Google Professional Machine Learning Engineer Question 277
Select 3Google Cloud PlatformYou are tasked with fine-tuning a foundational large language model (LLM) using Vertex AI Model Garden to generate domain-specific customer support responses. The dataset contains labeled examples of past customer queries and the corresponding responses. To optimize the fine-tuning process, which steps should you follow to ensure the model is effectively fine-tuned for this use case?
- A
Preprocess the dataset to ensure it is clean, tokenized, and formatted to align with the model's input requirements.
- B
Directly deploy the foundational model without fine-tuning and rely solely on prompt engineering to achieve domain-specific behavior.
- C
Use Vertex AI's training pipelines to fine-tune the model with the labeled dataset, specifying appropriate hyperparameters such as learning rate and batch size.
- D
Evaluate the fine-tuned model using a validation set and adjust hyperparameters if the performance does not meet expectations.
- E
Disable pre-trained weights of the foundational model during fine-tuning to ensure the model learns only from the new dataset.
Show answer and explanation
Correct answers: A, C, D
Explanation
Fine-tuning a foundational model on Vertex AI requires cleaning and preparing the dataset, leveraging training pipelines to customize the training process, and validating the model's performance to ensure it meets the use case requirements. Pre-trained weights should be retained to build on the model's existing knowledge, and prompt engineering alone is insufficient for training the model on domain-specific data.
- A. Correct.
Preprocessing the dataset is essential to ensure it matches the foundational model's input format, which helps prevent errors during fine-tuning and ensures optimal learning.
- B. Incorrect.
While prompt engineering can sometimes achieve domain-specific behavior, it does not train the model on the domain-specific data, which is necessary for your use case.
- C. Correct.
Using Vertex AI's training pipelines for fine-tuning allows you to customize the training process and optimize it for your dataset, including setting hyperparameters like learning rate and batch size.
- D. Correct.
Evaluating the fine-tuned model using a validation set ensures that the model performs well on unseen data, and adjusting hyperparameters can help improve performance if necessary.
- E. Incorrect.
Disabling pre-trained weights is not recommended for fine-tuning foundational models, as it would discard the knowledge the model has already learned from large-scale pre-training, which is crucial for adapting to new tasks efficiently.