wayground logo

Free Printable Worksheets

NEW

Font size

S
M
L
XL
Worksheets

Fine-Tuning Quiz Answers

Total questions: 20

Worksheet time: 10mins

Name
Class
Date
1.

What is the primary purpose of fine-tuning a pretrained model?

a)

To adapt models to domain-specific data

b)

To compress model weights for storage

c)

To replace pretraining with faster training

d)

To remove attention layers entirely

2.

In full fine-tuning, which parameters are typically updated?

a)

All parameters of the model

b)

Only embedding matrix weights

c)

Only normalization layer scales

d)

Only output layer parameters

3.

Which scenario most strongly justifies avoiding full fine-tuning due to feasibility?

a)

Model has public checkpoints available

b)

Dataset is very large and proprietary

c)

Evaluation metrics are already standardized

d)

Team has extensive GPU availability

4.

What is a common risk when fully fine-tuning large models on small datasets?

a)

Overfitting and high compute cost

b)

Underfitting and low variance

c)

Improved generalization without cost

d)

Stable training with fewer epochs

5.

What does PEFT stand for in model adaptation?

a)

Parameter-Efficient Fine-Tuning

b)

Parallel-Enabled Forward Training

c)

Probabilistic Encoder Fine-Tuning

d)

Partial-Embedding Feature Transfer

6.

In many PEFT methods, what is the status of the base model during training?

a)

Base model reinitialized

b)

Base model pruned

c)

Base model quantized

d)

Base model frozen

7.

Roughly what percentage of parameters can PEFT reduce training to compared with full fine-tuning?

a)

30–40% of parameters

b)

90–95% of parameters

c)

60–70% of parameters

d)

10–20% of parameters

8.

Which technique introduces low-rank adapters into attention layers to reduce trainable parameters?

a)

Adapters without attention

b)

Prompt Tuning

c)

LoRA

d)

Prefix Tuning

9.

In LoRA, what is inserted inside attention layers to capture updates efficiently?

a)

Low-rank matrices inside attention layers

b)

Sparse pruning masks across all layers

c)

High-rank projections in feedforward blocks

d)

Dropout units replacing value heads

10.

What is a general effect of adding PEFT modules to a model’s architecture?

a)

Replaces layer normalization mechanisms

b)

Increases tokenization vocabulary sizes

c)

Removes positional encodings from inputs

d)

Adds small trainable modules to model layers

11.

When are PEFT techniques particularly advantageous compared to full fine-tuning?

a)

When label noise is extremely high

b)

When latency is the only constraint

c)

When data is unlimited and cheap

d)

When compute resources are limited

12.

Which set of parameters are trained in some minimal PEFT variants like simple adapters?

a)

Only bias terms

b)

Only embedding positions

c)

Only key-value heads

d)

Only normalization gammas

13.

Which sequence best represents a practical fine-tuning workflow for deployment?

a)

Collect → Label → Test-only → Ship

b)

Tokenize → Pretrain → Quantize → Deploy

c)

Load → Choose PEFT → Train → Evaluate → Deploy

d)

Pretrain → Prune → Distill → Deploy

14.

What is a valid strategy to balance adaptation quality and efficiency for large models?

a)

Combining full fine-tuning + PEFT

b)

Using zero-shot without validation

c)

Training only embeddings forever

d)

Skipping hyperparameter search entirely

15.

Which application commonly benefits from fine-tuning for specialized behavior?

a)

GPU memory diagnostics

b)

General file compression

c)

Operating system bootloaders

d)

Domain-specific chatbots

16.

You must adapt a large language model to legal documents with tight GPU budgets. Which approach is most suitable?

a)

Replace attention with convolution layers

b)

Use PEFT to update small adapter modules

c)

Retrain the model from scratch on statutes

d)

Fully fine-tune all layers for many epochs

17.

A team reports that LoRA improved training speed but lost generalization. Which adjustment best addresses this first?

a)

Remove adapters and freeze embeddings

b)

Disable evaluation to save training time

c)

Increase LoRA rank and regularization strength

d)

Lower learning rate while removing decay

18.

During evaluation, a PEFT-tuned model overfits the training set. Which action is most reasonable next?

a)

Freeze more layers and remove adapters

b)

Train longer without checkpoints or metrics

c)

Increase batch size without changing anything

d)

Add validation early stopping and augment data

19.

Which statement best contrasts PEFT with full fine-tuning in deployment contexts?

a)

PEFT stores small deltas, easing versioning

b)

PEFT requires larger checkpoints for shipping

c)

PEFT mandates retraining tokenizers too

d)

PEFT always improves latency regardless

20.

A project needs quick iteration across multiple domains using one base model. Which setup is most effective?

a)

Maintain separate PEFT adapters per domain

b)

Train separate full models for each domain

c)

Use one fixed model without adapters

d)

Replace all attention with static prompts