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A model fine-tuned only for format-following (instruction tuning) still uses outdated or generic domain vocabulary.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?