September 2024 term · Large Language Models · BSDA5004

Large Language Models Quiz 2: 1 December 2024 (September 2024 term)

The IIT Madras BS Large Language Models (LLM) Quiz 2 paper sat on 1 Dec 2024, in the September 2024 term: 16 questions for 40 marks in 120 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.

Questions: 16
Marks: 40
Duration: 120 min
Numerical: 7
MCQ: 3
Written: 1
MSQ: 5

Take as mock test Practise with answers

Updated 25 Sept 2026

Official paper: IIT M DEGREE AN EXAM QDB2 01 Dec 2024 · No negative marking.

Question 1

+2 marksNumerical answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo =  { "deeper": 5,
2        "keener": 6,
3        "sweeter": 7,}

Note: Append identifier/special symbol </w> to each word at the end.

You will be learning a byte pair encoding, answer the given subquestions in that context:

How many tokens are there in the initial vocabulary?

Show answer

Correct answer: 10

Question 1 on its own page

Question 2

+3 marksOne correct option

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo =  { "deeper": 5,
2        "keener": 6,
3        "sweeter": 7,}

Note: Append identifier/special symbol </w> to each word at the end.

You will be learning a byte pair encoding, answer the given subquestions in that context:

Which of the following pairs has the least frequency before any merge?

Show answer

Correct answer

Question 2 on its own page

Question 3

+3 marksWritten answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo =  { "deeper": 5,
2        "keener": 6,
3        "sweeter": 7,}

Note: Append identifier/special symbol </w> to each word at the end.

You will be learning a byte pair encoding, answer the given subquestions in that context:

What is the most frequent byte-pair before the very first merge? Say the most frequent byte pair is (‘a’,‘b’), then enter “ab” (without quotes and white spaces). If there is a tie between two or more candidates, pick the one that occurs first in the original vocabulary.

Show answer

Correct answer: ee

Question 3 on its own page

Question 4

+2 marksNumerical answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo =  { "deeper": 5,
2        "keener": 6,
3        "sweeter": 7,}

Note: Append identifier/special symbol </w> to each word at the end.

You will be learning a byte pair encoding, answer the given subquestions in that context:

What is the frequency of the most frequent byte-pair before the very first merge?

Show answer

Correct answer: 18

Question 4 on its own page

Question 5

+2 marksNumerical answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo =  { "deeper": 5,
2        "keener": 6,
3        "sweeter": 7,}

Note: Append identifier/special symbol </w> to each word at the end.

You will be learning a byte pair encoding, answer the given subquestions in that context:

How many tokens in the vocabulary will have their frequency revised/reduced after first merge?

Show answer

Correct answer: 2

Question 5 on its own page

Question 6

+2 marksNumerical answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo =  { "deeper": 5,
2        "keener": 6,
3        "sweeter": 7,}

Note: Append identifier/special symbol </w> to each word at the end.

You will be learning a byte pair encoding, answer the given subquestions in that context:

After the first merge, how many tokens are there in the updated vocabulary?

Show answer

Correct answer: 11

Question 6 on its own page

Question 7

+2 marksNumerical answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo = {'simulation</w>': 4,
2      'suspicious</w>': 3,
3      'stringent</w>': 4,}

Note: Identifier/special symbol </w> is already appended.

Suppose the process of building the vocabulary is stopped after 4 merges with BPE tokenization . The following is the list of first 4 merges in that order:

Merge No.	Token 1	Token 2	Merged
1	‘i’	‘o’	‘io’
2	‘u’	‘s’	‘us’
3	‘s’	‘i’	‘si’
4	‘si’	‘m’	‘sim’

Table 1: First 4 merges

Based on the above data, answer the given subquestions.

In how many tokens will the word ‘simultaneous</w>’ be tokenized? If this word can not be tokenized with the given vocabulary, then enter -1.

Note: Identifier/special symbol </w> is already appended.

Show answer

Correct answer: 10

Question 7 on its own page

Question 8

+2 marksNumerical answer

Consider the following dictionary with the number of word occurrences in a corpus:

python

1wo = {'simulation</w>': 4,
2      'suspicious</w>': 3,
3      'stringent</w>': 4,}

Note: Identifier/special symbol </w> is already appended.

Suppose the process of building the vocabulary is stopped after 4 merges with BPE tokenization . The following is the list of first 4 merges in that order:

Merge No.	Token 1	Token 2	Merged
1	‘i’	‘o’	‘io’
2	‘u’	‘s’	‘us’
3	‘s’	‘i’	‘si’
4	‘si’	‘m’	‘sim’

Table 1: First 4 merges

Based on the above data, answer the given subquestions.

In how many tokens will the word ‘stereotypical</w>’ be tokenized? If this word can not be tokenized with the given vocabulary, then enter -1.

Note: Identifier/special symbol </w> is already appended.

Show answer

Correct answer: -1

Question 8 on its own page

Question 9

+2 marksOne or more correct options

Which of the following represent(s) the normalization step(s) in building a tokenizer for the English language?

Select all that apply.

A
removing accents from characters.
B
converting upper case letters to small case letters.
C
assigning integers to tokens.
D
splitting a sentence by whitespaces.
E
the final post-processing step where the tokenizer adds the special tokens.
F
none of these.

