Image Classification on CIFAR-100 (test) (Accuracy and Latency Profile)
83.52Accuracy (%)HARMONY
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HARMONYrho=0, Backbone=ViT-Small-16, N=30, Shard=32026.05 | 83.52 | 34.26 | |
| SplitGPlambda=0.9, rho=0, Backbone=ViT-Small-16, N=30, Shard=32026.05 | 83.06 | 37.4 | |
| SplitGP + FTlambda=0.2, rho=0, Backbone=ViT-Small-16, N=30, Shard=3, On-device fine-tuning=true2026.05 | 70.73 | 38.12 | |
| SplitGPlambda=0.2, rho=0, Backbone=ViT-Small-16, N=30, Shard=32026.05 | 70.62 | 33.65 | |
| HARMONYrho=97/300, Backbone=ViT-Small-16, N=30, Shard=32026.05 | 63.15 | 45.53 | |
| SplitGPlambda=0.9, rho=97/300, Backbone=ViT-Small-16, N=30, Shard=32026.05 | 62.76 | 49.54 | |
| SplitGPlambda=0.2, rho=97/300, Backbone=ViT-Small-16, N=30, Shard=32026.05 | 53.74 | 44.72 | |
| SplitGP + FTlambda=0.2, rho=97/300, Backbone=ViT-Small-16, N=30, Shard=3, On-device fine-tuning=true2026.05 | 53.55 | 50.71 |