Tokenization and vocabulary in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Tokenization Types

Which of the following best describes subword tokenization in natural language processing?

A Splitting text into individual characters only

B Breaking text into whole words without splitting

C Dividing text into smaller units than words, like syllables or parts of words

D Ignoring spaces and treating the entire text as one token

💡 Hint

Think about how tokenization helps handle unknown or rare words by breaking them down.

Subword tokenization breaks words into smaller meaningful parts, allowing models to understand rare or new words by their components.

❓ Predict Output

intermediate

Output of Tokenization Code

What is the output of the following Python code using a simple whitespace tokenizer?

Prompt Engineering / GenAI

text = "Machine learning is fun"
tokens = text.split()
print(tokens)

A ['Machine_learning_is_fun']

B ['Machine learning is fun']

C ['M', 'a', 'c', 'h', 'i', 'n', 'e']

D ['Machine', 'learning', 'is', 'fun']

💡 Hint

Remember what the split() method does by default.

The split() method without arguments splits the string at spaces, producing a list of words.

❓ Model Choice

advanced

Choosing Vocabulary Size for Tokenization

You want to train a language model on a large dataset with many rare words. Which vocabulary size is best to balance coverage and model size?

A Moderate vocabulary size (e.g., 30,000 tokens) with subword tokenization

B Very small vocabulary (e.g., 500 tokens) to reduce model size

C Very large vocabulary (e.g., 100,000 tokens) to cover all words exactly

D Vocabulary size does not affect model performance

💡 Hint

Think about how subword tokenization helps with rare words and model efficiency.

A moderate vocabulary with subword tokens balances coverage of rare words and keeps the model size manageable.

❓ Metrics

advanced

Evaluating Tokenizer Coverage

You have a tokenizer vocabulary of 10,000 tokens. After tokenizing a test set of 1,000 words, 50 words are split into multiple tokens. What is the approximate tokenization coverage percentage?

A 95%

B 5%

C 50%

D 100%

💡 Hint

Coverage means how many words are represented as single tokens.

If 50 words are split, then 950 words are covered as single tokens. Coverage = (950/1000)*100 = 95%.

🔧 Debug

expert

Identifying Tokenization Bug in Code

What error does the following code raise when trying to tokenize text using a vocabulary dictionary?

vocab = {"hello": 1, "world": 2}
text = "hello unknown world"
tokens = [vocab[word] for word in text.split()]
print(tokens)

A No error, output is [1, 0, 2]

B KeyError because 'unknown' is not in vocab

C TypeError because vocab is not iterable

D SyntaxError due to missing colon

💡 Hint

Check what happens when a word is not found in the dictionary keys.

Accessing a dictionary with a key that does not exist raises a KeyError in Python.