Sentence transformers in Prompt Engineering / GenAI - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the main purpose of sentence transformers?

Sentence transformers are used to convert sentences into vectors. What is the main reason for doing this?

A To represent sentences as fixed-size numerical vectors for similarity comparison

B To translate sentences from one language to another

C To generate new sentences from a given input

D To count the number of words in a sentence

💡 Hint

Think about how computers understand text for tasks like search or clustering.

Sentence transformers convert sentences into fixed-size vectors so that similar sentences have vectors close to each other. This helps in tasks like semantic search and clustering.

❓ Predict Output

intermediate

Output of embedding similarity calculation

Given two sentences embedded using a sentence transformer, what is the cosine similarity output?

Prompt Engineering / GenAI

from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity

model = SentenceTransformer('all-MiniLM-L6-v2')
sentences = ['I love machine learning.', 'Machine learning is great.']
embeddings = model.encode(sentences)
similarity = cosine_similarity([embeddings[0]], [embeddings[1]])[0][0]
print(round(similarity, 2))

A 0.15

B 0.85

C 1.00

D 0.50

💡 Hint

These sentences have similar meaning, so expect a high similarity but less than 1.

The cosine similarity between embeddings of similar sentences is close to 1 but not exactly 1. Here, it is about 0.85.

❓ Model Choice

advanced

Choosing the best sentence transformer model for large-scale semantic search

You want to build a semantic search engine that balances speed and accuracy on millions of sentences. Which sentence transformer model is best?

A 'all-MiniLM-L6-v2' for fast and reasonably accurate embeddings

B 'bert-base-uncased' for the most accurate embeddings but slower speed

C 'distilbert-base-uncased' for the smallest model but less accuracy

D 'gpt-3' for generating embeddings with zero-shot learning

💡 Hint

Consider the trade-off between speed and accuracy for large datasets.

'all-MiniLM-L6-v2' is optimized for fast embedding generation with good accuracy, making it suitable for large-scale semantic search.

❓ Hyperparameter

advanced

Effect of changing pooling strategy in sentence transformers

Sentence transformers use pooling to create sentence embeddings from token embeddings. What happens if you change from mean pooling to max pooling?

A Max pooling always improves embedding quality regardless of task

B Max pooling averages all token embeddings, smoothing the sentence representation

C Max pooling reduces embedding size by half compared to mean pooling

D Max pooling captures the most prominent features, possibly making embeddings more sensitive to key words

💡 Hint

Think about how max pooling selects values compared to averaging.

Max pooling picks the highest value per dimension, highlighting strong features, which can make embeddings more sensitive to important words.

🔧 Debug

expert

Why does this sentence transformer embedding code raise an error?

Consider this code snippet:

from sentence_transformers import SentenceTransformer
model = SentenceTransformer('all-MiniLM-L6-v2')
embeddings = model.encode('This is a test sentence')
print(embeddings.shape)

What error will this code raise and why?

A AttributeError because embeddings is a 1D array and has no attribute 'shape'

B ValueError because the model name is invalid

C No error; the code runs and prints the shape

D TypeError because encode expects a list of sentences, not a single string

💡 Hint

Check the documentation for the encode method input types and output.

The encode method accepts a single string or list of strings and returns a numpy array with shape. So no error occurs and shape prints correctly.