Why QA systems extract answers in NLP - Test Your Understanding

Practice - 5 Tasks

1fill in blank

easy

Complete the code to extract the answer from the model's output.

NLP

answer = model.predict(question, context).[1]()

Drag options to blanks, or click blank then click option'

Adecode

Bget_answer

Cstrip

Dextract

Attempts:

3 left

💡 Hint

Common Mistakes

Using methods that do not convert tokens to text, like 'extract' or 'get_answer'.

The decode() method converts the model's output tokens into a readable string answer.

2fill in blank

medium

Complete the code to tokenize the input question for the QA system.

NLP

inputs = tokenizer.[1](question, return_tensors='pt')

Drag options to blanks, or click blank then click option'

Atransform

Btokenize

Cparse

Dencode

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'tokenize' which returns tokens but not tensor format.

The encode() method converts the question text into tokens the model can understand.

3fill in blank

hard

Fix the error in the code to get the start position of the answer.

NLP

start_pos = outputs.start_logits.[1](dim=1).argmax()

Drag options to blanks, or click blank then click option'

Asum

Bmax

Csoftmax

Dmean

Attempts:

3 left

💡 Hint

Common Mistakes

Using 'max' directly on logits without softmax.

The softmax() function converts logits to probabilities before finding the max index.

4fill in blank

hard

Fill both blanks to extract the answer text from tokens.

NLP

answer_tokens = inputs.input_ids[0][[1]:[2]]
answer = tokenizer.decode(answer_tokens)

Drag options to blanks, or click blank then click option'

Astart_pos

Bend_pos

C0

Dlen(inputs.input_ids[0])

Attempts:

3 left

💡 Hint

Common Mistakes

Using fixed indices like 0 or full length instead of predicted positions.

We slice tokens from start_pos to end_pos to get the answer span.

5fill in blank

hard

Fill all three blanks to prepare inputs and get the answer from the QA model.

NLP

inputs = tokenizer.[1](question, context, return_tensors='pt')
outputs = model(**inputs)
start_pos = outputs.start_logits.[2](dim=1).argmax()
end_pos = outputs.end_logits.[3](dim=1).argmax()

Drag options to blanks, or click blank then click option'

Aencode

Bsoftmax

Dtokenize

Attempts:

3 left

💡 Hint

Common Mistakes

Skipping softmax or using tokenize instead of encode.

We encode inputs, then apply softmax to logits before argmax to find answer positions.