Decoder-only MLLMs tolerate simplified self-attention and FFN processing for visual tokens in about half of their layers, enabling a training-free FLOPs reduction method.
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
-
RedundancyLens: Revealing and Exploiting Visual Token Processing Redundancy for Efficient Decoder-Only MLLMs
Decoder-only MLLMs tolerate simplified self-attention and FFN processing for visual tokens in about half of their layers, enabling a training-free FLOPs reduction method.