Visual-token attention filtering improves structured pruning of vision-language models, keeping 94% of average benchmark accuracy after removing 20% of parameters.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SlimVLM: Sensitivity-aware Dynamic Structured Pruning with Adaptive Visual Token Selection for Efficient Vision-Language Models
Visual-token attention filtering improves structured pruning of vision-language models, keeping 94% of average benchmark accuracy after removing 20% of parameters.