LLM — da5004_2025T3_Q1_NA

Source: da5004_2025T3_Q1_NA.pdf · Academic questions: 20 · Separately counted children: 0
Provenance: Large_Language_Models/da5004_2025T3_Q1_NA.pdf · SHA-256 30260f7a668f593215b5bcbb39bb0c90bf56b6d294b34b5d17c7189495954c0d

Scope: Large Language Models (LLM) Section 10 examination content from page 1 to page 10. Preceding Section exam content on page 1 and administrative subject confirmation Question 172 on page 2 are excluded.

Questions

Question 173

Question ID: 6406531529852

Type: MCQ · Marks: 2
PDF pages: 2

What is the primary reason for dividing the dot product of the query and key vectors by the square root of the key vector’s dimension, $\sqrt{d_k}$?

Question 173 stem: Query-key dot product scaling by square root of key dimension

Figure 1. Question 173 stem: Query-key dot product scaling by square root of key dimension

Options

To increase the numerical precision of the attention scores.

Source option ID: 6406535121364

Label added for presentation.

To reduce the overall computational cost of the attention mechanism.

Source option ID: 6406535121365

Label added for presentation.

To ensure that the sum of the attention scores for each query equals 1.

Source option ID: 6406535121366

Label added for presentation.

To avoid numerical issues and loss of gradients during training.

Source option ID: 6406535121367

Label added for presentation.

Correct answer: To avoid numerical issues and loss of gradients during training.

Source evidence: [{"page": 2, "bbox": [23.62, 552.57, 119.02, 568.89], "type": "green source option", "detail": "Option ID 6406535121367 is highlighted green (#008000) in source PDF"}]

Question 174

Question ID: 6406531529853

Type: MCQ · Marks: 2
PDF pages: 2

Which of the following is true regarding sinusoidal encoding?

Options

The encoding vector of words present at even positions in the given text sequence uses sine, and the encoding vector of words present at odd positions uses cosine.

Source option ID: 6406535121368

Label added for presentation.

The encoding vector of words present at odd positions in the given text sequence uses sine, and the encoding vector of words present at even positions uses cosine.

Source option ID: 6406535121369

Label added for presentation.

The encoding vector for each word position contains values computed using both sine (for even dimensions) and cosine (for odd dimensions).

Source option ID: 6406535121370

Label added for presentation.

Correct answer: The encoding vector for each word position contains values computed using both sine (for even dimensions) and cosine (for odd dimensions).

Source evidence: [{"page": 2, "bbox": [23.62, 768.09, 119.02, 784.41], "type": "green source option", "detail": "Option ID 6406535121370 is highlighted green (#008000) in source PDF"}]

Question 175

Question ID: 6406531529855

Type: MCQ · Marks: 2
PDF pages: 3

We have a transformer model translating English to French. Suppose the target sequence is:

Target (French): ['je', 'suis', 'heureux']

Suppose the model predicted at time step $t$:
$t = 1$: 'je'
$t = 2$: 'content'
$t = 3$: 'heureux'

If teacher forcing is used, what will be the decoder input at time step 3?

Question 175 stem: Transformer English-to-French translation teacher forcing scenario

Figure 2. Question 175 stem: Transformer English-to-French translation teacher forcing scenario

Options

‘content’

Source option ID: 6406535121375

Label added for presentation.

‘je’

Source option ID: 6406535121376

Label added for presentation.

‘suis’

Source option ID: 6406535121377

Label added for presentation.

‘heureux’

Source option ID: 6406535121378

Label added for presentation.

Correct answer: ‘suis’

Source evidence: [{"page": 3, "bbox": [23.62, 311.05, 119.02, 327.37], "type": "green source option", "detail": "Option ID 6406535121377 is highlighted green (#008000) in source PDF"}]

Question 176

Question ID: 6406531529856

Type: MCQ · Marks: 2
PDF pages: 3

We have a decoder input of length 4 tokens: [x1, x2, x3, x4].
We use masked self-attention with the following pre-softmax attention scores:

\[\begin{bmatrix} 2 & 1 & 3 & 4 \\ 3 & 2 & 1 & 4 \\ 1 & 2 & 0 & 2 \\ 0 & 1 & 1 & 2 \end{bmatrix}\]

For token x3, the pre-softmax row is [1, 2, 0, 2]. After masking, what does it become?

