Visual prefixes account for a major portion of the per-layer computation in multimodal large language models (MLLMs), making visual-token pruning a direct approach to accelerating inference. Existing top-K methods typically evaluate tokens independently and apply a uniform budget to all inputs, overlooking both selecti...
Han-Sen Zhang, Lan He, Min Yao et al.· 0 citations
This work reformulate WSI token pruning as a sequential selection process, enabling the model to autonomously learn an optimal routing strategy rather than relying on static heuristics.
Jing-Zhi Chen, Lan He, Zehong Chen et al.· arXiv.org· 2 citations
A new evaluation framework that decouples MLLMs'capabilities to faithfully assess VTC quality is introduced, and the ZeroSense Benchmark is introduced to ensure low semantic correlation of testing samples.
Yonghan Gao, Zehong Chen, Lijian Xu et al.· 1 citation
DiffPrune is proposed, which gives token scores a direct meaning and implements an Information Throttler, which injects variance-preserving noise into visual tokens, where high-score tokens remain close to their original representations, while low-score tokens carry less original information.
Lan He, Min Yao, S. Young et al.· 1 citation· ⚡1
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
