quotes)

Question 12

Refer to the exhibit. A user deployed a custom model via OCI Data Science and registered it in the Model Catalog. They use the correct OCID but get this error. What is the most likely issue?

Accepted Answer

The model is not deployed to an endpoint. The error indicates that the user is trying to invoke a custom model via the OCI Data Science Model Catalog, but the model has not been deployed to an endpoint. In OCI, a model registered in the Model Catalog is just a metadata entry; it must be deployed to a managed endpoint (using OCI Data Science Model Deployment) to generate a serving URL that can accept inference requests. Without this deployment, any attempt to call the model using its OCID will fail because there is no active endpoint to route the request to.

Answer

The model is not fine-tuned

Answer

The compartment OCID is missing

Answer

The model is in a different region

Question 13

A developer sends this request but receives an error: "modelId not found". Which is the most likely cause?

Accepted Answer

The modelId "cohere.command" is deprecated.. The error 'modelId not found' indicates that the model identifier specified in the request does not exist or is no longer available. In Oracle Cloud Infrastructure (OCI) Generative AI service, model IDs like 'cohere.command' are deprecated and replaced with newer versions (e.g., 'cohere.command-r'). Using a deprecated modelId causes the service to reject the request because it cannot locate a valid model for that identifier.

Answer

The temperature parameter is out of range.

Answer

The compartment ID is incorrect.

Answer

The presencePenalty parameter is misspelled.

Question 14

The job fails with "InvalidParameter: trainingDatasetUri". What should the administrator check first?

Accepted Answer

Whether the bucket exists and the file is accessible.. The error 'InvalidParameter: trainingDatasetUri' indicates that the URI provided for the training dataset is invalid or inaccessible. In OCI Generative AI service, the training dataset must be stored in an OCI Object Storage bucket, and the URI must point to an existing, readable file. Therefore, the administrator should first verify that the bucket exists and the file is accessible, as this is the most common cause of this error.

Answer

Whether the compartment has sufficient budget.

Answer

Whether the model ID supports fine-tuning.

Answer

Whether the parameters JSON is correctly formatted.

Question 15

A startup needs to deploy an LLM for a simple FAQ chatbot on OCI with low latency. Which model choice is most appropriate?

Accepted Answer

Use a smaller, task-specific fine-tuned model.. For a simple FAQ chatbot requiring low latency, a smaller, task-specific fine-tuned model is most appropriate because it reduces inference time and computational overhead while maintaining high accuracy on the narrow domain of FAQs. Fine-tuning adapts the model to the specific question-answer patterns, avoiding the unnecessary complexity and latency of larger general-purpose models.

Answer

Use a medium-sized model with high precision.

Answer

Use an ensemble of models.

Answer

Use the largest available model for best quality.

Question 16

An application using OCI GenAI experiences high response times. Which change will most directly reduce latency?

Accepted Answer

Reduce the number of output tokens requested.. Reducing the number of output tokens requested directly decreases the amount of text the model must generate per request. Since latency in OCI GenAI is proportional to the total number of tokens processed (input + output), fewer output tokens mean fewer autoregressive decoding steps, which directly reduces response time.

Answer

Increase the max tokens parameter.

Answer

Switch to a fine-tuned version of the same model.

Answer

Enable batched inference.

Question 17

A developer integrates OCI GenAI into a mobile app to provide product descriptions. The responses sometimes include explanations or questions instead of the requested format. The developer is using a simple prompt: 'Describe product X.' The app expects a single paragraph. Which corrective action should the developer take?

Accepted Answer

Add a structured prompt with format instructions and an example.. Adding a structured prompt with format instructions and an example directly addresses the issue of the model producing off-format responses. By explicitly specifying the expected output (e.g., 'Provide a single paragraph description without questions or explanations') and including a few-shot example, the developer constrains the model's generation to match the desired format, leveraging prompt engineering to guide the LLM's behavior without changing model parameters or architecture.

Answer

Lower the temperature to 0 to make responses deterministic.

Answer

Increase the max tokens to allow longer responses.

Answer

Switch to a different model with better language understanding.

Question 18

A machine learning team is fine-tuning a 7B parameter Llama 2 model on a custom dataset of 10,000 documents using OCI Data Science and GPU instances. They encounter out-of-memory (OOM) errors during the fine-tuning process. They are using a batch size of 8 and a sequence length of 2048. They cannot increase the GPU memory. Which change should they prioritize to resolve the OOM?

Accepted Answer

Enable gradient accumulation with steps of 4 or more.. Gradient accumulation divides the batch into micro-batches, reducing per-step memory usage while maintaining the effective batch size. This directly addresses OOM errors without increasing GPU memory. Option B (mixed precision) can reduce memory but may not be sufficient with a batch size of 8 and sequence length 2048. Option C changes the model, which may not be necessary or feasible. Option D (reducing epochs) does not affect per-step memory and may still cause OOM.

Answer

Use mixed precision training (FP16).

Answer

Reduce the model size by using a 3B parameter version.

Answer

Decrease the number of training epochs.

Question 19

An enterprise wants to deploy a large language model for processing sensitive internal documents. They must ensure that data does not leave their OCI tenancy. Which OCI GenAI deployment option meets this requirement?

Accepted Answer

Dedicated AI Cluster with on-demand model hosting. A Dedicated AI Cluster with on-demand model hosting provides a fully isolated, single-tenant infrastructure within your OCI tenancy. This ensures that all data processing and model inference occur on dedicated hardware that never leaves your tenancy, meeting strict data residency and security requirements for sensitive internal documents.

Answer

Using a third-party model via OCI Marketplace

Answer

Accessing models through OCI Console only

Answer

Using the OCI GenAI API with the default endpoint

Question 20

A research team is using OCI Data Science and OCI GenAI to build a multilingual chatbot for customer service. They have training data in English, Spanish, and French. The model currently struggles with code-switching—users often mix languages in a single query (e.g., 'Quiero cancel my order'), and the model responds inconsistently, sometimes in English, sometimes mixing incorrectly. The team wants to improve performance on code-switching while maintaining fluency in each language. They have limited compute resources and cannot deploy separate models per language. Which approach should they take?

Accepted Answer

Fine-tune a multilingual model on a combined dataset that includes code-switching examples.. Fine-tuning a single multilingual model (e.g., Cohere Command with multilingual support) on a combined dataset that includes code-switching examples directly teaches the model to handle mixed-language inputs, improving performance while maintaining fluency in each language without requiring separate models. Option A is incorrect because training separate models per language would prevent any code-switching capability and increase resource usage. Option C is incorrect because language detection and translation loses the original code-switching context and adds latency. Option D is incorrect because multilingual embeddings improve retrieval but do not directly improve generation fluency for code-switching.

Answer

Train separate fine-tuned models for each language and route queries based on detected language.

Answer

Use language detection to route the query to a specific language model, then translate the response.

Answer

Use a multilingual embedding model for retrieval to improve context understanding.