List your company

Visual Question Answering on GQA (Accuracy and Average performance retention)

65.34AccuracyLLaVA-NeXT 8B
30.531239.568148.60557.6419Oct 6, 2025Nov 18, 2025Dec 31, 2025Feb 13, 2026Mar 28, 2026May 10, 2026Jun 23, 2026

Evaluation results

MethodLinks
LLaVA-NeXT 8BBackbone=LLaVA-NeXT 8B, Sparsity Level=0%, Pruning Method=Dense, Pruning Type=None2026.0365.34100
VanillaBackbone=LLaVA-1.5-13B, Retained Tokens=5762026.0663.3—
ATV-PruningBackbone=LLaVA-NeXT 8B, Sparsity Level=50%, Pruning Method=ATV-Pruning, Pruning Type=unstructured, uniform pruning2026.0363.2894
TAMPBackbone=LLaVA-NeXT 8B, Sparsity Level=50%, Pruning Method=TAMP, Pruning Type=unstructured, uniform pruning2026.0363.1592.67
FP32Model=LLaVA-1.6-vicuna-13B, Quantization=FP322025.1063.08—
SparseGPTBackbone=LLaVA-NeXT 8B, Sparsity Level=50%, Pruning Method=SparseGPT, Pruning Type=unstructured, uniform pruning2026.0362.8691.74
WandaBackbone=LLaVA-NeXT 8B, Sparsity Level=50%, Pruning Method=Wanda, Pruning Type=unstructured, uniform pruning2026.0362.6188.36
Qwen2-VL 7BBackbone=Qwen2-VL 7B, Sparsity Level=0%, Pruning Method=Dense, Pruning Type=None2026.0362.35100
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0662.3—
VanillaSource=CVPR’24, Average Token=576, Reroute=false2026.0661.9100
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0661.3—
Nüwa+RerouteSource=Ours, Average Token=192, Reroute=true2026.0661.1100.6
NüwaSource=ICLR’26, Average Token=192, Reroute=false2026.0660.999.2
Nüwa+RerouteSource=Ours, Average Token=128, Reroute=true2026.0660.499.8
NüwaSource=ICLR’26, Average Token=128, Reroute=false2026.0660.298.4
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0659.6—
VisionZipSource=CVPR’25, Average Token=192, Reroute=false2026.0659.397.9
FastVBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0659.1—
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0659.1—
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=1922026.0658.7—
PriorTRBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0658.7—
SparseGPTBackbone=LLaVA-NeXT 8B, Sparsity Level=60%, Pruning Method=SparseGPT, Pruning Type=unstructured, uniform pruning2026.0358.5276.24
Nüwa+RerouteSource=Ours, Average Token=64, Reroute=true2026.0658.596.7
NüwaSource=ICLR’26, Average Token=64, Reroute=false2026.0658.395.8
ATV-PruningBackbone=Qwen2-VL 7B, Sparsity Level=60%, Pruning Method=ATV-Pruning, Pruning Type=unstructured, uniform pruning2026.0358.1385.65
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0657.9—
PruMergeSource=ICCV’25, Average Token=128, Reroute=false2026.0657.894.7
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0657.8—
FastVBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0657.7—
ATV-PruningBackbone=LLaVA-NeXT 8B, Sparsity Level=60%, Pruning Method=ATV-Pruning, Pruning Type=unstructured, uniform pruning2026.0357.6177.01
SparseVLMSource=ICML’25, Average Token=192, Reroute=false2026.0657.695.4
VisionZipSource=CVPR’25, Average Token=128, Reroute=false2026.0657.696.9
TAMPBackbone=Qwen2-VL 7B, Sparsity Level=60%, Pruning Method=TAMP, Pruning Type=unstructured, uniform pruning2026.0357.4783.79
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=1282026.0657.2—
PDropSource=CVPR’25, Average Token=192, Reroute=false2026.0657.195.7
SparseGPTBackbone=Qwen2-VL 7B, Sparsity Level=60%, Pruning Method=SparseGPT, Pruning Type=unstructured, uniform pruning2026.0356.9885.3
FP32Model=LLaVA-1.6-mistral-7B, Quantization=FP322025.1056.59—
PDropSource=CVPR’25, Average Token=128, Reroute=false2026.065693.8
SparseVLMSource=ICML’25, Average Token=128, Reroute=false2026.065692.9
VisPrunerBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0655.9—
PruMergeSource=ICCV’25, Average Token=64, Reroute=false2026.0655.492.1
VisionZipSource=CVPR’25, Average Token=64, Reroute=false2026.0655.193.2
TAMPBackbone=LLaVA-NeXT 8B, Sparsity Level=60%, Pruning Method=TAMP, Pruning Type=unstructured, uniform pruning2026.0354.6273.75
WandaBackbone=Qwen2-VL 7B, Sparsity Level=60%, Pruning Method=Wanda, Pruning Type=unstructured, uniform pruning2026.0354.3777.78
SparseVLMSource=ICML’25, Average Token=64, Reroute=false2026.0653.891.5
FastVBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0653.7—
FastVSource=ECCV’24, Average Token=192, Reroute=false2026.0652.790.1
RegCacheModel=LLaVA-1.6-vicuna-13B, Quantization=4-bit2025.1051.4—
PDropBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0650.7—
SparseVLMBackbone=LLaVA-1.5-13B, Retained Tokens=642026.0650.6—
WandaBackbone=LLaVA-NeXT 8B, Sparsity Level=60%, Pruning Method=Wanda, Pruning Type=unstructured, uniform pruning2026.0349.8864.45
FastVSource=ECCV’24, Average Token=128, Reroute=false2026.0649.685.2
RepQ-ViTModel=LLaVA-1.6-vicuna-13B, Quantization=4-bit2025.1048.93—
RegCacheModel=LLaVA-1.6-mistral-7B, Quantization=4-bit2025.1046.21—
FastVSource=ECCV’24, Average Token=64, Reroute=false2026.0646.177.9
RepQ-ViTModel=LLaVA-1.6-mistral-7B, Quantization=4-bit2025.1044.63—
PDropSource=CVPR’25, Average Token=64, Reroute=false2026.0641.970.3
NaïveModel=LLaVA-1.6-vicuna-13B, Quantization=4-bit2025.1035.72—
NaïveModel=LLaVA-1.6-mistral-7B, Quantization=4-bit2025.1031.87—