Concept

Span Masking

Span masking is an input corruption technique in which non-overlapping spans of tokens are randomly sampled from a sequence, and each span is replaced by a single [MASK] token. This approach uniquely accommodates spans of length 0, where a [MASK] token is simply inserted at a chosen position in the sequence without removing any original tokens.

0

1

Updated 2026-04-17

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