DiFiT method proposed for distribution-fidelity-guided visual token pruning in multimodal LLMs
A new paper introduces DiFiT, a visual token pruning approach for multimodal large language models guided by distribution fidelity, noting pruning's effects go beyond answer accuracy.
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-11-13 05:00 · arXiv cs.CV
DiFiT: Distribution-Fidelity-guided Visual Token Pruning for Multimodal Large Language Models