Efficiency of Sequence Expansion
Consider a step-wise sequence generation process where two candidate sequences are:
The movie was great ⟨EOS⟩The movie was great and
In the next step, one of these sequences will be removed from the set of candidates to be expanded, while the other will be used to generate longer sequences. Identify which sequence is removed and analyze the primary benefit of this removal for the overall search process.
0
1
Tags
Ch.5 Inference - Foundations of Large Language Models
Foundations of Large Language Models
Foundations of Large Language Models Course
Computing Sciences
Analysis in Bloom's Taxonomy
Cognitive Psychology
Psychology
Social Science
Empirical Science
Science
Related
Notation for the Set of Complete Sequences
In a step-by-step sequence generation process, a set of candidate sequences is maintained and expanded at each step. Suppose at a given step, the current set of candidate sequences is:
["The cat sat", "The dog ran ⟨EOS⟩", "The cat slept on"]Assuming
⟨EOS⟩is a special token indicating the end of a sequence, which of these sequences will be used as a basis for generating longer sequences in the next step?Applying a Stopping Condition in Sequence Expansion
Efficiency of Sequence Expansion