Transformer generalization prediction on ViT-CIFAR10 generalization models (test)
0.9028Kendall's TauDNG-Encoder
Evaluation results
| Method | Links | |
|---|---|---|
| DNG-EncoderNumber of Parameters=~0.25M, GFLOPs=0.12, Peak Mem (MB)=25.36, Latency (ms)=4.382026.07 | 0.9028 | |
| NG-TNumber of Parameters=~0.25M, GFLOPs=44.69, Peak Mem (MB)=446.95, Latency (ms)=11.772026.07 | 0.8917 | |
| NG-GNNNumber of Parameters=~0.25M, GFLOPs=3.22, Peak Mem (MB)=54.49, Latency (ms)=5.942026.07 | 0.8844 |