List your company

Document Visual Question Answering on DocVQA (Accuracy)

95.75AccuracyBase
29.24246.508563.77581.0415Sep 25, 2025Nov 10, 2025Dec 27, 2025Feb 12, 2026Mar 31, 2026May 17, 2026Jul 3, 2026

Evaluation results

MethodLinks
BasePruning Ratio=100%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0595.75
ERASEPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0595.44
IVC-PrunePruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0595.31
ERASEPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0595
IVC-PrunePruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0594.97
DeFactoBackbone=Qwen2.5-VL-7B2025.0994
C-RADIOv4-HTokens per frame/tile=256, TTFT (ms)=468.2, LLM Backbone=9B Nemotron2026.0793.3
ERASEPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0593.21
RADIO1DTokens per frame/tile=192, TTFT (ms)=380.4, LLM Backbone=9B Nemotron2026.0793.2
RADIO1DTokens per frame/tile=256, TTFT (ms)=452.8, LLM Backbone=9B Nemotron2026.0793.2
RADIO1DTokens per frame/tile=224, TTFT (ms)=410.2, LLM Backbone=9B Nemotron2026.0793
SigLIP2-SO400mTokens per frame/tile=256, TTFT (ms)=440.5, LLM Backbone=9B Nemotron2026.0792.9
SigLIP2-gTokens per frame/tile=256, TTFT (ms)=517.6, LLM Backbone=9B Nemotron2026.0792.8
RADIO1DTokens per frame/tile=128, TTFT (ms)=373.2, LLM Backbone=9B Nemotron2026.0792.5
Qwen2.5-VLBackbone=Qwen2.5-VL-7B2025.0992
IVC-PrunePruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0591.02
RADIO1DTokens per frame/tile=64, TTFT (ms)=346.1, LLM Backbone=9B Nemotron2026.0790.6
PruneSIDPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0590.51
RADIO1DTokens per frame/tile=32, TTFT (ms)=335.1, LLM Backbone=9B Nemotron2026.0788.1
PruneSIDPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0585.77
Visual-SR1Backbone=Qwen2.5-VL-7B2025.0984.9
iLLaVAPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0584.89
VicropModel=Qwen2.5-VL 7B2025.0981.8
HiDeModel=Qwen2.5-VL 7B2025.0981.8
Qwen2.5-VL 7BModel=Qwen2.5-VL 7B2025.0981.7
iLLaVAPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0579.63
RADIO1DTokens per frame/tile=8, TTFT (ms)=329.7, LLM Backbone=9B Nemotron2026.0779.5
DARTPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0574.9
PruneSIDPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0568.86
iLLaVAPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0567.64
DeepEyesBackbone=Qwen2.5-VL-7B2025.0967.1
DARTPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0564.99
CDPrunerPruning Ratio=-65.11%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0563.06
GRITBackbone=Qwen2.5-VL-3B2025.0957
CDPrunerPruning Ratio=-72.77%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0554.53
VRCDdecoding length (L)=384, forward ratio (FR)=0.52026.0554.33
IGdecoding length (L)=384, forward ratio (FR)=0.52026.0554.05
Confidencedecoding length (L)=384, forward ratio (FR)=0.52026.0553.74
Margindecoding length (L)=384, forward ratio (FR)=0.52026.0553.74
Entropydecoding length (L)=384, forward ratio (FR)=0.52026.0552.93
VRCDdecoding length (L)=384, forward ratio (FR)=0.252026.0552.24
RADIO1DTokens per frame/tile=1, TTFT (ms)=327.0, LLM Backbone=9B Nemotron2026.0751.1
Margindecoding length (L)=384, forward ratio (FR)=0.252026.0550.57
IGdecoding length (L)=384, forward ratio (FR)=0.252026.0550.51
Confidencedecoding length (L)=384, forward ratio (FR)=0.252026.0550.41
Entropydecoding length (L)=384, forward ratio (FR)=0.252026.0549.93
VRCDdecoding length (L)=384, forward ratio (FR)=0.1252026.0549.22
Margindecoding length (L)=384, forward ratio (FR)=0.1252026.0546.54
IGdecoding length (L)=384, forward ratio (FR)=0.1252026.0546.48
Confidencedecoding length (L)=384, forward ratio (FR)=0.1252026.0546.45
Entropydecoding length (L)=384, forward ratio (FR)=0.1252026.0546.39
DARTPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0544.94
CDPrunerPruning Ratio=-84.43%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.0538.49
ViCropBackbone=LLaVA-1.5 (Vicuna-7B)2025.0931.8