List your company
ResearchFormerly WizWand

Image Classification on ImageNet 1K (val) (Speedup)

2.13SpeedupThreshold-ℓ2
0.50760.92881.351.7712Apr 16, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Threshold-ℓ2Backend=Triton (ours), Token Pruning Ratio=90%2026.042.13
DynamicViTBackend=Triton (ours), Token Pruning Ratio=90%2026.042.06
EViTBackend=Triton (ours), Token Pruning Ratio=90%2026.041.94
ATSBackend=Triton (ours), Token Pruning Ratio=90%2026.041.65
Threshold-ℓ2Backend=Triton (ours), Token Pruning Ratio=50%2026.041.37
DynamicViTBackend=Triton (ours), Token Pruning Ratio=50%2026.041.33
EViTBackend=Triton (ours), Token Pruning Ratio=50%2026.041.28
ATSBackend=Triton (ours), Token Pruning Ratio=50%2026.041.14
Threshold-ℓ2Backend=PyTorch (padded), Token Pruning Ratio=50%2026.040.62
Threshold-ℓ2Backend=PyTorch (padded), Token Pruning Ratio=90%2026.040.62
DynamicViTBackend=PyTorch (padded), Token Pruning Ratio=50%2026.040.61
DynamicViTBackend=PyTorch (padded), Token Pruning Ratio=90%2026.040.61
EViTBackend=PyTorch (padded), Token Pruning Ratio=50%2026.040.6
EViTBackend=PyTorch (padded), Token Pruning Ratio=90%2026.040.6
ATSBackend=PyTorch (padded), Token Pruning Ratio=50%2026.040.57
ATSBackend=PyTorch (padded), Token Pruning Ratio=90%2026.040.57