New benchmarks and pruning methods target multimodal model reasoning and efficiency
Two new arXiv papers address multimodal large language models: VIGOR benchmarks whether visual rationales are correct beyond answer accuracy, and DiFiT introduces distribution-fidelity-guided visual token pruning to cut inference costs.
Read the full story at arXiv cs.CV ↗
Timeline · 2 reports
- 2026-11-13 05:00 · arXiv cs.CV
DiFiT: Distribution-Fidelity-guided Visual Token Pruning for Multimodal Large Language Models - 2026-11-11 05:00 · arXiv cs.CV
VIGOR: Benchmarking Visual Rationale Correctness in Multimodal Large Language Models