Task Statement 3.3: Describe the Training and Fine-Tuning Process for FMs. AIF-C01 Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A small e-commerce company wants to build a model that classifies product reviews as positive, neutral, or negative. They only have 300 labeled reviews, which is not enough to train a model from scratch. They decide to use a pre-trained foundation model that has learned general language patterns from a massive corpus. Which concept BEST describes why this approach can work with such a small dataset?

   • AData augmentation, because the 300 reviews can be synthetically expanded to create more training examples.
   • BContinuous pre-training, because the model can be further trained on the 300 reviews to learn the task.
   • CTransfer learning, because the pre-trained model already understands language structure and sentiment-related concepts.
   • DKnowledge distillation, because the pre-trained model can be compressed to fit the small dataset.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A research lab is building a new foundation model from scratch for a specialized scientific domain. They have collected a massive corpus of unlabeled scientific papers. Their goal is to create a model that understands general language and scientific concepts, which they will later fine-tune for specific tasks. What is the FIRST training stage they should perform?

   • AKnowledge distillation from a smaller existing model.
   • BFine-tuning on a small labeled dataset of scientific tasks.
   • CPre-training on the large unlabeled corpus to learn general language patterns and scientific knowledge.
   • DRLHF using expert scientists to rank the model's outputs.
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   What is the main goal of fine-tuning a pre-trained foundation model?

   • ATo further train the model on domain-specific unlabeled data
   • BTo adapt the model to a specific task or domain using a smaller labeled dataset
   • CTo reduce the model's size while preserving performance
   • DTo train the model from scratch on a completely new dataset
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   Which type of data is most commonly used in continuous pre-training?

   • ALabeled examples for a specific task
   • BInstruction-response pairs
   • CHuman preference rankings
   • DDomain-specific unlabeled text
   Reveal Solution Discussion Select an answer first
6. 25expert · hard

   A multinational corporation is fine-tuning a foundation model for internal HR support. The training data includes employee records from multiple countries, and the legal team has determined that data from the European Union must not leave the EU region. The company uses a cloud provider with regions in the EU and the US. What is the MOST appropriate data governance approach?

   • ADe-identify all employee data before fine-tuning so it can be processed anywhere.
   • BObtain consent from all EU employees to process their data in the US.
   • CStore and process the EU employee data only in an EU region, and use separate fine-tuning jobs for EU and non-EU data.
   • DUse a single fine-tuning job in the US region and rely on encryption to protect the data.
   Reveal Solution Discussion Select an answer first