AI-102 Question 490
Select 3You are developing a customer support chatbot using an Azure OpenAI model. The pre-trained model is effective but does not fully understand your organization's specific terminology and domain-specific questions. You decide to fine-tune the model. Which of the following actions are necessary when fine-tuning an Azure OpenAI model?
- A
Prepare a dataset in JSONL format containing your domain-specific data.
- B
Use the Azure OpenAI Studio to upload the dataset and initiate the fine-tuning process.
- C
Manually modify the pre-trained model's architecture to add new layers for domain-specific tasks.
- D
Validate the fine-tuned model by testing it against a set of representative queries.
- E
Directly deploy the fine-tuned model to production without any validation.
Show answer and explanation
Correct answers: A, B, D
Explanation
Fine-tuning an Azure OpenAI model involves uploading a properly structured dataset in JSONL format, using Azure OpenAI Studio to initiate the fine-tuning process, and validating the model's performance before deployment. Modifying the model's architecture is not part of the fine-tuning process, and skipping validation can result in poor performance.
- A. Correct.
Correct: Azure OpenAI models require a JSONL (JSON Lines) formatted dataset for fine-tuning, ensuring the data is properly labeled and structured.
- B. Correct.
Correct: Azure OpenAI Studio provides tools for uploading datasets and managing the fine-tuning process.
- C. Incorrect.
Incorrect: Fine-tuning does not involve modifying the architecture of the pre-trained model. The model's architecture remains fixed, and only its weights are adjusted based on the fine-tuning data.
- D. Correct.
Correct: Testing the fine-tuned model with representative queries is essential to ensure it performs as expected and aligns with business requirements.
- E. Incorrect.
Incorrect: Deploying the fine-tuned model without validation is not recommended, as it may lead to suboptimal performance or unexpected behavior in production.