Doge 40M MoE checkpoint
Doge uses wsd_scheduler as the training scheduler, which divides the learning rate into three stages: warmup, stable, and decay. It allows us to continue training on any new dataset from any checkpoint in the stable stage without spikes in training.
Here are the initial learning rates required to continue training at each checkpoint:
- Doge-40M: 8e-3
- Doge-40M-MoE: 8e-3
| Model | Learning Rate | Schedule | Warmup Steps | Stable Steps |
|---|---|---|---|---|
| Doge-40M | 8e-3 | wsd_scheduler | 2000 | 4000 |
| Doge-40M-MoE | 8e-3 | wsd_scheduler | 2000 | 4000 |