Question 176 stem: Decoder masked self-attention pre-softmax scores matrix and token x3 row masking

Figure 3. Question 176 stem: Decoder masked self-attention pre-softmax scores matrix and token x3 row masking

Options

[1, 2, 0, 2]

Option A: [1, 2, 0, 2]

Figure 4. Option A: [1, 2, 0, 2]

Source option ID: 6406535121379

Label added for presentation.

[1, 2, 0, 0]

Option B: [1, 2, 0, 0]

Figure 5. Option B: [1, 2, 0, 0]

Source option ID: 6406535121380

Label added for presentation.

[1, 2, 0, -\infty]

Option C: [1, 2, 0, -inf]

Figure 6. Option C: [1, 2, 0, -inf]

Source option ID: 6406535121381

Label added for presentation.

[-\infty, 2, 0, 2]

Option D: [-inf, 2, 0, 2]

Figure 7. Option D: [-inf, 2, 0, 2]

Source option ID: 6406535121382

Label added for presentation.

[-\infty, -\infty, -\infty, 2]

Option E: [-inf, -inf, -inf, 2]

Figure 8. Option E: [-inf, -inf, -inf, 2]

Source option ID: 6406535121383

Label added for presentation.

Correct answer: [1, 2, 0, -\infty]

Source evidence: [{"page": 3, "bbox": [23.62, 691.78, 119.02, 708.1], "type": "green source option", "detail": "Option ID 6406535121381 is highlighted green (#008000) in source PDF"}]

Question 177

Question ID: 6406531529857

Type: MCQ · Marks: 2
PDF pages: 4

In the context of language model training, teacher forcing is a technique where:

Options

The correct target tokens from the training data are used as input for the next time step, regardless of the model’s prediction.

Source option ID: 6406535121384

Label added for presentation.

The model’s predictions are used as the input for the next time step, even if they are incorrect.

Source option ID: 6406535121385

Label added for presentation.

The model is forced to learn without any pre-trained weights.

Source option ID: 6406535121386

Label added for presentation.

The model is forced to correct the input data.

Source option ID: 6406535121387

Label added for presentation.

Correct answer: The correct target tokens from the training data are used as input for the next time step, regardless of the model’s prediction.

Source evidence: [{"page": 4, "bbox": [23.62, 113.81, 119.02, 130.13], "type": "green source option", "detail": "Option ID 6406535121384 is highlighted green (#008000) in source PDF"}]

Question 178

Question ID: 6406531529860

Type: MCQ · Marks: 2
PDF pages: 4

Given a vocabulary of size $|V|$, how many sequences of maximum length 5 are possible?

Question 178 stem: Maximum length 5 sequences for vocabulary size |V|

Figure 9. Question 178 stem: Maximum length 5 sequences for vocabulary size |V|

Options

$|V|^5$

Option A: |V|^5

Figure 10. Option A: |V|^5

Source option ID: 6406535121397

Label added for presentation.

$5 \cdot |V|$

Option B: 5 * |V|

Figure 11. Option B: 5 * |V|

Source option ID: 6406535121398

Label added for presentation.

$\sum_{k=1}^5 |V|^k$

$Option C: sum_{k=1}^5 |V|^k$

Figure 12. Option C: sum_{k=1}^5 |V|^k

Source option ID: 6406535121399

Label added for presentation.

$\frac{|V|^5 - 1}{|V| - 1}$

Option D: (|V|^5 - 1) / (|V| - 1)

Figure 13. Option D: (|V|^5 - 1) / (|V| - 1)

Source option ID: 6406535121400

Label added for presentation.

