Short Answer

Dynamic K/V Cache in Transformer Decoding

Imagine a Transformer-based language model is generating a response. Compare the composition of the key and value vector set used to generate the first token of the response with the set used to generate the tenth token of the response. Explain why this composition changes.

0

1

Updated 2025-10-07

Contributors are:

Who are from:

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