A learned router sends about 90% of tokens through a linear projection path instead of quadratic attention, keeps the MLP for every token, and matches dense Transformer accuracy at roughly 85% of the FLOPs.
Title resolution pending
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
-
DTRNet: Dynamic Token Routing Network to Reduce Quadratic Costs in Transformers
A learned router sends about 90% of tokens through a linear projection path instead of quadratic attention, keeps the MLP for every token, and matches dense Transformer accuracy at roughly 85% of the FLOPs.