Skip to content
ml-recall

#vision-transformer

5 questions from Attention & Transformers, Landmark Architectures

Timed test

5 questions · 5 min · score and explanations at the end

Take a test
Q 1 / 5Score 0 / 0

Question 1: Short answer

mediumAI-checkedAttention & Transformers

A vision Transformer splits square images into 16×1616 \times 16 patches. You raise img_size from 96 to 192 pixels and keep everything else fixed. Ignoring the <cls> token, by what factor does the cost of computing the self-attention score matrices QK⊤\mathbf{Q}\mathbf{K}^\top in each layer grow?

a whole number