List your company
ResearchFormerly WizWand

Multimodal Question Answering on MMBench (dev)

77.3AccuracyQwen
55.35661.05366.7572.447Aug 1, 2025
Updated 6mo ago

Evaluation results

MethodLinks
QwenBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=100%2025.0877.3100
HiPrune++Backbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.087698.7
HiPruneBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.0875.998.7
VisionZipBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.0874.997.7
FastVBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=33.3%2025.0874.497.4
HiPrune++Backbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.0874.297.1
HiPruneBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.0873.797.1
VisionZipBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.0873.596.2
FastVBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=22.2%2025.0872.495.9
HiPrune++Backbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.0870.593
HiPruneBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.0869.693
VisionZipBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.0867.891.5
FastVBackbone=Qwen2.5-VL-3B-Instruct, Token Retain Ratio=11.1%2025.0856.286.4