LLM — da5004_2025T2_ET_FN

Source: da5004_2025T2_ET_FN.pdf · Academic questions: 17 · Separately counted children: 0
Provenance: Large_Language_Models/da5004_2025T2_ET_FN.pdf · SHA-256 838e6b1ea02e15a09eb284d2f3db7950d356652b030197a1e328941d9dbf50f3

Scope: Large Language Models (LLM) section (Section 6, Section Id: 64065399797) spanning from page 1 at y=237.40 to page 12 at y=426.10 (Question 154 answer). Preceding Section 5 content on page 1 (y=0 to y=232.20) and succeeding Section 7 (DL(CV)) content on page 12 (y=431.30 to y=841.89) are excluded as out of subject scope.

Questions

Question 138

Question ID: 6406531422514

Type: MCQ · Marks: 2
PDF pages: 2

Given the input string:

moonlight

And the following vocabulary of subword tokens with their corresponding log-probabilities:

Subword	Log-Probability (base e)
moon	-0.4
light	-1.0
moonlight	-1.8
moo	-0.3
nlight	-0.6
n	-2.0
li	-0.5
ght	-0.5

Using the Viterbi algorithm (as used in the SentencePiece tokenizer), determine the most probable tokenization of the input string. The probability of a tokenized sequence is the sum of the log-probabilities of the selected subwords. Only subwords from the vocabulary may be used.

Input string, subword vocabulary log-probabilities table, and Viterbi tokenization problem description

Figure 1. Input string, subword vocabulary log-probabilities table, and Viterbi tokenization problem description

Options

[moonlight]

Option A token sequence [moonlight]

Figure 2. Option A token sequence [moonlight]

Source option ID: 6406534756305

Label added for presentation.

[moo, nlight]

Option B token sequence [noo, nlight]

Figure 3. Option B token sequence [noo, nlight]

Source option ID: 6406534756306

Label added for presentation.

[moon, light]

Option C token sequence [moon, light]

Figure 4. Option C token sequence [moon, light]

Source option ID: 6406534756307

Label added for presentation.

[moon, li, ght]

Option D token sequence [moon, li, ght]

Figure 5. Option D token sequence [moon, li, ght]

Source option ID: 6406534756308

Label added for presentation.

Correct answer: [moo, nlight]

Source evidence: [{"page": 2, "bbox": [119.02, 707.13, 131.02, 719.13], "type": "green source option", "detail": "Green marked option 6406534756306 (B)"}]

Question 139

Question ID: 6406531422515

Type: MCQ · Marks: 2
PDF pages: 3

Match the given attention masks with their corresponding variants of attention mechanisms. In each figure, the colored cells indicate positions where a query token ($Q_i$) attends to a key token ($K_j$), while the white cells represent positions that are not attended. Choose the correct option.

Attention mask diagrams (a), (b), (c), (d) for attention pattern matching

Figure 6. Attention mask diagrams (a), (b), (c), (d) for attention pattern matching

Options

(a) : Random Local Attention,
(b) : Strided Local Attention,
(c) : Sparse Block Attention,
(d) : Local + Global Attention

Option A attention mechanism matching list

Figure 7. Option A attention mechanism matching list

Source option ID: 6406534756309

Label added for presentation.

(a) : Strided Local Attention,
(b) : Local + Global Attention,
(c) : Random Local Attention,
(d) : Sparse Block Attention

Option B attention mechanism matching list

Figure 8. Option B attention mechanism matching list

Source option ID: 6406534756310

Label added for presentation.

(a) : Strided Local Attention,
(b) : Random Local Attention,
(c) : Local + Global Attention,
(d) : Sparse Block Attention

Option C attention mechanism matching list

Figure 9. Option C attention mechanism matching list

Source option ID: 6406534756311

Label added for presentation.

