Formula

Formula for Input Embedding Composition

The final input embedding for a sequence token, denoted as e\mathbf{e}, is calculated by summing three distinct component vectors: the token embedding x\mathbf{x}, the positional embedding epos\mathbf{e}_{\text{pos}}, and the segment embedding eseg\mathbf{e}_{\text{seg}}. This composition is represented by the formula: e=x+epos+eseg\mathbf{e} = \mathbf{x} + \mathbf{e}_{\text{pos}} + \mathbf{e}_{\text{seg}}

0

1

Updated 2026-06-27

Contributors are:

Who are from:

Tags

Ch.1 Pre-training - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences

Related