Long-trajectory rollout on FNO-1e-3 (test)
0.0004L2 Relative Error @ 20 stepsAOT-POT-FT-S
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| AOT-POT-FT-SEvaluation protocol=Fine-tuning, Model scale=Small, Resolution=128x1282026.05 | 0.0004 | 0.0005 | 0.001 | 0.003 | 0.004 | |
| AOT-POT-SEvaluation protocol=Pre-trained, Model scale=Small, Resolution=128x1282026.05 | 0.0008 | 0.001 | 0.0015 | 0.0064 | 0.0154 | |
| DPOT-FT-SEvaluation protocol=Fine-tuning, Model scale=Small, Resolution=128x1282026.05 | 0.0012 | 0.0014 | 0.0029 | 0.0093 | 0.0291 | |
| MoE-POT-FT-SEvaluation protocol=Fine-tuning, Model scale=Small, Resolution=128x1282026.05 | 0.0013 | 0.0015 | 0.0028 | 0.0101 | 0.0305 | |
| DPOT-SEvaluation protocol=Pre-trained, Model scale=Small, Resolution=128x1282026.05 | 0.0015 | 0.0016 | 0.0047 | 0.0169 | 0.0669 | |
| MoE-POT-SEvaluation protocol=Pre-trained, Model scale=Small, Resolution=128x1282026.05 | 0.0017 | 0.002 | 0.0049 | 0.0193 | 0.0702 |