Show answer

Correct answers

A
removing accents from characters.
B
converting upper case letters to small case letters.

Question 9 on its own page

Question 10

+2 marksOne or more correct options

Choose all the aspects of the pre-training datasets that impact the model’s performance.

Select all that apply.

A
Quality
B
Size
C
Diversity
D
None of these

Show answer

Correct answers

A
Quality
B
Size
C
Diversity

Question 10 on its own page

Question 11

+4 marksOne or more correct options

The strikeout words in the passage given below denote the words to be dropped from the original sentence.

“Metacognition ~~is an awareness~~ of one's thought processes and an understanding of the patterns behind them. The term comes from the root word meta, meaning ”beyond”, or ”on top of”. Metacognition ~~can take many forms~~, such as reflecting on one's ways of thinking, and knowing when and how oneself and others use ~~particular strategies~~ for problem-solving. There are generally two components of metacognition: cognitive conceptions ~~and cognitive regulation~~ system.”

Which of the following represents the correct target sequence, with sentinel tokens, to the baseline model (according to T5 study) that uses the pre-training denoising objective? The characters inside the square brackets are the sentinel tokens and [z] represents the end of the sentinel token in a sentence

Select all that apply.

A
[v] is an awareness [w] can take many forms [x] particular strategies [y] and cognitive regulation
B
[v] is an awareness [w] can take many forms [x] particular strategies [y] and cognitive regulation [z]
C
[v] is an awareness [w] can take many forms [x] and cognitive regulation [y] particular strategies
D
[v] is an awareness [w] particular strategies [x] can take many forms [y] and cognitive regulation
E
[a] is an awareness [b] can take many forms [c] particular strategies [d] and cognitive regulation
F
[a] is an awareness [b] can take many forms [c] particular strategies [d] and cognitive regulation[z]
G
[a] is an awareness [b] can take many forms [c] and cognitive regulation
H
None of these.

Show answer

Correct answers

B
[v] is an awareness [w] can take many forms [x] particular strategies [y] and cognitive regulation [z]
F
[a] is an awareness [b] can take many forms [c] particular strategies [d] and cognitive regulation[z]

Question 11 on its own page

Question 12

+3 marksNumerical answer

Suppose you are working on prefix language modeling. The sequence length is 32 and the first two tokens represent the task specific prefix. How many non-infinity elements are there in the mask for computing attention scores?

Show answer

Correct answer: 529

Question 12 on its own page

Question 13

+3 marksOne correct option

Which of the following components in the data pre-processing pipeline removes pages that contain bad words?

A
Language Identification
B
Exact Deduplication
C
Fuzzy Deduplication
D
ML classifiers for quality filtering
E
Simple heuristics to detect toxic contents

Show answer

Correct answer

E
Simple heuristics to detect toxic contents

Question 13 on its own pageAlso asked in Quiz 2 24 Mar 2024

Question 14

+2 marksOne correct option

Which of the following is an important implication of the scaling law:

A
Increasing the size of the model only can improve test performance, irrespective of the size of the training data.
B
Increasing the size of the data only can improve test performance, irrespective of the size of the model.
C
Both the size of the model and that of the data, have to be increased appropriately to improve test performance.
D
None of these.

Show answer

Correct answer

C
Both the size of the model and that of the data, have to be increased appropriately to improve test performance.

Question 14 on its own page

Question 15

+3 marksOne or more correct options

Which of the following are the design choices for building a large language model?

Select all that apply.

A
Activation function in feed forward layer
B
Training dataset.
C
Positional encoding
D
Attention mechanism.
E
None of these.

Show answer

Correct answers

A
Activation function in feed forward layer
B
Training dataset.
C
Positional encoding
D
Attention mechanism.

Question 15 on its own page

Question 16

+3 marksOne or more correct options

Consider an ideal data preprocessing pipeline to prepare a dataset. Which of the following sentences or sets of paragraphs will NOT pass through the pipeline as it is?

Select all that apply.

A
“You idiot. All those people from the country Sumatrazia are taking our jobs. If I get a chance, I will wipe that country off of the face of the planet.”
B
Rajesh is an amazing businessman. He holds a savings account with Dena bank. His phone number is 9876543210.
C
Deep learning is a subset of Machine Learning.
Deep learning is a subset of Machine Learning.
D
India’s space agency ISRO says it has successfully brought back into Earth’s orbit a part of the rocket that carried its historic Moon mission recently.
E
Last Saturday, the Board of Control for Cricket in India (BCCI) officially communicated to the International Cricket Council (ICC) its decision to not send a team for the 2025 ICC Champions Trophy to be held in Pakistan from February 19 to March 9.
F
None of these.

Show answer

Correct answers

A
“You idiot. All those people from the country Sumatrazia are taking our jobs. If I get a chance, I will wipe that country off of the face of the planet.”
B
Rajesh is an amazing businessman. He holds a savings account with Dena bank. His phone number is 9876543210.
C
Deep learning is a subset of Machine Learning.
Deep learning is a subset of Machine Learning.

Question 16 on its own page