(a) : Sparse Block Attention,
(b) : Local + Global Attention,
(c) : Strided Local Attention,
(d) : Random Local Attention

Option D attention mechanism matching list

Figure 10. Option D attention mechanism matching list

Source option ID: 6406534756312

Label added for presentation.

Correct answer: (a) : Strided Local Attention,
(b) : Local + Global Attention,
(c) : Random Local Attention,
(d) : Sparse Block Attention

Source evidence: [{"page": 3, "bbox": [119.02, 506.43, 131.02, 518.43], "type": "green source option", "detail": "Green marked option 6406534756310 (B)"}]

Question 140

Question ID: 6406531422516

Type: MSQ · Marks: 3
PDF pages: 4

Which of the following statements about subword tokenizers (e.g., Byte Pair Encoding,
SentencePiece) is true?

Options

They always split words into individual characters.

Source option ID: 6406534756313

Label added for presentation.

They help handle rare words by breaking them into smaller units.

Source option ID: 6406534756314

Label added for presentation.

They require the vocabulary to contain every possible word in the language.

Source option ID: 6406534756315

Label added for presentation.

They can reduce the overall vocabulary size compared to word-level
tokenizers.

Source option ID: 6406534756316

Label added for presentation.

Correct answer: They help handle rare words by breaking them into smaller units.; They can reduce the overall vocabulary size compared to word-level
tokenizers.

Source evidence: [{"page": 4, "bbox": [119.02, 152.06, 131.02, 164.06], "type": "green source option", "detail": "Green marked option 6406534756314 (B)"}, {"page": 4, "bbox": [119.02, 193.7, 131.02, 205.7], "type": "green source option", "detail": "Green marked option 6406534756316 (D)"}]

Question 141

Question ID: 6406531422517

Type: MSQ · Marks: 3
PDF pages: 4

A GPT-based model is fine-tuned for story generation. During evaluation, you test different
decoding strategies on the same prompt:

Method A: Every run produces the exact same story, even when executed multiple times.
Method B: The story changes slightly in each run, but overall structure and key events remain
similar.
Method C: The story is wildly different each time, sometimes even changing the genre.

Based on the observations, select the most likely decoding strategies for each method:

Options

Method A: Greedy decoding

Source option ID: 6406534756317

Label added for presentation.

Method A: Top-K sampling with K = 50

Source option ID: 6406534756318

Label added for presentation.

Method B: Top-P sampling with p = 0.9

Source option ID: 6406534756319

Label added for presentation.

Method C: Top-P sampling with p = 0.5 and high temperature T = 1.5

Source option ID: 6406534756320

Label added for presentation.

Method C: Beam search with K = 5

Source option ID: 6406534756321

Label added for presentation.

Correct answer: Method A: Greedy decoding; Method B: Top-P sampling with p = 0.9; Method C: Top-P sampling with p = 0.5 and high temperature T = 1.5

Source evidence: [{"page": 4, "bbox": [119.02, 481.82, 131.02, 493.82], "type": "green source option", "detail": "Green marked option 6406534756317 (A)"}, {"page": 4, "bbox": [119.02, 523.49, 131.02, 535.49], "type": "green source option", "detail": "Green marked option 6406534756319 (C)"}, {"page": 4, "bbox": [119.02, 544.34, 131.02, 556.34], "type": "green source option", "detail": "Green marked option 6406534756320 (D)"}]

Question 142

Question ID: 6406531422518

Type: MSQ · Marks: 3
PDF pages: 4

Which of the following are valid design choices when building a transformer-based language
model?

Options

Number of transformer layers

Source option ID: 6406534756322

Label added for presentation.

Learning rate schedule

Source option ID: 6406534756323

Label added for presentation.

Font style of the source code

Source option ID: 6406534756324

Label added for presentation.

Vocabulary size

Source option ID: 6406534756325

Label added for presentation.

Type of normalization (e.g., LayerNorm)

Source option ID: 6406534756326

Label added for presentation.