Correct answer: $\sum_{k=1}^5 |V|^k$

Source evidence: [{"page": 4, "bbox": [23.62, 503.77, 119.02, 520.09], "type": "green source option", "detail": "Option ID 6406535121399 is highlighted green (#008000) in source PDF"}]

Question 179

Question ID: 6406531529854

Type: MCQ · Marks: 3
PDF pages: 4, 5

You have a mini-batch of 2 samples, each with 2 features:

\[X = \begin{bmatrix} 1 & 2 \\ 3 & 4 \end{bmatrix}\]

• Rows represent samples
• Columns represent features

If Batch Normalization is applied across the batch for each feature (assume $\gamma = 1, \beta = 0$ and $\epsilon = 0$), what is the resulting normalized mini-batch?

Question 179 stem: Mini-batch of 2 samples with 2 features and batch normalization

Figure 14. Question 179 stem: Mini-batch of 2 samples with 2 features and batch normalization

Options

$\begin{bmatrix} -1 & -1 \\ 1 & 1 \end{bmatrix}$

Option A: [[-1, -1], [1, 1]]

Figure 15. Option A: [[-1, -1], [1, 1]]

Source option ID: 6406535121371

Label added for presentation.

$\begin{bmatrix} -0.707 & -0.707 \\ 0.707 & 0.707 \end{bmatrix}$

Option B: [[-0.707, -0.707], [0.707, 0.707]]

Figure 16. Option B: [[-0.707, -0.707], [0.707, 0.707]]

Source option ID: 6406535121372

Label added for presentation.

$\begin{bmatrix} -0.707 & 0.707 \\ -0.707 & 0.707 \end{bmatrix}$

Option C: [[-0.707, 0.707], [-0.707, 0.707]]

Figure 17. Option C: [[-0.707, 0.707], [-0.707, 0.707]]

Source option ID: 6406535121373

Label added for presentation.

$\begin{bmatrix} 1 & 1 \\ -1 & -1 \end{bmatrix}$

Option D: [[1, 1], [-1, -1]]

Figure 18. Option D: [[1, 1], [-1, -1]]

Source option ID: 6406535121374

Label added for presentation.

Correct answer: $\begin{bmatrix} -1 & -1 \\ 1 & 1 \end{bmatrix}$

Source evidence: [{"page": 5, "bbox": [23.62, 227.0, 119.02, 243.32], "type": "green source option", "detail": "Option ID 6406535121371 is highlighted green (#008000) in source PDF"}]

Question 180

Question ID: 6406531529858

Type: MCQ · Marks: 3
PDF pages: 5

For a vocabulary of size $V$ and a fixed sequence length $L$, exhaustive decoding evaluates all possible sequences of length $L$. If we increase the sequence length to $L + 1$, how many additional sequences will need to be evaluated?

Question 180 stem: Exhaustive decoding sequence length increase from L to L+1

Figure 19. Question 180 stem: Exhaustive decoding sequence length increase from L to L+1

Options

Same, no change

Option A: Same, no change

Figure 20. Option A: Same, no change

Source option ID: 6406535121388

Label added for presentation.

$V^L$ more sequences

Option B: V^L more sequences

Figure 21. Option B: V^L more sequences

Source option ID: 6406535121389

Label added for presentation.

$(V - 1) \cdot V^L$ more sequences

Option C: (V - 1) * V^L more sequences

Figure 22. Option C: (V - 1) * V^L more sequences

Source option ID: 6406535121390

Label added for presentation.

$V$ more sequences

Option D: V more sequences

Figure 23. Option D: V more sequences

Source option ID: 6406535121391

Label added for presentation.

One more sequence

Option E: One more sequence

Figure 24. Option E: One more sequence

Source option ID: 6406535121392

Label added for presentation.

Correct answer: $(V - 1) \cdot V^L$ more sequences

Source evidence: [{"page": 5, "bbox": [23.62, 698.8, 119.02, 715.12], "type": "green source option", "detail": "Option ID 6406535121390 is highlighted green (#008000) in source PDF"}]

Question 181

Question ID: 6406531529861

Type: MCQ · Marks: 3
PDF pages: 6

For a vocabulary size of 10, how many beams will there be in greedy search, beam search with beam size 3, and exhaustive search at step 1 and step 3, respectively?

