Task Statement 3.3: Describe the Training and Fine-Tuning Process for FMs. AIF-C01 Practice Questions (Page 8)

Questions 36–40

1. 36foundation · easy

   What is a key benefit of transfer learning?

   • AIt makes the model smaller and faster
   • BIt guarantees higher accuracy than training from scratch
   • CIt eliminates the need for any labeled data
   • DIt reduces the need for extensive training data and compute for the new task
   Reveal Solution Discussion Select an answer first
2. 37application · medium

   A healthcare startup wants to fine-tune a pre-trained foundation model to generate patient-friendly summaries of medical discharge notes. They have collected 500 discharge notes from a single hospital, but the notes contain many typographical errors and inconsistent formatting. The team plans to clean the data before fine-tuning. Which additional data-related consideration is MOST important for the model to generalize well to patients from different regions?

   • ALabel each note with the patient's diagnosis to provide additional supervision signals.
   • BRemove all notes that contain any typographical errors to ensure perfect input quality.
   • CEnsure the dataset includes notes from multiple hospitals and diverse patient demographics.
   • DIncrease the dataset size to at least 10,000 notes to meet a minimum threshold for fine-tuning.
   Reveal Solution Discussion Select an answer first
4. 38foundation · easy

   What is a risk of using fine-tuning data that is not representative of the target population?

   • AThe model will definitely overfit the training data
   • BThe model will require more computational resources
   • CThe model may become larger than necessary
   • DThe model may perform poorly on underrepresented groups
   Reveal Solution Discussion Select an answer first
5. 39application · medium

   A national bank is fine-tuning a foundation model to answer customer questions about loan products. The fine-tuning dataset was created using only responses from customers in urban areas with high incomes. The bank plans to deploy the model to all customers, including those in rural areas and with lower incomes. What is the MOST likely problem with this approach?

   • AThe model will not be able to understand the language used by rural customers.
   • BThe model will require continuous pre-training on the urban dataset before fine-tuning.
   • CThe model may not generalize well to the broader customer population, leading to biased or inaccurate responses.
   • DThe model will be too large to deploy to all customers.
   Reveal Solution Discussion Select an answer first
6. 40application · medium

   A mobile app company has a large, high-performing foundation model that powers a real-time language translation feature. The model is too large and slow to run on users' devices, and the company wants to deploy a smaller model that runs efficiently on-device while maintaining most of the translation quality. They have access to the large model's outputs on a diverse set of translation tasks. Which approach should they use?

   • ATrain a smaller model to mimic the large model's outputs on the diverse translation dataset.
   • BFine-tune the large model on a smaller dataset of common phrases to reduce its size.
   • CUse continuous pre-training on a smaller corpus to make the model more efficient.
   • DApply RLHF to the large model to make its outputs more concise.
   Reveal Solution Discussion Select an answer first