Correct answer: Number of transformer layers; Learning rate schedule; Vocabulary size; Type of normalization (e.g., LayerNorm)

Source evidence: [{"page": 4, "bbox": [119.02, 722.75, 131.02, 734.75], "type": "green source option", "detail": "Green marked option 6406534756322 (A)"}, {"page": 4, "bbox": [119.02, 743.57, 131.02, 755.57], "type": "green source option", "detail": "Green marked option 6406534756323 (B)"}, {"page": 4, "bbox": [119.02, 785.21, 131.02, 797.21], "type": "green source option", "detail": "Green marked option 6406534756325 (D)"}, {"page": 4, "bbox": [119.02, 806.03, 131.02, 818.03], "type": "green source option", "detail": "Green marked option 6406534756326 (E)"}]

Question 143

Question ID: 6406531422519

Type: MSQ · Marks: 3
PDF pages: 5

Which of the following statements about Top-k and Nucleus (Top-p) sampling are correct?

Options

Top-k keeps only the k most probable tokens before sampling.

Source option ID: 6406534756327

Label added for presentation.

Top-p dynamically chooses the smallest set of tokens whose cumulative
probability exceeds p.

Source option ID: 6406534756328

Label added for presentation.

Top-k is always a subset of Top-p.

Source option ID: 6406534756329

Label added for presentation.

Top-p always has a fixed number of candidate tokens.

Source option ID: 6406534756330

Label added for presentation.

Correct answer: Top-k keeps only the k most probable tokens before sampling.; Top-p dynamically chooses the smallest set of tokens whose cumulative
probability exceeds p.

Source evidence: [{"page": 5, "bbox": [119.02, 134.24, 131.02, 146.24], "type": "green source option", "detail": "Green marked option 6406534756327 (A)"}, {"page": 5, "bbox": [119.02, 155.06, 131.02, 167.06], "type": "green source option", "detail": "Green marked option 6406534756328 (B)"}]

Question 144

Question ID: 6406531422520

Type: MSQ · Marks: 3
PDF pages: 5

Which of the following preprocessing steps are commonly required when preparing data for
transformer models like BERT?

Options

Lowercasing text (for uncased models)

Source option ID: 6406534756331

Label added for presentation.

Splitting into subword units using BPE or WordPiece

Source option ID: 6406534756332

Label added for presentation.

Removing all punctuation marks

Source option ID: 6406534756333

Label added for presentation.

Adding position indices for each token

Source option ID: 6406534756334

Label added for presentation.

Correct answer: Lowercasing text (for uncased models); Splitting into subword units using BPE or WordPiece; Adding position indices for each token

Source evidence: [{"page": 5, "bbox": [119.02, 370.58, 131.02, 382.58], "type": "green source option", "detail": "Green marked option 6406534756331 (A)"}, {"page": 5, "bbox": [119.02, 391.4, 131.02, 403.4], "type": "green source option", "detail": "Green marked option 6406534756332 (B)"}, {"page": 5, "bbox": [119.02, 433.04, 131.02, 445.04], "type": "green source option", "detail": "Green marked option 6406534756334 (D)"}]

Question 145

Question ID: 6406531422521

Type: SA · Marks: 3
PDF pages: 5, 6

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

Token	Logit
cat	1.5
sat	0.5
on	0.0
mat	-0.5

You apply Low Temperature Sampling with temperature $T = 2.0$. After applying the softmax on the temperature-scaled logits, what is the probability assigned to the token “cat”? (Give your answer up to 2 decimal places)

Vocabulary logits table and low temperature sampling setup for token cat

Figure 11. Vocabulary logits table and low temperature sampling setup for token cat

Correct answer: 0.38 to 0.44

Source evidence: [{"page": 6, "bbox": [22.87, 105.0, 160.0, 130.0], "type": "explicit field", "detail": "Official source Range answer: 0.38 to 0.44"}]

Shared context 146 to 147

Question ID: 6406531422522

Portal container ID: 6406531422522

Type: Comprehension
PDF pages: 6

A vocabulary $V = \{\text{tea}, \text{you}, \text{enjoy}, \text{often}\}$ is associated with the following embedding matrix:

\[E = \begin{bmatrix} 0 & 1 \\ 1 & 0 \\ 1 & 1 \\ -1 & 1 \end{bmatrix}\]

Words are indexed in the order shown above: “tea” (index 0), “you” (1), “enjoy” (2), and “often” (3). No positional encodings are used.

