Optimization Convergence on MoE-FMNIST (train)
0.632Final Training Loss GradientSyncSingle
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SyncSingleMomentum=None2026.06 | 0.632 | 0.309 | |
| ASADPRECMomentum=None2026.06 | 0.744 | 0.279 | |
| SyncMultiMomentum=None2026.06 | 0.826 | 0.314 |
| Method | Links | ||
|---|---|---|---|
| SyncSingleMomentum=None2026.06 | 0.632 | 0.309 | |
| ASADPRECMomentum=None2026.06 | 0.744 | 0.279 | |
| SyncMultiMomentum=None2026.06 | 0.826 | 0.314 |