Figure from the original question paper

(A) [a] your mind [b] ask for brief replies [c] demonstrate the format[d] less the model has to guess [z]

(D) [v] your mind [w] ask for brief replies [x] demonstrate the format[y] less the model has to guess [z]

(B) [a] your mind [b] ask for brief replies [c] demonstrate the format[d] less the model has to guess

(C) [v] your mind [w] ask for brief replies [x] demonstrate the format[y] less the model has to guess

(E) [a] your mind [b] demonstrate the format [c] ask for brief replies[d] less the model has to guess

January 2025 term · Large Language Models · BSDA5004

Large Language Models End Term: 13 April 2025, Set 1 (January 2025 term)

The IIT Madras BS Large Language Models (LLM) End Term paper sat on 13 Apr 2025, in the January 2025 term, set 1: 15 questions for 33 marks in 180 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.

Questions: 15
Marks: 33
Duration: 180 min
MCQ: 3
MSQ: 7
Numerical: 4
Written: 1

Take as mock test Practise with answers

Updated 25 Sept 2026

Official paper: IIT M DEGREE FN EXAM QDB2 13 Apr 2025 · No negative marking.

Question 1

+2 marksOne correct option

Suppose we are given a Model A with N layers and a dataset with D tokens. Choose the correct statements according to the scaling law.

A
Increasing the model size requires a proportionate increase in the dataset sizeto reduce the test loss
B
Increasing the model size requires a proportionate increase in the dataset sizeto reduce the training loss
C
Increasing the model size requires us to double the dataset size to reduce thetest loss
D
Increasing the model size requires us to double the dataset size to reduce thetrain loss

Show answer

Correct answer

A
Increasing the model size requires a proportionate increase in the dataset sizeto reduce the test loss

Question 1 on its own pageAlso asked in End Term 13 Apr 2025

Question 2

+2 marksOne correct option

The statement that KV caching is not helpful during inference in encoder models like BERT is

A
TRUE
B
FALSE

Show answer

Correct answer

A
TRUE

Question 2 on its own pageAlso asked in End Term 13 Apr 2025

Question 3

+2 marksOne correct option

Consider the problem of length extrapolation using the Absolute Position Encoding (APE) scheme. Suppose the context length of a model during training is 512. Which of the following approaches allows the model to extrapolate to a context length of 1024 tokens

A
Fixed Sinusoidal Encoding
B
Parameterized APE
C
ALiBi
D
T5

Show answer

Correct answer

A
Fixed Sinusoidal Encoding

Question 3 on its own pageAlso asked in End Term 13 Apr 2025

Question 4

+2 marksOne or more correct options

Select correct statements about sinusoidal positional encoding:

Select all that apply.

A
It produces a unique encoding for each time-step (token’s position in asentence)
B
Distance between two tokens at time-steps 3 and 4 is independent of thelength of the input sequence.
C
It is non-deterministic.
D
None of these.

Show answer

Correct answers

A
It produces a unique encoding for each time-step (token’s position in asentence)
B
Distance between two tokens at time-steps 3 and 4 is independent of thelength of the input sequence.

Question 4 on its own pageAlso asked in End Term 13 Apr 2025

Question 5

+2 marksOne or more correct options

Suppose we have a dataset for machine translation tasks with thousands of samples. Suppose a team considers training the transformer model. The model could be trained using two approaches
A. Autoregressive training B. Teacher forcing
Choose the correct statements

Select all that apply.

A
Approach A helps the model to converge faster than approach B
B
Approach B helps the model to converge faster than approach A
C
One can start the training with approach B first and then switch to approach Aafter some training steps
D
Once the training starts with approach A and then switching to approach Bafter some training steps cannot be done

Show answer

Correct answers

B
Approach B helps the model to converge faster than approach A
C
One can start the training with approach B first and then switch to approach Aafter some training steps

Question 5 on its own pageAlso asked in End Term 13 Apr 2025

Question 6

+2 marksOne or more correct options

Select correct statements regarding character level tokenization:
Note: Assume that the character level vocabulary has English uppercase letters, lowercase letters, digits, special symbols, etc.

