Chipmunk speeds up diffusion transformer inference by recomputing, at each step, only the top percent of attention and MLP activation columns that change most between steps, caching the rest in column-sparse GPU kernels.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Chipmunk: Training-Free Acceleration of Diffusion Transformers with Dynamic Column-Sparse Deltas
Chipmunk speeds up diffusion transformer inference by recomputing, at each step, only the top percent of attention and MLP activation columns that change most between steps, caching the rest in column-sparse GPU kernels.