The parameters of the attention layer are:

\[W_Q = \begin{bmatrix} 0.5 & 1 \\ -0.5 & 0.5 \end{bmatrix}, \quad W_K = \begin{bmatrix} 1 & 0.5 \\ -0.5 & 1 \end{bmatrix}, \quad W_V = \begin{bmatrix} 1 & 0.2 \\ 0.5 & 0.5 \end{bmatrix}, \quad W_O = \begin{bmatrix} 0.5 & 0.5 \\ 0.5 & 1 \end{bmatrix}\]

For the input sequence “you enjoy tea often”. The computed attention matrix $A = \text{softmax}\left(\frac{Q K^T}{\sqrt{d_k}}\right)$ is given as:

\[A = \begin{bmatrix} 0.55 & 0.32 & 0.11 & 0.02 \\ 0.43 & 0.25 & 0.21 & 0.10 \\ 0.39 & 0.39 & 0.16 & 0.07 \\ 0.24 & 0.40 & 0.20 & 0.17 \end{bmatrix}\]

Answer the given subquestions. If you believe the information is insufficient, enter -1 as your answer.

Vocabulary embeddings, projection matrices W_Q, W_K, W_V, W_O, and computed attention matrix A

Figure 12. Vocabulary embeddings, projection matrices W_Q, W_K, W_V, W_O, and computed attention matrix A

Question 146

Question ID: 6406531422523

Portal container ID: 6406531422522

Type: SA · Marks: 3
PDF pages: 6, 7

Shared context: ctx-146-147

For the input “you enjoy tea often”, compute the final representation of the word “tea” after the
attention layer (i.e., after applying WQ, WK, WV , attention weights, and WO). Enter the sum of the
elements in the resulting vector.

Correct answer: 0.98 to 1.06

Source evidence: [{"page": 7, "bbox": [22.87, 180.0, 160.0, 205.0], "type": "explicit field", "detail": "Official source Range answer: 0.98 to 1.06"}]

Question 147

Question ID: 6406531422524

Portal container ID: 6406531422522

Type: SA · Marks: 2
PDF pages: 7

Shared context: ctx-146-147

Suppose the input sentence is “tea you enjoy often”. Using the same matrices and processing
method, what is the attention score (i.e., softmax entry from A) for the query word “tea” attending
to key word “often”?

Correct answer: 0.04 to 0.10

Source evidence: [{"page": 7, "bbox": [22.87, 420.0, 160.0, 445.0], "type": "explicit field", "detail": "Official source Range answer: 0.04 to 0.10"}]

Shared context 148 to 150

Question ID: 6406531422530

Portal container ID: 6406531422530

Type: Comprehension
PDF pages: 7, 8

Suppose we have context length $T = 30$ and the number of blocks $n = 5$ for the block attention mechanism.

The masking matrix $M$ is defined as:

\[M_{ij} = \begin{cases} 1 & \text{if } \pi\left(\lfloor i/6 \rfloor\right) = \lfloor j/6 \rfloor \\ 0 & \text{otherwise} \end{cases}\]

where $i$ is row indices and $j$ is columns indexes both starting from 0.

Here, $\pi$ is a permutation of $\{0, 1, \dots, (n-1)\}$, which determines the allowed query-key block pairs. Specifically, if a query token $i$ lies in block $r = \lfloor i/6 \rfloor$, then it can attend only to key tokens in block $\pi(r)$.