Select all that apply.

A
The size of the vocabulary with character-level tokenization will be larger thanthat with word-level tokenization if the corpus consists of all the books published on Harry Potter (in English).
B
The vocabulary will not expand if new English words/sentences are added tothe corpus.
C
No issue of handling unknown tokens as long as the unknown tokens are fromEnglish vocabulary
D
None of these.

Show answer

Correct answers

B
The vocabulary will not expand if new English words/sentences are added tothe corpus.
C
No issue of handling unknown tokens as long as the unknown tokens are fromEnglish vocabulary

Question 6 on its own pageAlso asked in End Term 13 Apr 2025

Question 7

+2 marksOne or more correct options

Which of the following are 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 7 on its own pageAlso asked in End Term 13 Apr 2025

Question 8

+3 marksOne or more correct options

Select all that apply.

A
[a] your mind [b] ask for brief replies [c] demonstrate the format[d] less the model has to guess [z]
B
[a] your mind [b] ask for brief replies [c] demonstrate the format[d] less the model has to guess
C
[v] your mind [w] ask for brief replies [x] demonstrate the format[y] less the model has to guess
D
[v] your mind [w] ask for brief replies [x] demonstrate the format[y] less the model has to guess [z]
E
[a] your mind [b] demonstrate the format [c] ask for brief replies[d] less the model has to guess
F
None of these

Show answer

Correct answers

A
[a] your mind [b] ask for brief replies [c] demonstrate the format[d] less the model has to guess [z]
D
[v] your mind [w] ask for brief replies [x] demonstrate the format[y] less the model has to guess [z]

Question 8 on its own page

Question 9

+2 marksNumerical answer

Show answer

Correct answer: 164

Question 9 on its own page

Question 10

+2 marksNumerical answer

Consider following distribution generated by a GPT model after feeding the sentence “I love Manali because it is so”:

Consider following distribution generated by a GPT model after feeding the sentence “I love Manali because it is so”:
Please note that the words are listed in decreasing order of their probabilities.
Based on the above data, answer the given subquestions

If you are using the Top-P sampling method, what should be the minimum value of P to make sure that “hilly” could be selected as an output?

Show answer

Correct answer: 0.83 (accepted within ±0.005)

Question 10 on its own page

Question 11

+2 marksOne or more correct options

Consider following distribution generated by a GPT model after feeding the sentence “I love Manali because it is so”:

What value of P will allow “bad” to be selected in the output?

Select all that apply.

A
1
B
0.9
C
0.95
D
0.005
E
0.75
F
0.88

Show answer

Correct answers

A
1
B
0.9
C
0.95

Question 11 on its own page

Question 12

+2 marksNumerical answer

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

Consider the following dictionary with the number of word occurrences in a corpus:
You will be working with word piece algorithm, answer the given subquestions in that context:

How many tokens are there in the initial vocabulary?

Show answer

Correct answer: 11

Question 12 on its own page

Question 13

+3 marksWritten answer

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

Consider the following dictionary with the number of word occurrences in a corpus:
You will be working with word piece algorithm, answer the given subquestions in that context:

Which pair will be merged in the very first merge? Say the 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 outlined in the question.
NOTE: Enter the exact answer without any extra space in the beginning or at the end.

Show answer

Correct answer: la

Question 13 on its own page

Question 14

+3 marksNumerical answer

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

Consider the following dictionary with the number of word occurrences in a corpus:
You will be working with word piece algorithm, answer the given subquestions in that context:

What is the score of the pair merged in the very first merge?

Show answer

Correct answer: 0.33 (accepted within ±0.005)

Question 14 on its own page

Question 15

+2 marksOne or more correct options

Based on the above data, answer the given subquestions

Of these 4 blocks, choose the block(s) that is (are) proper permutation of 4 blocks.

Select all that apply.

A
a
B
b
C
c
D
d
E
None of these

Show answer

Correct answers

Question 15 on its own page