LLM — da5004_2025T2_Q1_NA

Source: da5004_2025T2_Q1_NA.pdf · Academic questions: 21 · Separately counted children: 0
Provenance: Large_Language_Models/da5004_2025T2_Q1_NA.pdf · SHA-256 a8f64888d57debe8682701d05682f6028b33f9120ac354ab933639265a872afa

Scope: Section 8: LLM (Degree Level : Large Language Models (Computer Based Exam)), physical pages 1 to 15. The in-scope subject begins on physical page 1 with Section 8 header (Section Id 64065391701) at y=491.39 and terminates on physical page 15 immediately preceding Section 9: DL(CV) (Section Id 64065391702) at y=406.50.

Questions

Question 135

Question ID: 6406531279474

Type: Multiple Choice Question · Marks: 2
PDF pages: 2

Transformers process input tokens:

Options

One at a time (sequentially)

Source option ID: 6406534313782

In reverse order

Source option ID: 6406534313783

All at once (in parallel)

Source option ID: 6406534313784

Only after seeing the full input

Source option ID: 6406534313785

Correct answer: All at once (in parallel)

Source evidence: [{"page": 2, "bbox": [23.62, 586.6, 119.02, 602.92], "type": "green source option", "detail": "Option 6406534313784 is rendered in green (color 32768)"}]

Question 136

Question ID: 6406531279475

Type: Multiple Choice Question · Marks: 2
PDF pages: 3

What is the purpose of the softmax function in the attention mechanism?

Options

Normalize attention scores to a probability distribution

Source option ID: 6406534313786

Add non-linearity to the model

Source option ID: 6406534313787

To predict the correct class for the loss function

Source option ID: 6406534313788

Remove redundant features from the input

Source option ID: 6406534313789

Correct answer: Normalize attention scores to a probability distribution

Source evidence: [{"page": 3, "bbox": [23.62, 114.1, 119.02, 130.42], "type": "green source option", "detail": "Option 6406534313786 is rendered in green (color 32768)"}]

Question 137

Question ID: 6406531279476

Type: Multiple Choice Question · Marks: 2
PDF pages: 3

What is the main difference between GPT and BERT pre-training objectives?

Options

GPT uses Masked Language Modeling, BERT uses Causal Language Modeling

Source option ID: 6406534313790

GPT uses Causal Language Modeling, BERT uses Masked Language Modeling

Source option ID: 6406534313791

Both use Masked Language Modeling

Source option ID: 6406534313792

Both use Causal Language Modeling

Source option ID: 6406534313793

Correct answer: GPT uses Causal Language Modeling, BERT uses Masked Language Modeling

Source evidence: [{"page": 3, "bbox": [23.62, 338.62, 119.02, 354.94], "type": "green source option", "detail": "Option 6406534313791 is rendered in green (color 32768)"}]

Question 138

Question ID: 6406531279477

Type: Multiple Choice Question · Marks: 2
PDF pages: 3

In Top-K sampling for language generation, increasing the value of K typically has which of the following effects?

Options

It makes the output more deterministic and repetitive.

Source option ID: 6406534313794

It reduces the probability of selecting high-frequency words.

Source option ID: 6406534313795

It increases the diversity of the generated text but may reduce coherence if K is too large.

Source option ID: 6406534313796

It guarantees grammatical correctness by focusing on top-ranked tokens only.

Source option ID: 6406534313797

Correct answer: It increases the diversity of the generated text but may reduce coherence if K is too large.

Source evidence: [{"page": 3, "bbox": [23.62, 579.46, 119.02, 595.78], "type": "green source option", "detail": "Option 6406534313796 is rendered in green (color 32768)"}]

Question 139

Question ID: 6406531279478

Type: Multiple Choice Question · Marks: 2
PDF pages: 4

Why are residual connections important in transformer architectures?

Options

They reduce memory consumption.

Source option ID: 6406534313798

They add extra cost of compute by adding batch normalization.