Equivalently, if we partition queries and keys into blocks:

\[Q = (Q_0, Q_1, \dots, Q_{n-1}), \quad K = (K_{\pi(0)}, K_{\pi(1)}, \dots, K_{\pi(n-1)})\]

then block $Q_r$ attends only to block $K_{\pi(r)}$.

This block-restricted attention reduces computation by limiting interactions to the query-key block pairs specified by $\pi$.

Answer the given sub-questions using the attention mask $M \in \{0, 1\}^{T \times T}$ image, where black squares represent entries equal to 1 and white squares represent entries equal to 0.

Block attention mechanism definition, permutation matrix, and attention mask M image

Figure 13. Block attention mechanism definition, permutation matrix, and attention mask M image

Question 148

Question ID: 6406531422531

Portal container ID: 6406531422530

Type: MCQ · Marks: 2
PDF pages: 8, 9

Shared context: ctx-148-150

Choose the correct representation of $\pi$ for the given mask image.

Prompt to choose correct representation of permutation pi for given mask image

Figure 14. Prompt to choose correct representation of permutation pi for given mask image

Options

$\pi = (0, 2, 3, 1, 4)$

Option A permutation mapping

Figure 15. Option A permutation mapping

Source option ID: 6406534756348

Label added for presentation.

$\pi = (0, 6, 12, 18, 24)$

Option B permutation mapping

Figure 16. Option B permutation mapping

Source option ID: 6406534756349

Label added for presentation.

$\pi = (0, 3, 1, 2, 4)$

Option C permutation mapping

Figure 17. Option C permutation mapping

Source option ID: 6406534756350

Label added for presentation.

All of these

Option D permutation mapping

Figure 18. Option D permutation mapping

Source option ID: 6406534756351

Label added for presentation.

Correct answer: $\pi = (0, 3, 1, 2, 4)$

Source evidence: [{"page": 9, "bbox": [119.02, 111.15, 131.02, 123.15], "type": "green source option", "detail": "Green marked option 6406534756350 (C)"}]

Question 149

Question ID: 6406531422532

Portal container ID: 6406531422530

Type: SA · Marks: 1
PDF pages: 9

Shared context: ctx-148-150

How many permutations of $\pi$ are possible for $n = 5$?

Prompt for total number of permutations of pi for n = 5

Figure 19. Prompt for total number of permutations of pi for n = 5

Correct answer: 120

Source evidence: [{"page": 9, "bbox": [22.87, 410.0, 100.0, 435.0], "type": "explicit field", "detail": "Official source Equal answer: 120"}]

Question 150

Question ID: 6406531422533

Portal container ID: 6406531422530

Type: SA · Marks: 2
PDF pages: 9

Shared context: ctx-148-150

What percentage of entries in the attention matrix are zero (i.e., the sparsity)?

Correct answer: 80

Source evidence: [{"page": 9, "bbox": [22.87, 610.0, 100.0, 635.0], "type": "explicit field", "detail": "Official source Equal answer: 80"}]

Shared context 151 to 154

Question ID: 6406531422525

Portal container ID: 6406531422525

Type: Comprehension
PDF pages: 9, 10

Consider the short sentence:
“large language models are cool”
A transformer processes this sequence using the naive relative positional embedding method. The transformer’s embedding dimension is $d_{\text{model}} = 4$, and the number of tokens is $T = 5$. We use zero-based indexing for token positions:

0: large, 1: language, 2: models, 3: are, 4: cool

Note: In this context, the relative position ($k$) between two tokens is:
Relative position = $j - i$
rows = current token $i$, cols = other token $j$
Positive values mean the other token is after the current token, negative values mean it is before, and 0 means it is the same token.