Options

Step 1:
Greedy: 1, Beam: 1, Exhaustive: 1
Step 3:
Greedy: 1, Beam: 1, Exhaustive: 1

Source option ID: 6406535121401

Label added for presentation.

Step 1:
Greedy: 1, Beam: 3, Exhaustive: 10
Step 3:
Greedy: 1, Beam: 3, Exhaustive: 10

Source option ID: 6406535121402

Label added for presentation.

Step 1:
Greedy: 1, Beam: 3, Exhaustive: 1
Step 3:
Greedy: 1, Beam: 3, Exhaustive: 100

Source option ID: 6406535121403

Label added for presentation.

Step 1:
Greedy: 1, Beam: 3, Exhaustive: 10
Step 3:
Greedy: 1, Beam: 3, Exhaustive: 1000

Source option ID: 6406535121404

Label added for presentation.

Correct answer: Step 1:
Greedy: 1, Beam: 3, Exhaustive: 10
Step 3:
Greedy: 1, Beam: 3, Exhaustive: 1000

Source evidence: [{"page": 6, "bbox": [23.62, 359.08, 119.02, 375.4], "type": "green source option", "detail": "Option ID 6406535121404 is highlighted green (#008000) in source PDF"}]

Question 182

Question ID: 6406531529859

Type: MSQ · Marks: 2
PDF pages: 6

Which of the following decoding strategies is/are inherently non-deterministic?

Options

Beam

Source option ID: 6406535121393

Label added for presentation.

Greedy

Source option ID: 6406535121394

Label added for presentation.

Nucleus

Source option ID: 6406535121395

Label added for presentation.

Exhaustive

Source option ID: 6406535121396

Label added for presentation.

Correct answer: Nucleus

Source evidence: [{"page": 6, "bbox": [23.62, 648.88, 119.02, 665.2], "type": "green source option", "detail": "Option ID 6406535121395 is highlighted green (#008000) in source PDF"}]

Question 183

Question ID: 6406531529862

Type: MSQ · Marks: 2
PDF pages: 6, 7

Which of the following statements about multiple attention heads in a Transformer are true?

Options

They allow learning diverse local/global relations.

Source option ID: 6406535121405

Label added for presentation.

They reduce memory usage compared to one head.

Source option ID: 6406535121406

Label added for presentation.

Increasing the number of heads while keeping dmodel fixed, decreases the dimensionality handled by each head.

Source option ID: 6406535121407

Label added for presentation.

All heads necessarily learn completely independent information about the input.

Source option ID: 6406535121408

Label added for presentation.

Correct answer: They allow learning diverse local/global relations.; Increasing the number of heads while keeping dmodel fixed, decreases the dimensionality handled by each head.

Source evidence: [{"page": 7, "bbox": [23.62, 17.5, 119.02, 33.82], "type": "green source option", "detail": "Option ID 6406535121405 is highlighted green (#008000) in source PDF"}, {"page": 7, "bbox": [23.62, 59.15, 119.02, 75.47], "type": "green source option", "detail": "Option ID 6406535121407 is highlighted green (#008000) in source PDF"}]

Question 184

Question ID: 6406531529863

Type: MSQ · Marks: 4
PDF pages: 7

Which of the following statements about the Transformer architecture are true?

Options

Masking (look-ahead mask) is applied in every decoder self-attention layer, not just the first one.

Source option ID: 6406535121409

Label added for presentation.

Token embeddings are used in the encoder but not in the decoder.

Source option ID: 6406535121410

Label added for presentation.

Positional encoding is required only in the encoder, not in the decoder.

Source option ID: 6406535121411

Label added for presentation.

Token embeddings are used in both the encoder and the decoder.

Source option ID: 6406535121412

Label added for presentation.

Positional encodings are added to the input embeddings in both the encoder and the decoder.