Source option ID: 6406534313799

They help in training deep networks by enabling gradient flow.

Source option ID: 6406534313800

They remove the need for layer normalization.

Source option ID: 6406534313801

Correct answer: They help in training deep networks by enabling gradient flow.

Source evidence: [{"page": 4, "bbox": [23.62, 155.74, 119.02, 172.06], "type": "green source option", "detail": "Option 6406534313800 is rendered in green (color 32768)"}]

Question 140

Question ID: 6406531279479

Type: Multiple Select Question · Marks: 3
PDF pages: 4

Which of the following statements are true regarding causal language modeling (CLM)?

Options

The model only attends to past and current tokens during training.

Source option ID: 6406534313802

The model is trained by predicting the next token in a sequence.

Source option ID: 6406534313803

The model uses bidirectional context.

Source option ID: 6406534313804

The CLM objective is commonly used for encoder-only models.

Source option ID: 6406534313805

Correct answer: The model only attends to past and current tokens during training.; The model is trained by predicting the next token in a sequence.

Source evidence: [{"page": 4, "bbox": [23.62, 375.76, 119.02, 392.08], "type": "green source option", "detail": "Option 6406534313802 is rendered in green (color 32768)"}, {"page": 4, "bbox": [23.62, 396.58, 119.02, 412.9], "type": "green source option", "detail": "Option 6406534313803 is rendered in green (color 32768)"}]

Question 141

Question ID: 6406531279480

Type: Multiple Select Question · Marks: 3
PDF pages: 4

Which of the following are valid reasons why transformer-based large language models are widely used in natural language processing?

Options

Transformers process input sequences in parallel, enabling faster training.

Source option ID: 6406534313806

They use recurrence to remember long-term dependencies more effectively than LSTMs.

Source option ID: 6406534313807

Transformers are limited to short text inputs due to their architecture.

Source option ID: 6406534313808

Large transformer models can be fine-tuned for various NLP tasks using a single pre-trained model.

Source option ID: 6406534313809

Correct answer: Transformers process input sequences in parallel, enabling faster training.; Large transformer models can be fine-tuned for various NLP tasks using a single pre-trained model.

Source evidence: [{"page": 4, "bbox": [23.62, 595.78, 119.02, 612.1], "type": "green source option", "detail": "Option 6406534313806 is rendered in green (color 32768)"}, {"page": 4, "bbox": [23.62, 674.56, 119.02, 690.88], "type": "green source option", "detail": "Option 6406534313809 is rendered in green (color 32768)"}]

Question 142

Question ID: 6406531279481

Type: Multiple Select Question · Marks: 3
PDF pages: 5

Which elements are included in BERT’s input representation for Next Sentence Prediction?

Options

[CLS] token at the beginning of input.

[CLS] token at the beginning of input option block

Figure 1. [CLS] token at the beginning of input option block

Source option ID: 6406534313810

[SEP] token to separate sentence A and sentence B.

[SEP] token to separate sentence A and sentence B option block

Figure 2. [SEP] token to separate sentence A and sentence B option block

Source option ID: 6406534313811

Segment embeddings to distinguish sentence A and sentence B.

Segment embeddings to distinguish sentence A and sentence B option block

Figure 3. Segment embeddings to distinguish sentence A and sentence B option block

Source option ID: 6406534313812

Position embeddings are not required for Next Sentence Prediction.

Position embeddings are not required for Next Sentence Prediction option block

Figure 4. Position embeddings are not required for Next Sentence Prediction option block

Source option ID: 6406534313813

Correct answer: [CLS] token at the beginning of input.; [SEP] token to separate sentence A and sentence B.; Segment embeddings to distinguish sentence A and sentence B.

