Concept

Input Representation for a Single Token in Autoregressive Generation

During autoregressive generation, for the token at the current position i', its embedding is computed. This embedding, which typically combines the token's semantic meaning and its positional information, serves as the initial input representation that is fed into the stack of Transformer layers for processing.

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