Source option ID: 6406535121413

Label added for presentation.

Correct answer: Masking (look-ahead mask) is applied in every decoder self-attention layer, not just the first one.; Token embeddings are used in both the encoder and the decoder.; Positional encodings are added to the input embeddings in both the encoder and the decoder.

Source evidence: [{"page": 7, "bbox": [23.62, 311.82, 119.02, 328.14], "type": "green source option", "detail": "Option ID 6406535121409 is highlighted green (#008000) in source PDF"}, {"page": 7, "bbox": [23.62, 390.6, 119.02, 406.92], "type": "green source option", "detail": "Option ID 6406535121412 is highlighted green (#008000) in source PDF"}, {"page": 7, "bbox": [23.62, 411.42, 119.02, 427.74], "type": "green source option", "detail": "Option ID 6406535121413 is highlighted green (#008000) in source PDF"}]

Question 185

Question ID: 6406531529864

Type: MSQ · Marks: 4
PDF pages: 7, 8

A transformer decoder uses a vocabulary of size 5. At step 1, the predicted token probabilities are:

Token	Probability
A	0.40
B	0.30
C	0.20
D	0.05
E	0.05

Which statement(s) is/are correct?

Question 185 stem: Transformer decoder predicted token probabilities at step 1

Figure 25. Question 185 stem: Transformer decoder predicted token probabilities at step 1

Options

Beam search with beam size 2 will keep either A or B as per probability whereas Top-2 sampling will keep both A and B.

Source option ID: 6406535121414

Label added for presentation.

Beam search with beam size 2 will keep both A and B whereas Top-2 sampling will keep either A or B.

Source option ID: 6406535121415

Label added for presentation.

Beam search with beam size 2 will randomly choose any two tokens, whereas Top-2 sampling will always pick the two least probable tokens.

Source option ID: 6406535121416

Label added for presentation.

Beam search with beam size 2 will always keep the top-2 most probable tokens, whereas Top-2 sampling will restrict choices to top-2 tokens and then sample one probabilistically.

Source option ID: 6406535121417

Label added for presentation.

Correct answer: Beam search with beam size 2 will keep both A and B whereas Top-2 sampling will keep either A or B.; Beam search with beam size 2 will always keep the top-2 most probable tokens, whereas Top-2 sampling will restrict choices to top-2 tokens and then sample one probabilistically.

Source evidence: [{"page": 7, "bbox": [23.62, 780.8, 119.02, 797.12], "type": "green source option", "detail": "Option ID 6406535121415 is highlighted green (#008000) in source PDF"}, {"page": 8, "bbox": [23.62, 54.64, 119.02, 70.96], "type": "green source option", "detail": "Option ID 6406535121417 is highlighted green (#008000) in source PDF"}]

Question 186

Question ID: 6406531529865

Type: SA · Marks: 3
PDF pages: 8

A Transformer with single self-attention layer has batch size $B = 4$, context length (sequence length) $T = 12$, and number of heads $h = 2$. How many scalar attention scores are computed in total across two heads and all sequences for one forward pass?

Question 186 stem: Total scalar attention scores calculation for B=4, T=12, h=2

Figure 26. Question 186 stem: Total scalar attention scores calculation for B=4, T=12, h=2

Correct answer: 1152

Source evidence: [{"page": 8, "bbox": [22.87, 439.63, 61.41, 457.99], "type": "explicit field", "detail": "Official key provides Possible Answers: 1152"}]

Question 187

Question ID: 6406531529866

Type: SA · Marks: 3
PDF pages: 8

A Transformer model processes a sequence of $T = 1000$ tokens. Within the self-attention layer, the attention scores are stored as 32-bit floating-point numbers (float32). We compare two configurations that differ in the number of attention heads, $h$:

Model A: $h = 8$
Model B: $h = 16$

Using the given conversion factors (1 byte = 8 bits, 1 MB = $10^6$ bytes), calculate the percentage increase in memory (in MB) required to store all attention scores when moving from Model A to Model B.