Source evidence: [{"page": 5, "bbox": [23.62, 120.03, 119.02, 136.35], "type": "green source option", "detail": "Option 6406534313810 is rendered in green (color 32768)"}, {"page": 5, "bbox": [23.62, 161.6, 119.02, 177.92], "type": "green source option", "detail": "Option 6406534313811 is rendered in green (color 32768)"}, {"page": 5, "bbox": [23.62, 203.93, 119.02, 220.25], "type": "green source option", "detail": "Option 6406534313812 is rendered in green (color 32768)"}]

Question 143

Question ID: 6406531279482

Type: Multiple Select Question · Marks: 3
PDF pages: 5

How does Top-p (nucleus) sampling differ from Top-K sampling in language generation?

Options

It samples only from a fixed number of tokens at each step.

Source option ID: 6406534313814

It samples from the smallest set of tokens whose cumulative probability exceeds p.

Source option ID: 6406534313815

It always selects the top-p tokens with equal probability.

Source option ID: 6406534313816

It guarantees diversity by selecting all low-probability tokens.

Source option ID: 6406534313817

Correct answer: It samples from the smallest set of tokens whose cumulative probability exceeds p.

Source evidence: [{"page": 5, "bbox": [23.62, 424.63, 119.02, 440.95], "type": "green source option", "detail": "Option 6406534313815 is rendered in green (color 32768)"}]

Question 144

Question ID: 6406531279483

Type: Short Answer Question · Marks: 3
PDF pages: 6

Consider the embedding vector for a word, $x = [0.5, -0.4, 0.3, -0.2]^T$. Suppose the word is at position 3 in the given sentence. Add the corresponding position embedding $p$ to the word embedding to get $h$, i.e. the sum of the elements in $h = x + p$. Use the fixed sinusoidal position embedding vector calculated using the formula given below:

\[\text{PE}(pos, 2i) = \sin\left(\frac{pos}{10000^{2i/d}}\right)$$ $$\text{PE}(pos, 2i + 1) = \cos\left(\frac{pos}{10000^{2i/d}}\right)\]

What is the value of $h[0] - h[1]$ i.e. substract second element from the first element of $h$ vector ? (Answer correct up to 2 values after the decimal)

Sinusoidal positional embedding formulation and question parameters for token position 3

Figure 5. Sinusoidal positional embedding formulation and question parameters for token position 3

Correct answer: 1.98 to 2.08

Source evidence: [{"page": 6, "bbox": [22.87, 404.1, 180.0, 422.5], "type": "explicit field", "detail": "Possible Answers: 1.98 to 2.08"}]

Question 145

Question ID: 6406531279484

Type: Short Answer Question · Marks: 3
PDF pages: 7

Suppose a language model outputs the following logits for a vocabulary of size 4:

Token	Logit
the	2.0
sky	1.0
is	0.5
blue	-1.0

You apply temperature scaling with temperature $T = 0.5$. After applying the softmax on the temperature-scaled logits, what is the probability assigned to the token “the”? (Give your answer up to 2 decimal places)

Vocabulary token logits table and temperature scaling parameter T=0.5 for softmax probability calculation

Figure 6. Vocabulary token logits table and temperature scaling parameter T=0.5 for softmax probability calculation

Correct answer: 0.81 to 0.87

Source evidence: [{"page": 7, "bbox": [22.87, 405.6, 180.0, 424.0], "type": "explicit field", "detail": "Possible Answers: 0.81 to 0.87"}]

Question 146

Question ID: 6406531279485

Type: Short Answer Question · Marks: 3
PDF pages: 7

A transformer generates the following attention weights for a token over four positions: (0.1, 0.2, 0.5, 0.2). The corresponding value vectors are:

\[v_1 = [1, 0], \quad v_2 = [0, 1], \quad v_3 = [1, 1], \quad v_4 = [2, 1]\]

Compute the output vector from the attention mechanism by taking the weighted sum of the value vectors, and return the first element of the resulting vector. (Round to 2 decimal places)

Attention weights and value vectors v1, v2, v3, v4 for weighted sum output vector calculation

Figure 7. Attention weights and value vectors v1, v2, v3, v4 for weighted sum output vector calculation

