wayground logo

Free Printable Worksheets

NEW

Font size

S
M
L
XL
Worksheets

Page 1

Total questions: 10

Worksheet time: 5mins

Name
Class
Date
1.

Which model family is described as encoder-only in modern NLP architectures?

a)

Autoregressive decoders like GPT-4

b)

Encoder–decoder hybrids like T5

c)

Diffusion-based latent models

d)

Transformer encoder models like BERT

2.

Which statement best characterizes a strictly autoregressive decoder during generation?

a)

It computes all output tokens in parallel

b)

It encodes the full input before any decoding begins

c)

It simultaneously attends to future and past tokens

d)

It only sees previously generated tokens sequentially

3.

Top-p (nucleus) sampling primarily aims to achieve which outcome in text generation?

a)

Ensure identical outputs across repeated runs

b)

Guarantee the most likely token every time

c)

Balance creativity with coherence in outputs

d)

Reduce vocabulary size during tokenization

4.

Which decoding method is most likely to produce deterministic and focused outputs?

a)

Beam search with high diversity penalty

b)

Top-p sampling with high nucleus

c)

Random sampling with high temperature

d)

Greedy decoding with low temperature

5.

Which penalty discourages the model from repeating words it has already used frequently in the session?

a)

Top-k truncation of candidate set

b)

Frequency penalty applied during decoding

c)

Presence penalty applied only to new tokens

d)

Temperature scaling before softmax

6.

Tokenization in large language models is best described as which process?

a)

Sampling words from a probability distribution

b)

Converting text into discrete subword tokens

c)

Compressing sentences into fixed-length vectors

d)

Parsing syntax using a context-free grammar

7.

Which configuration most increases randomness and diversity in generated text?

a)

High temperature with random sampling

b)

Low temperature with greedy decoding

c)

Top-p value near zero with sampling

d)

Beam width of one with repetition penalty

8.

Which pair correctly matches popular proprietary model families with their category?

a)

BERT and T5 as proprietary models

b)

GPT-4o and Gemini as proprietary models

c)

Claude and LLaMA as proprietary models

d)

LLaMA and Mistral as proprietary models

9.

External tools like search and code interpreters are typically used in conjunction with LLMs to achieve which benefit?

a)

Augment capabilities for tasks requiring tools

b)

Replace the model’s internal reasoning entirely

c)

Reduce token usage during tokenization steps

d)

Increase training data diversity automatically

10.

Setting temperature to a very low value (e.g., 0.1) during decoding most likely has what effect?

a)

Outputs become more deterministic and focused

b)

Outputs explore more diverse continuations

c)

Vocabulary is expanded during inference

d)

Model ignores previously generated tokens