wayground logo

Free Printable Worksheets

NEW

Font size

S
M
L
XL
Worksheets

Mid-Level MCQs on Transformers and LLMs (Part 1)

Total questions: 25

Worksheet time: 13mins

Name
Class
Date
1.

Which component of the Transformer is responsible for capturing relationships between tokens regardless of distance?

a)

Attention mechanism

b)

Positional encoding

c)

Feed-forward network

2.

What is the main advantage of self-attention in Transformers?

a)

It reduces training data requirements

b)

It processes all tokens in parallel

c)

It uses fewer parameters than RNNs

3.

Which of the following best defines an embedding?

a)

A compressed text file

b)

A numerical vector representing semantic meaning

c)

A special token representing sentence boundaries

4.

Vector databases store embeddings primarily to support:

a)

GPU acceleration

b)

Fast semantic search

c)

Long-term backup

5.

In Retrieval-Augmented Generation (RAG), the retriever’s main role is to:

a)

Generate final answers

b)

Convert prompts to embeddings

c)

Fetch relevant documents

6.

Which training step adjusts model parameters to minimize loss?

a)

Forward pass

b)

Tokenization

c)

Backward pass

7.

What does “fine-tuning” typically modify in a pretrained model?

a)

Tokenizer

b)

Internal weights

c)

GPU configuration

8.

Which method is used to represent token sequences numerically before entering a Transformer?

a)

Positional indexing

b)

Word clouds

c)

One-hot encoding

9.

In a multi-agent LLM system, agents primarily interact through:

a)

Static JSON files

b)

GPU kernel calls

c)

Shared memory and messages

10.

Which loss function is commonly used for language modeling?

a)

Cross-entropy loss

b)

Mean squared error

c)

Hinge loss

11.

Function calling in LLMs mainly helps with:

a)

Reducing GPU load

b)

Improving tokenization speed

c)

Structured and deterministic outputs

12.

Which similarity metric is most commonly used in vector databases?

a)

Manhattan distance

b)

Cosine similarity

c)

Hamming distance

13.

The key difference between RNNs and Transformers is that Transformers:

a)

Work only on images

b)

Use recurrence for sequence modeling

c)

Use attention instead of recurrence

14.

What is the main benefit of batching during model training?

a)

Increases dataset size

b)

Improves computational efficiency

c)

Reduces model parameters

15.

Which of the following is NOT a common use of embeddings?

a)

Increasing batch size

b)

Recommendation systems

c)

Semantic search

16.

In RAG, a reranker is used to:

a)

Generate final summaries

b)

Reorder retrieved results based on relevance

c)

Clean up documents before indexing

17.

Which component of the Transformer helps in capturing nonlinear relationships?

a)

Layer normalization

b)

Feed-forward layers

c)

Softmax operation

18.

What is the output of the softmax layer in classification tasks?

a)

Embedding vectors

b)

Normalized probabilities

c)

Raw logits

19.

During fine-tuning, overfitting can be reduced using:

a)

More parameters

b)

Larger learning rate

c)

Dropout

20.

What does the attention score represent?

a)

The gradient magnitude during backprop

b)

Relevance of one token to another

c)

How likely a token is to be masked

21.

Which indexing method is commonly used in vector databases for fast similarity search?

a)

KD-tree

b)

Bloom filter

c)

HNSW graph

d)

Hash maps

22.

In model evaluation, what does perplexity primarily measure?

a)

Model size in parameters

b)

Training dataset diversity

c)

GPU memory consumption

d)

How well the model predicts text

23.

In a multi-agent LLM system, what is the planner agent mainly responsible for?

a)

Serving the user interface layer

b)

Generating embeddings for documents

c)

Decomposing tasks into subgoals

d)

Translating languages between agents

24.

Which tokenization method is used in most modern large language models?

a)

Character splitting only

b)

Whitespace segmentation

c)

Byte-Pair Encoding (BPE)

d)

Rule-based tokenization

25.

What is the primary purpose of positional encoding in transformer models?

a)

Improve gradient stability in layers

b)

Inject order information into tokens

c)

Encode overall sentence length

d)

Reduce hallucinations during decoding