Correct answer: 0.9 to 1.1

Source evidence: [{"page": 7, "bbox": [22.87, 749.0, 180.0, 767.3], "type": "explicit field", "detail": "Possible Answers: 0.9 to 1.1"}]

Shared context 147 to 149

Question ID: 6406531279486

Type: Comprehension
PDF pages: 9

Question Description

Embedding dimension : $d_{\text{model}} = 4$
Input sequence (sequence length = 2, $d_{\text{model}} = 4$) :
$$X = \begin{bmatrix} 1 & 0 & 1 & 0 \\ 0 & 2 & 0 & 2 \end{bmatrix}$$
Number of heads: 2 (each head operates on 2 dimensions)
For both the heads $d_K = d_Q = d_V = 2$
Weight matrices for $W_Q, W_K, W_V$ first head :
$$W_Q^1, W_K^1, W_V^1 = \begin{bmatrix} 1 & 0 \\ 0 & 1 \\ 1 & 0 \\ 0 & 1 \end{bmatrix}$$
Weight matrices for $W_Q, W_K, W_V$ second head :
$$W_Q^2, W_K^2, W_V^2 = \begin{bmatrix} 0 & 1 \\ 1 & 0 \\ 0 & 1 \\ 1 & 0 \end{bmatrix}$$
The output from both heads are concatenated, then projected by:
$$W_O = \begin{bmatrix} 0.5 & 0 & 0 & 0 \\ 0 & 0.5 & 0 & 0 \\ 0 & 0 & 0.5 & 0 \\ 0 & 0 & 0 & 0.5 \end{bmatrix}$$
$$\text{Scaled Dot-Product Attention}(Q, K, V) = \text{softmax}\left(\frac{Q K^T}{\sqrt{d_K}}\right) V$$

Comprehension data: embedding dimensions, input sequence matrix X, projection matrices WQ, WK, WV, WO, and attention formula

Figure 8. Comprehension data: embedding dimensions, input sequence matrix X, projection matrices WQ, WK, WV, WO, and attention formula

Based on the above data, answer the given subquestions.

Question 147

Question ID: 6406531279487

Type: Multiple Choice Question · Marks: 3
PDF pages: 9, 10

Shared context: ctx-147-149

Select the scaled dot-product attention for the first head:

Options

\[\begin{bmatrix} 0 & 0 \\ 1 & 1 \end{bmatrix}\]

Candidate attention matrix for head 1 option A

Figure 9. Candidate attention matrix for head 1 option A

Source option ID: 6406534313821

\[\begin{bmatrix} 0 & 0 \\ 6 & 12 \end{bmatrix}\]

Candidate attention matrix for head 1 option B

Figure 10. Candidate attention matrix for head 1 option B

Source option ID: 6406534313822

\[\begin{bmatrix} 1.46 & 2.92 \\ 4.53 & 9.08 \end{bmatrix}\]

Candidate attention matrix for head 1 option C

Figure 11. Candidate attention matrix for head 1 option C

Source option ID: 6406534313823

\[\begin{bmatrix} 1.98 & 3.76 \\ 4.02 & 8.31 \end{bmatrix}\]

Candidate attention matrix for head 1 option D

Figure 12. Candidate attention matrix for head 1 option D

Source option ID: 6406534313824

Correct answer: $$\begin{bmatrix} 0 & 0 \\ 6 & 12 \end{bmatrix}$$

Source evidence: [{"page": 10, "bbox": [23.62, 101.75, 119.02, 118.07], "type": "green source option", "detail": "Option 6406534313822 is rendered in green (color 32768)"}]

Question 148

Question ID: 6406531279488

Type: Multiple Choice Question · Marks: 3
PDF pages: 10

Shared context: ctx-147-149

Select the scaled dot-product attention for the second head:

Options

\[\begin{bmatrix} 1.34 & 2.23 \\ 0.66 & 4.77 \end{bmatrix}\]

Candidate attention matrix for head 2 option A

Figure 13. Candidate attention matrix for head 2 option A

Source option ID: 6406534313825

\[\begin{bmatrix} 2.33 & 1.32 \\ 2.94 & 3.76 \end{bmatrix}\]

