Model serving for NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

What is the primary purpose of model serving in NLP?

Model serving is a key step after training an NLP model. What is its main goal?

A To visualize the training loss and accuracy curves

B To clean and preprocess the input text before training

C To deploy the trained model so it can make predictions on new text data in real time or batch

D To train the model on a larger dataset for better accuracy

💡 Hint

Think about what happens after a model is ready and you want to use it.

Model serving means making the trained NLP model available to accept new text inputs and return predictions, often in real time.

❓ Predict Output

intermediate

What output does this Flask model serving code produce?

Given this simple Flask app serving an NLP sentiment model, what is the JSON response when sending POST with text 'I love this!'?

NLP

from flask import Flask, request, jsonify
app = Flask(__name__)

@app.route('/predict', methods=['POST'])
def predict():
    data = request.json
    text = data.get('text', '')
    # Dummy sentiment prediction
    sentiment = 'positive' if 'love' in text else 'negative'
    return jsonify({'sentiment': sentiment})

# Assume app.run() is called elsewhere

A {"error": "No text provided"}

B {"sentiment": "negative"}

C 500 Internal Server Error

D {"sentiment": "positive"}

💡 Hint

Check if the word 'love' is in the input text.

The code returns 'positive' sentiment if the word 'love' is found in the input text, otherwise 'negative'.

❓ Hyperparameter

advanced

Which batch size is best for serving an NLP model with low latency?

You want to serve an NLP model that responds quickly to single user requests. Which batch size should you choose?

A Batch size of 1 to minimize waiting time per request

B Batch size of 64 to maximize GPU throughput

C Batch size of 128 to reduce memory usage

D Batch size of 256 to increase model accuracy

💡 Hint

Think about how batch size affects response time for individual requests.

For low latency serving, batch size 1 is best because it processes each request immediately without waiting to fill a batch.

❓ Metrics

advanced

Which metric best measures NLP model serving quality in production?

To monitor an NLP model serving system, which metric directly reflects user experience quality?

A Loss value during training

B Average response latency (time to get prediction)

C Number of model parameters

D Training accuracy on the original dataset

💡 Hint

Think about what users notice when using a model service.

Users care about how fast the model responds, so average response latency is a key serving metric.

🔧 Debug

expert

Why does this NLP model serving code raise a KeyError?

Examine this snippet from a model serving function. Why does it raise a KeyError?

NLP

def serve_model(request_json):
    text = request_json['input_text']
    # Model prediction code here
    return {'prediction': 'positive'}

# Called with: serve_model({'text': 'Hello world'})

A The key 'input_text' does not exist in the input dictionary, causing KeyError

B The function is missing a return statement

C The model prediction code is incomplete and causes runtime error

D The input dictionary is empty, causing KeyError

💡 Hint

Check the keys used to access the input dictionary.

The function expects 'input_text' key, but the input dictionary has 'text' key, so accessing 'input_text' raises KeyError.