DeepCrossAttention learns input-dependent weighted combinations of previous layer outputs for queries, keys, and values, yielding better perplexity-per-training-time than standard residual transformers.
Implicit bias of large depth networks: a notion of rank for nonlinear functions
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
extension 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
extension 1polarities
extend 1representative citing papers
citing papers explorer
-
DeepCrossAttention: Supercharging Transformer Residual Connections
DeepCrossAttention learns input-dependent weighted combinations of previous layer outputs for queries, keys, and values, yielding better perplexity-per-training-time than standard residual transformers.