Candidate attention matrix for head 2 option B

Figure 14. Candidate attention matrix for head 2 option B

Source option ID: 6406534313826

\[\begin{bmatrix} 1.96 & 1.68 \\ 3.04 & 2.32 \end{bmatrix}\]

Candidate attention matrix for head 2 option C

Figure 15. Candidate attention matrix for head 2 option C

Source option ID: 6406534313827

\[\begin{bmatrix} 2.65 & 2.60 \\ 2.58 & 2.32 \end{bmatrix}\]

Candidate attention matrix for head 2 option D

Figure 16. Candidate attention matrix for head 2 option D

Source option ID: 6406534313828

Correct answer: $$\begin{bmatrix} 1.34 & 2.23 \\ 0.66 & 4.77 \end{bmatrix}$$

Source evidence: [{"page": 10, "bbox": [23.62, 401.65, 119.02, 417.97], "type": "green source option", "detail": "Option 6406534313825 is rendered in green (color 32768)"}]

Question 149

Question ID: 6406531279489

Type: Multiple Choice Question · Marks: 2
PDF pages: 10, 11

Shared context: ctx-147-149

Concatenate the outputs from both the attention heads, then apply the output projection matrix Wo to produce the final output of the multi-head attention mechanism. Select the correct result of this operation.

Options

\[\begin{bmatrix} 0.73 & 1.46 \\ 2.26 & 4.54 \\ 2.33 & 1.32 \\ 2.94 & 3.76 \end{bmatrix}\]

Candidate multi-head attention projection matrix option A

Figure 17. Candidate multi-head attention projection matrix option A

Source option ID: 6406534313829

\[\begin{bmatrix} 0.99 & 1.88 \\ 2.01 & 4.16 \\ 1.96 & 1.68 \\ 3.04 & 2.32 \end{bmatrix}\]

Candidate multi-head attention projection matrix option B

Figure 18. Candidate multi-head attention projection matrix option B

Source option ID: 6406534313830

\[\begin{bmatrix} 0.00 & 0.00 \\ 0.50 & 0.50 \\ 2.65 & 2.60 \\ 2.58 & 2.32 \end{bmatrix}\]

Candidate multi-head attention projection matrix option C

Figure 19. Candidate multi-head attention projection matrix option C

Source option ID: 6406534313831

\[\begin{bmatrix} 0.00 & 0.00 \\ 3.00 & 6.00 \\ 1.34 & 2.23 \\ 0.66 & 4.77 \end{bmatrix}\]

Candidate multi-head attention projection matrix option D

Figure 20. Candidate multi-head attention projection matrix option D

Source option ID: 6406534313832

Correct answer: $$\begin{bmatrix} 0.00 & 0.00 \\ 3.00 & 6.00 \\ 1.34 & 2.23 \\ 0.66 & 4.77 \end{bmatrix}$$

Source evidence: [{"page": 11, "bbox": [23.62, 355.9, 119.02, 372.22], "type": "green source option", "detail": "Option 6406534313832 is rendered in green (color 32768)"}]

Shared context 150 to 152

Question ID: 6406531279490

Type: Comprehension
PDF pages: 12

Apply “add and norm” layer to the output of the multihead attention in which :

Input sequence (sequence length = 2, $d_{\text{model}} = 4$) to Multihead Attention:
$$X = \begin{bmatrix} 1 & 2 & 3 & 4 \\ 2 & 1 & 3 & 4 \end{bmatrix}$$
Multi head Attention Output :
$$\text{MHA}(X) = \begin{bmatrix} 2 & 3 & 1 & 2 \\ 5 & 4 & 1 & 2 \end{bmatrix}$$
$$\text{LayerNorm}(x) = \frac{x - \mu}{\sigma} \odot \gamma + \beta$$

Comprehension data: Add and Norm layer instructions, input sequence matrix X, and multi-head attention output matrix MHA(X)

Figure 21. Comprehension data: Add and Norm layer instructions, input sequence matrix X, and multi-head attention output matrix MHA(X)

Based on the above data, answer the given subquestions.

