sophia_300m_1 / README.md
KaiyueWen's picture
Upload folder using huggingface_hub
673256d verified
# Model Card
- Source: [https://arxiv.org/abs/2509.02046](https://arxiv.org/abs/2509.02046)
- Optimizer: `sophia`
- Model size: `300m`
- Data size: `6B`
## Best configuration
| Hyperparameter | Value |
|---|---|
| beta1 | `0.9` |
| beta2 | `0.9` |
| epsilon | `1e-07` |
| gamma | `0.0125` |
| learning_rate | `0.004` |
| max_grad_norm | `1` |
| min_lr_ratio | `0` |
| train_batch_size | `128` |
| warmup | `4000` |
| weight_decay | `0.1` |