List your company
ResearchFormerly WizWand

Visual Question Answering on OKVQA (Accuracy)

64.56AccuracyVanilla
21.649632.789843.9355.0702Sep 25, 2025Nov 6, 2025Dec 19, 2025Jan 31, 2026Mar 15, 2026Apr 27, 2026Jun 9, 2026
Updated 4mo ago

Evaluation results

MethodLinks
VanillaBackbone=Qwen2.5-VL-7B, Token budget=100%2026.0464.56
FastVBackbone=Qwen2.5-VL-7B, Token budget=25%2026.0463.89
DARTBackbone=Qwen2.5-VL-7B, Token budget=25%2026.0463.83
DivPruneBackbone=Qwen2.5-VL-7B, Token budget=25%2026.0463.46
InternVL-2.0Params # trainable=8B, Text Backbone=AR, Image Backbone=-2026.0562.9
SLVR-7BReasoning Paradigm=Visual + Semantic Latent2026.0561.8
DASHBackbone=Qwen2.5-VL-7B, Token budget=25%2026.0461.73
DeFactoBackbone=Qwen2.5-VL-7B2025.0961.7
Qwen2.5-VL-7BReasoning Paradigm=Text-only2026.0558.9
Qwen2.5-VLBackbone=Qwen2.5-VL-7B2025.0958.9
LoRAModel=Qwen2.5-VL-3B, Optimization=FlatPO2026.0653.8
LoRAModel=Qwen2.5-VL-3B, Optimization=SAM2026.0652.2
LoRAModel=Qwen2.5-VL-3B, Optimization=GAM2026.0652
Prefix TuningModel=Qwen2.5-VL-3B, Optimization=FlatPO2026.0652
Prefix TuningModel=Qwen2.5-VL-3B, Optimization=SAM2026.0651.6
LoRAModel=Qwen2.5-VL-3B, Optimization=Base2026.0651.1
Prefix TuningModel=Qwen2.5-VL-3B, Optimization=GAM2026.0650.9
LVRReasoning Paradigm=Visual Latent2026.0550.6
Prefix TuningModel=Qwen2.5-VL-3B, Optimization=Base2026.0650.4
Visual-SR1Reasoning Paradigm=Text CoT2026.0549
Visual-SR1Backbone=Qwen2.5-VL-7B2025.0949
FlamingoParams # trainable=9B, Text Backbone=AR, Image Backbone=-2026.0544.7
GRITReasoning Paradigm=Text CoT2026.0543.1
GRITBackbone=Qwen2.5-VL-3B2025.0943.1
DeepEyesReasoning Paradigm=Text + Visual Cropping2026.0540
DeepEyesBackbone=Qwen2.5-VL-7B2025.0940
DualDiff (256x256)Params # trainable=2B, Text Backbone=FM, Image Backbone=FM2026.0528.5
DualDiffTrainable Params=2B, Training resolution=2562026.0528.5
DualDiff (512x512)Params # trainable=2B, Text Backbone=FM, Image Backbone=FM2026.0525.3
DualDiffTrainable Params=2B, Training resolution=5122026.0525.3
CM3LeonParams # trainable=7B, Text Backbone=AR, Image Backbone=AR2026.0523.8
CM3LeonTrainable Params=7B2026.0523.8
FullFlowParams # trainable=130M, Text Backbone=FM, Image Backbone=FM2026.0523.3
FullFlowTrainable Params=130M2026.0523.3