Question 150

Question ID: 6406531279491

Type: Short Answer Question · Marks: 1
PDF pages: 12, 13

Shared context: ctx-150-152

For the given input matrix X and multihead attention output MHA(X), apply a residual connection and store the result in matrix R, and finally compute the sum of all elements in matrix R.

Correct answer: 36

Source evidence: [{"page": 13, "bbox": [22.87, 16.8, 180.0, 35.1], "type": "explicit field", "detail": "Possible Answers: 36"}]

Question 151

Question ID: 6406531279492

Type: Short Answer Question · Marks: 2
PDF pages: 13

Shared context: ctx-150-152

To apply the layer normalization on the matrix R. How many learning parameters ($\theta$) will be required ?

Question stem: Layer normalization learnable parameters count calculation for matrix R

Figure 22. Question stem: Layer normalization learnable parameters count calculation for matrix R

Correct answer: 8

Source evidence: [{"page": 13, "bbox": [22.87, 299.3, 180.0, 317.7], "type": "explicit field", "detail": "Possible Answers: 8"}]

Question 152

Question ID: 6406531279493

Type: Short Answer Question · Marks: 3
PDF pages: 13

Shared context: ctx-150-152

Initialize $\gamma = 2, \beta = 1$ and use $\epsilon = 0$ and apply the layer normalization on the matrix R and sum of all the elements of the final output.

Question stem: Layer normalization initialization parameters gamma=2, beta=1, epsilon=0 and output sum calculation

Figure 23. Question stem: Layer normalization initialization parameters gamma=2, beta=1, epsilon=0 and output sum calculation

Correct answer: 8

Source evidence: [{"page": 13, "bbox": [22.87, 598.4, 180.0, 616.7], "type": "explicit field", "detail": "Possible Answers: 8"}]

Shared context 153 to 155

Question ID: 6406531279494

Type: Comprehension
PDF pages: 14

The table presents the conditional probability distribution over vocabulary tokens at each timestep during sequence generation. Each column corresponds to a timestep (from 1 to 5), and the values represent the probability of selecting each token given the tokens chosen in all previous timesteps. For timestep t, the values in the column represent:

\[P(\text{token}_t \mid \text{token}_1, \text{token}_2, \dots, \text{token}_{t-1})\]

timesteps	1	2	3	4	5
the	0.07	0.19	0.20	0.42	0.08
is	0.04	0.13	0.12	0.26	0.08
sky	0.23	0.12	0.35	0.05	0.15
[other]	0.03	0.18	0.16	0.16	0.46
very	0.22	0.01	0.07	0.11	0.15
blue	0.33	0.25	0.08	0.05	0.08

Table 1: Probability distribution over vocabulary tokens across timesteps 1 to 5

Figure 24. Table 1: Probability distribution over vocabulary tokens across timesteps 1 to 5

Based on the above data, answer the given subquestions.

Question 153

Question ID: 6406531279495

Type: Short Answer Question · Marks: 2
PDF pages: 14

Shared context: ctx-153-155

In exhaustive search, at timestep t=1, we run the decoder once to obtain probability distributions over all tokens in the vocabulary. Given the 6 tokens available in the table in the main question, how many times must we run the decoder at timestep t=4?

Correct answer: 216

Source evidence: [{"page": 14, "bbox": [22.87, 672.0, 180.0, 690.4], "type": "explicit field", "detail": "Possible Answers: 216"}]

Question 154

Question ID: 6406531279496

Type: Short Answer Question · Marks: 1
PDF pages: 14, 15

Shared context: ctx-153-155

How many total sequences of exactly length 5 are possible according to exhaustive search?

