Tokenization and vocabulary in Prompt Engineering / GenAI - Model Pipeline Trace

Model Pipeline - Tokenization and vocabulary

This pipeline shows how raw text is changed into tokens and then mapped to a vocabulary for a language model to understand and use.

Data Flow - 3 Stages

1Raw Text Input

1 sentence (string)→Input sentence to be processed→1 sentence (string)

"Hello, how are you?"

↓

2Tokenization

1 sentence (string)→Split sentence into smaller pieces called tokens→List of tokens (e.g., 6 tokens)

["Hello", ",", "how", "are", "you", "?"]

↓

3Vocabulary Mapping

List of tokens (6 tokens)→Convert tokens to numbers using a vocabulary dictionary→List of token IDs (6 integers)

[1543, 12, 78, 45, 89, 7]

Training Trace - Epoch by Epoch

Loss
2.3 |*****
1.85|****
1.4 |***
1.1 |**
0.85|*

Epoch	Loss ↓	Accuracy ↑	Observation
1	2.30	0.15	Model starts with high loss and low accuracy as it learns token patterns.
2	1.85	0.35	Loss decreases and accuracy improves as vocabulary mapping becomes clearer.
3	1.40	0.55	Model better understands token sequences, improving predictions.
4	1.10	0.70	Vocabulary usage is more accurate, loss continues to drop.
5	0.85	0.80	Model converges well on token patterns and vocabulary.

Prediction Trace - 3 Layers

Layer 1: Input Sentence

Layer 2: Tokenization

Layer 3: Vocabulary Mapping

Model Quiz - 3 Questions

What does tokenization do in this pipeline?

A Trains the model to predict next words

B Converts tokens into numbers

C Splits text into smaller pieces called tokens

D Calculates accuracy of the model

Key Insight

Tokenization breaks text into manageable pieces, and vocabulary mapping turns these pieces into numbers. This process helps the model learn language patterns effectively, shown by decreasing loss and increasing accuracy during training.

Practice

1. What does tokenization do in natural language processing?

easy

A. Converts tokens into images

B. Breaks text into smaller pieces called tokens

C. Removes all punctuation from text

D. Combines multiple texts into one

Solution

1. Step 1: Understand the role of tokenization

   Tokenization splits text into smaller parts called tokens, like words or subwords.

2. Step 2: Compare options with tokenization definition

   Only Breaks text into smaller pieces called tokens correctly describes breaking text into tokens.

3. Final Answer:

   Breaks text into smaller pieces called tokens -> Option B

4. Quick Check:

   Tokenization = splitting text [OK]

Hint: Tokenization means splitting text into pieces [OK]

Common Mistakes:

• Thinking tokenization changes text to images
• Confusing tokenization with removing punctuation
• Believing tokenization merges texts

2. Which of the following is the correct way to represent a token ID in Python?

easy

A. token_id = 'word'

B. token_id = {word: 1}

C. token_id = [word]

D. token_id = 123

Solution

1. Step 1: Understand token ID representation

   Token IDs are numbers representing tokens, so they should be integers.

2. Step 2: Check each option's type

   token_id = 123 assigns an integer 123, which is correct. Others use strings, lists, or dictionaries incorrectly.

3. Final Answer:

   token_id = 123 -> Option D

4. Quick Check:

   Token ID = number [OK]

Hint: Token IDs are numbers, not words or lists [OK]

Common Mistakes:

• Using strings instead of numbers for token IDs
• Confusing token IDs with token text
• Using lists or dictionaries wrongly

3. Given the vocabulary {'hello': 1, 'world': 2, '!': 3}, what is the token ID list for the text 'hello world!'?

medium

A. [1, 2, 3]

B. [0, 1, 2]

C. ['hello', 'world', '!']

D. [3, 2, 1]

Solution

1. Step 1: Map each word to its token ID

   'hello' maps to 1, 'world' maps to 2, and '!' maps to 3 according to the vocabulary.

2. Step 2: Create the token ID list in order

   The text 'hello world!' becomes [1, 2, 3].

3. Final Answer:

   [1, 2, 3] -> Option A

4. Quick Check:

   Text tokens = [1, 2, 3] [OK]

Hint: Match words to IDs in order [OK]

Common Mistakes:

• Mixing up token order
• Using token text instead of IDs
• Assigning wrong IDs from vocabulary

4. What is wrong with this tokenization code snippet?

vocab = {'hi': 1, 'there': 2}
text = 'hi there'
tokens = [vocab[word] for word in text.split() if word in vocab]

medium

A. It will raise a KeyError if a word is missing

B. It correctly tokenizes the text

C. It ignores words not in vocabulary

D. It uses split() incorrectly on the text

Solution

1. Step 1: Analyze the list comprehension

   The code splits text and includes only words found in vocab, skipping others.

2. Step 2: Identify behavior on unknown words

   Words not in vocab are ignored, which may lose information.

3. Final Answer:

   It ignores words not in vocabulary -> Option C

4. Quick Check:

   Unknown words skipped = ignoring [OK]

Hint: Check if unknown words are skipped or cause errors [OK]

Common Mistakes:

• Assuming KeyError will happen due to 'if' check
• Thinking split() is wrong here
• Missing that unknown words are ignored silently

5. You have a vocabulary with tokens: {'I':1, 'love':2, 'AI':3, '.':4}. How would you tokenize the sentence 'I love AI!' considering the exclamation mark is not in the vocabulary?

hard

A. Add '!' to vocabulary with new ID and tokenize as [1, 2, 3, 5]

B. Replace '!' with '.' and tokenize as [1, 2, 3, 4]

C. Ignore '!' and tokenize as [1, 2, 3]

D. Raise an error because '!' is unknown

Solution

1. Step 1: Understand vocabulary coverage

   The vocabulary lacks '!', so it must be added to handle the sentence fully.

2. Step 2: Add '!' with a new token ID

   Assign '!' a new ID (e.g., 5) and tokenize the sentence as [1, 2, 3, 5].

3. Final Answer:

   Add '!' to vocabulary with new ID and tokenize as [1, 2, 3, 5] -> Option A

4. Quick Check:

   Unknown token added = new ID [OK]

Hint: Add unknown tokens to vocabulary before tokenizing [OK]

Common Mistakes:

• Ignoring unknown tokens silently
• Replacing unknown tokens incorrectly
• Assuming error without handling unknown tokens