Efficiency Benchmarking on ImageNet-1K 1.0 (test)
17.8GPU Latency (ms)Isomorphic Pruning
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Isomorphic PruningArchitecture=DeiT-Tiny, Params (M)=3.08, Size (MB)=11.74, Peak Mem (MB)=792, GPU batch size=256, CPU batch size=82026.01 | 17.8 | 1.36 | 11.68 | 1.44 | |
| AgenticPrunerArchitecture=DeiT-Tiny, Params (M)=2.90, Size (MB)=11.07, Peak Mem (MB)=828, GPU batch size=256, CPU batch size=82026.01 | 21.52 | 1.13 | 9.19 | 1.82 | |
| DeiT-Tiny†Architecture=DeiT-Tiny, Params (M)=5.91, Size (MB)=22.55, Peak Mem (MB)=680, GPU batch size=256, CPU batch size=82026.01 | 24.29 | 1 | 16.77 | 1 | |
| AgenticPrunerArchitecture=ResNet-50, Params (M)=13.29, Size (MB)=50.84, Peak Mem (MB)=1926, GPU batch size=256, CPU batch size=82026.01 | 64.35 | 1.29 | 175.19 | 0.55 | |
| Isomorphic PruningArchitecture=ResNet-50, Params (M)=15.31, Size (MB)=58.55, Peak Mem (MB)=2344, GPU batch size=256, CPU batch size=82026.01 | 71.1 | 1.17 | 679.97 | 0.14 | |
| ResNet-50†Architecture=ResNet-50, Params (M)=25.56, Size (MB)=97.70, Peak Mem (MB)=2937, GPU batch size=256, CPU batch size=82026.01 | 82.97 | 1 | 96.89 | 1 | |
| AgenticPrunerArchitecture=ResNet-101, Params (M)=25.58, Size (MB)=97.89, Peak Mem (MB)=2541, GPU batch size=256, CPU batch size=82026.01 | 115.92 | 1.11 | 197.74 | 3.97 | |
| Isomorphic PruningArchitecture=ResNet-101, Params (M)=29.22, Size (MB)=111.78, Peak Mem (MB)=2521, GPU batch size=256, CPU batch size=82026.01 | 121.73 | 1.06 | 181.09 | 4.33 | |
| ResNet-101†Architecture=ResNet-101, Params (M)=44.55, Size (MB)=170.34, Peak Mem (MB)=3013, GPU batch size=256, CPU batch size=82026.01 | 129.18 | 1 | 784.2 | 1 | |
| AgenticPrunerArchitecture=ConvNeXt-S, Params (M)=48.56, Size (MB)=185.23, Peak Mem (MB)=3509, GPU batch size=256, CPU batch size=82026.01 | 372.6 | 1.41 | 337.82 | 1.07 | |
| Isomorphic PruningArchitecture=ConvNeXt-S, Params (M)=47.32, Size (MB)=180.51, Peak Mem (MB)=3573, GPU batch size=256, CPU batch size=82026.01 | 372.88 | 1.41 | 332.98 | 1.09 | |
| ConvNeXt-Base†Architecture=ConvNeXt-Base, Params (M)=88.59, Size (MB)=337.95, Peak Mem (MB)=5038, GPU batch size=256, CPU batch size=82026.01 | 525.74 | 1 | 363.11 | 1 |