Image Classification on ImageNet-1K (val) (Peak Throughput)
25,348Peak Throughput (images/s)Triton (ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| Triton (ours)Model=DeiT-Ti, Backend=Triton, Pruning Ratio=50%, Memory Constraint=40 GB2026.04 | 25,348 | |
| UnprunedModel=DeiT-Ti, Pruning Ratio=0%, Memory Constraint=40 GB2026.04 | 19,499 | |
| Triton (ours)Model=DeiT-S, Backend=Triton, Pruning Ratio=50%, Memory Constraint=40 GB2026.04 | 12,999 | |
| UnprunedModel=DeiT-S, Pruning Ratio=0%, Memory Constraint=40 GB2026.04 | 9,791 | |
| PaddedModel=DeiT-Ti, Backend=PyTorch (padded), Pruning Ratio=50%, Memory Constraint=40 GB2026.04 | 7,017 | |
| Triton (ours)Model=DeiT-B, Backend=Triton, Pruning Ratio=50%, Memory Constraint=40 GB2026.04 | 5,268 | |
| PaddedModel=DeiT-S, Backend=PyTorch (padded), Pruning Ratio=50%, Memory Constraint=40 GB2026.04 | 4,593 | |
| UnprunedModel=DeiT-B, Pruning Ratio=0%, Memory Constraint=40 GB2026.04 | 3,749 | |
| PaddedModel=DeiT-B, Backend=PyTorch (padded), Pruning Ratio=50%, Memory Constraint=40 GB2026.04 | 2,303 |