NCP-GENL LLM Architecture Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  To filter out stop words from the input text.

  Why it's wrong here

  Masking in Transformers is about temporal causality, not linguistic content filtering. Stop words are handled during the preprocessing or tokenization stages. The attention mask is a structural mechanism used to enforce the autoregressive nature of the generation process, which is unrelated to the semantics of the words being processed.

• ✓

  To prevent the model from attending to future tokens in the sequence.

  Why this is correct

  The causal mask is a triangular matrix applied to the attention scores that sets values for future tokens to negative infinity. This ensures that when the softmax is applied, the weights for these positions become zero, effectively hiding the future context from the model during training and ensuring autoregressive integrity.

• ✗

  To increase the randomness of the model's predictions.

  Why it's wrong here

  Randomness in model predictions is controlled by parameters like temperature, top-k, or top-p sampling during inference, not by the attention mask. The attention mask is a deterministic architectural constraint designed to enforce sequence ordering; it does not introduce or manage stochasticity in the model's output generation process.

• ✗

  To reduce the computation of the attention mechanism by 50%.

  Why it's wrong here

  While masking does effectively compute fewer values, it is not an efficiency-focused optimization; it is a correctness-focused architectural requirement. The computation reduction is a side effect of ignoring future positions, but the primary goal is ensuring that the model does not violate the causal dependencies of language generation.