A two-layer transformer with relative positional encoding keeps its induction head active across the whole sequence, while absolute positional encoding loses it in the second half.
Birth of a transformer: A memory viewpoint.Advances in Neural Information Processing Systems, 36, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
contradiction 1
citation-polarity summary
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1roles
contradiction 1polarities
contest 1representative citing papers
citing papers explorer
-
Rethinking Associative Memory Mechanism in Induction Head
A two-layer transformer with relative positional encoding keeps its induction head active across the whole sequence, while absolute positional encoding loses it in the second half.