AIF-C01 Study Platform

Lesson 9 Quiz — Fine-Tuning & Training Methods (10 easy + 10 hard)

20 questions.

Q1easymcq

Training a model from scratch on a massive, general dataset is called:

Q2easymcq

Taking an already-pre-trained FM and further training it on a smaller, task-specific labeled dataset is called:

Q3easymcq

Training a smaller model to mimic a larger model's behavior is called:

Q4easymcq

Fine-tuning on (instruction, ideal response) pairs to improve direction-following is called:

Q5easymcq

What is RLHF?

Q6easymcq

Further training a model on large amounts of unlabeled, domain-specific data (without task labels) is called:

Q7easymcq

Specializing a general model for a specific industry's vocabulary (e.g., legal or medical) is called:

Q8easymcq

Which data-prep concern refers to whether you're actually allowed to train on a given dataset?

Q9easymcq

Training data that doesn't reflect the real-world distribution of inputs the model will see is a failure of:

Q10easymcq

Which term is the broader umbrella concept underlying all fine-tuning approaches?

Q11hardmcq

A company wants to shift a general FM toward legal-industry terminology using a large volume of unlabeled contracts, without task-specific labels. Best approach?

Q12hardmulti

Which of the following are legitimate data-preparation concerns for fine-tuning covered in this lesson?

Select all that apply.

Q13hardmcq

A team wants their assistant to sound more helpful and safe based on human reviewers ranking sample outputs. Which technique directly matches this?

Q14hardordering

Order a typical model-specialization progression from earliest/broadest to latest/most specific.

1.Fine-tuning on labeled task-specific data
2.Pre-training the base FM
3.RLHF alignment
4.Continuous pre-training on domain data
Q15hardmatching

Match each term to its definition.

Instruction tuning
Domain adaptation
Distillation
Transfer learning
Q16hardmcq

A startup has a tight budget and a narrow, well-defined task. Per the general rule of thumb from Lessons 7 and 9, what should they try before fine-tuning?

Q17hardmcq

A fine-tuning dataset is heavily skewed toward one customer segment's phrasing, causing poor performance for other segments in production. This is primarily a failure of:

Q18hardmulti

Which statements about pre-training are TRUE?

Select all that apply.

Q19hardmcq

A model fine-tuned only for format-following (instruction tuning) still uses outdated or generic domain vocabulary. What's the more targeted fix?

Q20hardmcq

Why does fine-tuning generally require far less data than pre-training?