Correct answer: 7776

Source evidence: [{"page": 15, "bbox": [22.87, 113.4, 180.0, 131.7], "type": "explicit field", "detail": "Possible Answers: 7776"}]

Question 155

Question ID: 6406531279497

Type: Short Answer Question · Marks: 2
PDF pages: 15

Shared context: ctx-153-155

If we use top-k sampling with k=2 at timestep 1, what is the normalized probability of selecting token “sky” at the timestep=1?

Correct answer: 0.41

Source evidence: [{"page": 15, "bbox": [22.87, 363.6, 180.0, 381.9], "type": "explicit field", "detail": "Possible Answers: 0.41"}]

Answer key

Source question	Source identity	Correct answer	PDF pages
135	6406531279474	All at once (in parallel)	2
136	6406531279475	Normalize attention scores to a probability distribution	3
137	6406531279476	GPT uses Causal Language Modeling, BERT uses Masked Language Modeling	3
138	6406531279477	It increases the diversity of the generated text but may reduce coherence if K is too large.	3
139	6406531279478	They help in training deep networks by enabling gradient flow.	4
140	6406531279479	The model only attends to past and current tokens during training.; The model is trained by predicting the next token in a sequence.	4
141	6406531279480	Transformers process input sequences in parallel, enabling faster training.; Large transformer models can be fine-tuned for various NLP tasks using a single pre-trained model.	4
142	6406531279481	[CLS] token at the beginning of input.; [SEP] token to separate sentence A and sentence B.; Segment embeddings to distinguish sentence A and sentence B.	5
143	6406531279482	It samples from the smallest set of tokens whose cumulative probability exceeds p.	5
144	6406531279483	1.98 to 2.08	6
145	6406531279484	0.81 to 0.87	7
146	6406531279485	0.9 to 1.1	7
147	6406531279487	$$\begin{bmatrix} 0 & 0 \\ 6 & 12 \end{bmatrix}$$	9, 10
148	6406531279488	$$\begin{bmatrix} 1.34 & 2.23 \\ 0.66 & 4.77 \end{bmatrix}$$	10
149	6406531279489	$$\begin{bmatrix} 0.00 & 0.00 \\ 3.00 & 6.00 \\ 1.34 & 2.23 \\ 0.66 & 4.77 \end{bmatrix}$$	10, 11
150	6406531279491	36	12, 13
151	6406531279492	8	13
152	6406531279493	8	13
153	6406531279495	216	14
154	6406531279496	7776	14, 15
155	6406531279497	0.41	15

Extraction notes

Status: verified
Count reconciliation: Section 8 declares 16 questions and 50 marks. The 16 portal question units correspond to 1 administrative subject-confirmation question (Question 134, 0 marks), 12 standalone academic questions (Questions 135-139 [5 MCQs x 2 = 10 marks], Questions 140-143 [4 MSQs x 3 = 12 marks], Questions 144-146 [3 SAs x 3 = 9 marks]), and 3 comprehension blocks (Questions 147-149 [3+3+2 = 8 marks], Questions 150-152 [1+2+3 = 6 marks], and Questions 153-155 [2+1+2 = 5 marks]). Total academic questions reconcile to exactly 21 (12 standalone + 9 comprehension subquestions). Marks sum to 10 + 12 + 9 + 8 + 6 + 5 = 50, matching the declared Section Marks of 50.

Excluded source entries:

{"classification": "out-of-scope-section", "page": 1, "bbox": [0.0, 0.0, 595.3, 491.39], "reason": "Preceding section exam content (options 6406534313773 to 6406534313775, and Question 133) prior to Section 8 LLM header", "identity": "preceding-section"}
{"classification": "subject-confirmation", "page": 2, "bbox": [22.87, 80.0, 580.0, 280.0], "reason": "Administrative subject confirmation asking candidate to confirm subject and cross check hall ticket", "identity": "134"}
{"classification": "out-of-scope-section", "page": 15, "bbox": [0.0, 406.5, 595.3, 841.9], "reason": "Subsequent Section 9 DL(CV) exam header terminating LLM section", "identity": "following-section-9"}