Skip to content
ml-recall

#decoding

11 questions from Search & Planning, Probabilistic & Graphical Models, Language Modeling, LLM Fundamentals & Prompting

Timed test

11 questions · 11 min · score and explanations at the end

Take a test
Q 1 / 11Score 0 / 0

Question 1: Short answer

mediumAI-checkedSearch & Planning

Beam search keeps the kk most probable prefixes at each step. Ignore <eos> and the length penalty, so every output has exactly T′=3T' = 3 tokens from a vocabulary of ∣Y∣=5|\mathcal{Y}| = 5. What is the smallest beam size kk that guarantees beam search returns the most probable sequence, i.e. makes it an exhaustive search?

a whole number