Dynamic Mixture of Experts (DynMoE) implementation on New-feature tasks
60.4Session DurationPithTrain
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PithTrainFramework=PithTrain2026.05 | 60.4 | 41.9 | 76 | 146 | 76.4 | |
| PithTrainFramework=PithTrain2026.05 | 63 | 39.9 | 90 | 176.6 | 107.7 | |
| TorchTitanFramework=TorchTitan2026.05 | 71.4 | 51.9 | 87 | 164.6 | 85.3 | |
| Megatron-LMFramework=Megatron-LM2026.05 | 83.8 | 49.1 | 199 | 208 | 115.2 | |
| Megatron-LMFramework=Megatron-LM2026.05 | 88.5 | 58.7 | 145 | 188.7 | 117 | |
| TorchTitanFramework=TorchTitan2026.05 | 140.6 | 94.4 | 197 | 228.8 | 161.3 |