Question 187 stem: Percentage increase in memory from Model A (h=8) to Model B (h=16)

Figure 27. Question 187 stem: Percentage increase in memory from Model A (h=8) to Model B (h=16)

Correct answer: 99 to 101

Source evidence: [{"page": 8, "bbox": [22.87, 761.64, 91.56, 780.0], "type": "explicit field", "detail": "Official key provides Possible Answers: 99 to 101"}]

Question 188

Question ID: 6406531529867

Type: SA · Marks: 2
PDF pages: 9

In a Transformer model with hidden size $d_{\text{model}} = 512$, number of heads $h = 8$, and key/query dimension $d_k = 64$, how many trainable parameters are there in the projection matrix $W_Q$ for a given head?

Question 188 stem: Trainable parameters in projection matrix W_Q for head dimension d_k=64

Figure 28. Question 188 stem: Trainable parameters in projection matrix W_Q for head dimension d_k=64

Correct answer: 32768

Source evidence: [{"page": 9, "bbox": [22.87, 296.88, 69.12, 315.24], "type": "explicit field", "detail": "Official key provides Possible Answers: 32768"}]

Shared context 189 to 192

Question ID: 6406531529868

Type: COMPREHENSION
PDF pages: 9

You are analyzing a GPT-style decoder-only Transformer trained as a causal language model with the following setup:

• Sequence length: $T = 1024$ tokens per input sequence.
• Batch size: $B = 32$.
• Hidden dimension: $d_{\text{model}} = 768$.
• Number of attention heads: $h = 8$.
• $d_k = \frac{d_{\text{model}}}{h}$
• Training uses next-token prediction with teacher forcing and a causal mask.
• The input sequence does not include a beginning-of-sequence (BOS) token.

Answer the given sub-questions:

Comprehension passage: GPT-style decoder-only Transformer causal language model setup

Figure 29. Comprehension passage: GPT-style decoder-only Transformer causal language model setup

Question 189

Question ID: 6406531529869

Type: SA · Marks: 3
PDF pages: 9, 10

Shared context: ctx-189-192

How many next-token prediction targets are generated during training a full batch?

Correct answer: 32736

Source evidence: [{"page": 10, "bbox": [22.87, 171.31, 61.41, 189.67], "type": "explicit field", "detail": "Official key provides Possible Answers: 32736"}]

Question 190

Question ID: 6406531529870

Type: SA · Marks: 2
PDF pages: 10

Shared context: ctx-189-192

For one sequence, how many non-zero attention scores remain in the masked attention score matrix (per head)?

Correct answer: 524800

Source evidence: [{"page": 10, "bbox": [22.87, 421.51, 69.12, 439.87], "type": "explicit field", "detail": "Official key provides Possible Answers: 524800"}]

Question 191

Question ID: 6406531529871

Type: SA · Marks: 2
PDF pages: 10

Shared context: ctx-189-192

The output projection matrix $W_O$ maps the concatenated head outputs dimension back to $d_{\text{model}}$ dimension. How many trainable parameters does $W_O$ contain?

Question 191 stem: Output projection matrix W_O trainable parameters question

Figure 30. Question 191 stem: Output projection matrix W_O trainable parameters question

Correct answer: 589824

Source evidence: [{"page": 10, "bbox": [22.87, 731.82, 69.12, 750.18], "type": "explicit field", "detail": "Official key provides Possible Answers: 589824"}]

Question 192

Question ID: 6406531529872

Type: MSQ · Marks: 2
PDF pages: 10

Shared context: ctx-189-192

[Question stem truncated in source PDF; document ends after question header]

Correct answer: Answer not verifiable from the source PDF

Source evidence: [{"page": 10, "bbox": [22.87, 772.5, 458.53, 808.14], "type": "absent", "detail": "Question 192 header appears at the bottom of final page 10 of the source PDF, but the question stem, options, and official answer key are truncated and absent before the document ends"}]

