#autoregressive-generation

2 questions from Attention & Transformers, Language Modeling

Timed test

2 questions · 2 min · score and explanations at the end

Q 1 / 2Score 0 / 0

Shuffle questions

Question 1: Single choice

mediumAI-checkedAttention & Transformers

A decoder-only transformer generates text one token at a time and uses a KV cache. What does the cache store?

Keys and values of earlier tokens, for every attention layer1
The queries of all earlier tokens, so attention need not recompute them2
The attention weight matrices of the earlier steps3
The output logits of the earlier steps4

Keys: 1–9 select · Enter check / next