Token embeddings (rows correspond to tokens in order):

\[X = \begin{bmatrix} 1 & 1 & 1 & 1 \\ 2 & 2 & 2 & 2 \\ 3 & 3 & 3 & 3 \\ 4 & 4 & 4 & 4 \\ 5 & 5 & 5 & 5 \end{bmatrix}\]

Denote the $i$-th row by $x_i$.

Positional embedding definition: For a relative position $k \in \mathbb{Z}$,

\[p_{j-i} = \left[ \frac{j-i}{10}, \frac{j-i}{10}, \frac{j-i}{10}, \frac{j-i}{10} \right]\]

Based on the above data, answer the given subquestions.

Relative positional embedding definition, token embeddings matrix X, and relative position vector formula

Figure 20. Relative positional embedding definition, token embeddings matrix X, and relative position vector formula

Question 151

Question ID: 6406531422526

Portal container ID: 6406531422525

Type: MCQ · Marks: 1
PDF pages: 10, 11

Shared context: ctx-151-154

Choose the sequence of relative positions (j — i) for the token "models" with respect to each token in the sentence.

Prompt for relative positions sequence for token models

Figure 21. Prompt for relative positions sequence for token models

Options

(0, 1, 2, 3, 4)

Source option ID: 6406534756338

Label added for presentation.

(-1, 0, 1, 2, 3)

Source option ID: 6406534756339

Label added for presentation.

(-2, -1, 0, 1, 2)

Source option ID: 6406534756340

Label added for presentation.

(-3, -2, -1, 0, 1)

Source option ID: 6406534756341

Label added for presentation.

Correct answer: (-2, -1, 0, 1, 2)

Source evidence: [{"page": 11, "bbox": [119.02, 39.14, 131.02, 51.14], "type": "green source option", "detail": "Green marked option 6406534756340 (C)"}]

Question 152

Question ID: 6406531422527

Portal container ID: 6406531422525

Type: MCQ · Marks: 1
PDF pages: 11

Shared context: ctx-151-154

Choose the value of $p_{j-i}$ for the current token “model” and other token “large”.

$Prompt for positional embedding vector p_{j-i} for current token model and other token large$

Figure 22. Prompt for positional embedding vector p_{j-i} for current token model and other token large

Options

[0.2, 0.2, 0.2, 0.2]

Option A vector [0.2, 0.2, 0.2, 0.2]

Figure 23. Option A vector [0.2, 0.2, 0.2, 0.2]

Source option ID: 6406534756342

Label added for presentation.

[-0.2, -0.2, -0.2, -0.2]

Option B vector [-0.2, -0.2, -0.2, -0.2]

Figure 24. Option B vector [-0.2, -0.2, -0.2, -0.2]

Source option ID: 6406534756343

Label added for presentation.

[-0.3, -0.3, -0.3, -0.3]

Option C vector [-0.3, -0.3, -0.3, -0.3]

Figure 25. Option C vector [-0.3, -0.3, -0.3, -0.3]

Source option ID: 6406534756344

Label added for presentation.

[-0.4, -0.4, -0.4, -0.4]

Option D vector [-0.4, -0.4, -0.4, -0.4]

Figure 26. Option D vector [-0.4, -0.4, -0.4, -0.4]

Source option ID: 6406534756345

Label added for presentation.

Correct answer: [-0.2, -0.2, -0.2, -0.2]

Source evidence: [{"page": 11, "bbox": [119.02, 295.89, 131.02, 307.89], "type": "green source option", "detail": "Green marked option 6406534756343 (B)"}]

Question 153

Question ID: 6406531422528

Portal container ID: 6406531422525

Type: SA · Marks: 3
PDF pages: 11, 12

Shared context: ctx-151-154

With a naive implementation, for each token embedding, we need a way to combine (say, addition) relative position embeddings. For token index $i \in \{0, \dots, T-1\}$,

\[h_i = x_i + \sum_{j=0}^{T-1} p_{j-i}\]

Using the naive formula above, compute the final combined embedding $h_i$ for the token "language" (i.e., $i = 1$). Submit the sum of all the elements.