Answer key

Source question	Source identity	Correct answer	PDF pages
173	6406531529852	To avoid numerical issues and loss of gradients during training.	2
174	6406531529853	The encoding vector for each word position contains values computed using both sine (for even dimensions) and cosine (for odd dimensions).	2
175	6406531529855	‘suis’	3
176	6406531529856	[1, 2, 0, -\infty]	3
177	6406531529857	The correct target tokens from the training data are used as input for the next time step, regardless of the model’s prediction.	4
178	6406531529860	$\sum_{k=1}^5 \\|V\\|^k$	4
179	6406531529854	$\begin{bmatrix} -1 & -1 \\ 1 & 1 \end{bmatrix}$	4, 5
180	6406531529858	$(V - 1) \cdot V^L$ more sequences	5
181	6406531529861	Step 1: Greedy: 1, Beam: 3, Exhaustive: 10 Step 3: Greedy: 1, Beam: 3, Exhaustive: 1000	6
182	6406531529859	Nucleus	6
183	6406531529862	They allow learning diverse local/global relations.; Increasing the number of heads while keeping dmodel fixed, decreases the dimensionality handled by each head.	6, 7
184	6406531529863	Masking (look-ahead mask) is applied in every decoder self-attention layer, not just the first one.; Token embeddings are used in both the encoder and the decoder.; Positional encodings are added to the input embeddings in both the encoder and the decoder.	7
185	6406531529864	Beam search with beam size 2 will keep both A and B whereas Top-2 sampling will keep either A or B.; Beam search with beam size 2 will always keep the top-2 most probable tokens, whereas Top-2 sampling will restrict choices to top-2 tokens and then sample one probabilistically.	7, 8
186	6406531529865	1152	8
187	6406531529866	99 to 101	8
188	6406531529867	32768	9
189	6406531529869	32736	9, 10
190	6406531529870	524800	10
191	6406531529871	589824	10
192	6406531529872	Answer not verifiable from the source PDF	10

Extraction notes

Status: verified-source-limitation
Count reconciliation: Section 10 declares 18 portal questions and 50 section marks. The 18 portal containers comprise: 1 administrative confirmation question (Q172, 0 marks, excluded), 16 standalone questions (Q173: 2 marks, Q174: 2 marks, Q175: 2 marks, Q176: 2 marks, Q177: 2 marks, Q178: 2 marks, Q179: 3 marks, Q180: 3 marks, Q181: 3 marks, Q182: 2 marks, Q183: 2 marks, Q184: 4 marks, Q185: 4 marks, Q186: 3 marks, Q187: 3 marks, Q188: 2 marks; total standalone = 41 marks), and 1 comprehension portal container (Q189-192: Q189: 3 marks, Q190: 2 marks, Q191: 2 marks, Q192: 2 marks; total comprehension = 9 marks). Total academic questions = 16 standalone + 4 comprehension subquestions = 20 academic questions. Total marks = 41 + 9 = 50 marks, exactly reconciling the declared section marks and 18 portal containers (1 confirmation + 16 standalone + 1 comprehension container = 18 portal questions).

Question 192 (MSQ, 2 marks, Question Id: 6406531529872) appears at the bottom of Page 10 with only its question header ('Question Number : 192 Question Id : 6406531529872 Question Type : MSQ Correct Marks : 2 Max. Selectable Options : 0'). The question stem, options, and official answer are truncated before the document ends at page 10. In accordance with the extraction contract, the answer is recorded with kind 'absent' and text exactly 'Answer not verifiable from the source PDF'.

Excluded source entries:

{"classification": "preceding-section-content", "page": 1, "bbox": [0.0, 0.0, 595.0, 380.0], "reason": "Preceding section examination question regarding contours of function f with options 6406535121340 and 6406535121341, appearing prior to Section 10 LLM header", "identity": "preceding-section"}
{"classification": "subject-confirmation", "page": 2, "bbox": [22.87, 36.07, 569.03, 285.0], "reason": "Administrative subject confirmation asking candidate to confirm subject and cross check hall ticket", "identity": "172"}