StagFormer staggers transformer layers one time step apart with delayed cross-attention, enabling depth-parallel decoding at quality comparable to a deeper baseline.
Alternating updates for efficient transformers
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
StagFormer: Time Staggering Transformer Decoding for RunningLayers In Parallel
StagFormer staggers transformer layers one time step apart with delayed cross-attention, enabling depth-parallel decoding at quality comparable to a deeper baseline.