Prompt for combined embedding h_i calculation for token language

Figure 27. Prompt for combined embedding h_i calculation for token language

Correct answer: 10

Source evidence: [{"page": 12, "bbox": [22.87, 55.0, 100.0, 75.0], "type": "explicit field", "detail": "Official source Equal answer: 10"}]

Question 154

Question ID: 6406531422529

Portal container ID: 6406531422525

Type: SA · Marks: 3
PDF pages: 12

Shared context: ctx-151-154

Compute $e_{ij}$ for the current token “models” and other token “large”.

\[e_{ij} = x_i W_Q (x_j W_K + p_{j-i})^T\]

Assume $W_Q = I$ and $W_K = 2I$ where $I$ is an identity matrix. Answer $e_{ij}$ correct upto 1 digit after the decimal.

$Prompt for relative attention score e_{ij} calculation for tokens models and large$

Figure 28. Prompt for relative attention score e_{ij} calculation for tokens models and large

Correct answer: 21.3 to 22.0

Source evidence: [{"page": 12, "bbox": [22.87, 405.0, 150.0, 430.0], "type": "explicit field", "detail": "Official source Range answer: 21.3 to 22.0"}]

Answer key

Source question	Source identity	Correct answer	PDF pages
138	6406531422514	[moo, nlight]	2
139	6406531422515	(a) : Strided Local Attention, (b) : Local + Global Attention, (c) : Random Local Attention, (d) : Sparse Block Attention	3
140	6406531422516	They help handle rare words by breaking them into smaller units.; They can reduce the overall vocabulary size compared to word-level tokenizers.	4
141	6406531422517	Method A: Greedy decoding; Method B: Top-P sampling with p = 0.9; Method C: Top-P sampling with p = 0.5 and high temperature T = 1.5	4
142	6406531422518	Number of transformer layers; Learning rate schedule; Vocabulary size; Type of normalization (e.g., LayerNorm)	4
143	6406531422519	Top-k keeps only the k most probable tokens before sampling.; Top-p dynamically chooses the smallest set of tokens whose cumulative probability exceeds p.	5
144	6406531422520	Lowercasing text (for uncased models); Splitting into subword units using BPE or WordPiece; Adding position indices for each token	5
145	6406531422521	0.38 to 0.44	5, 6
146	6406531422523	0.98 to 1.06	6, 7
147	6406531422524	0.04 to 0.10	7
148	6406531422531	$\pi = (0, 3, 1, 2, 4)$	8, 9
149	6406531422532	120	9
150	6406531422533	80	9
151	6406531422526	(-2, -1, 0, 1, 2)	10, 11
152	6406531422527	[-0.2, -0.2, -0.2, -0.2]	11
153	6406531422528	10	11, 12
154	6406531422529	21.3 to 22.0	12

Extraction notes

Status: verified
Count reconciliation: Section 6 declares 12 portal questions and 40 section marks. The 12 portal containers comprise: 1 administrative confirmation question (Q137, 0 marks, excluded), 8 standalone questions (Q138: 2 marks, Q139: 2 marks, Q140: 3 marks, Q141: 3 marks, Q142: 3 marks, Q143: 3 marks, Q144: 3 marks, Q145: 3 marks; total standalone = 22 marks), and 3 comprehension portal containers (Q146-147: 5 marks; Q148-150: 5 marks; Q151-154: 8 marks; comprehension subquestions total 18 marks). Total academic questions = 8 standalone + 9 comprehension subquestions = 17 academic questions. Total marks = 22 + 18 = 40 marks, exactly reconciling the declared section marks and question containers.

Excluded source entries:

{"classification": "subject-confirmation", "page": 1, "bbox": [22.87, 689.2, 542.5, 841.89], "reason": "Administrative subject confirmation asking candidate to confirm subject and cross check hall ticket", "identity": "137"}