Views
No views yet
| Baseline | Subspace | |
|---|---|---|
| Parameters | 1,279,228,550 | 1,479,545,460 |
| Layers | 48 | 48 |
| Width | 1250 | 1500 |
| Heads | 25 | 25 |
| MLP Ratio | 4.0 | 3.0 |
| Q/K/V Fractions | - | 0.1/0.1/0.8 |
| Readout | - | v |
| Training Steps | 87,999 | 93,999 |
| Context Length | 2048 | 2048 |
| Vocab Size | 151,665 | 151,665 |
| Best Val Loss | 2.5319 | 2.5254 |
baseline/model.pt - Baseline transformer checkpointsubspace/model.pt - Subspace transformer checkpointbaseline/config.json - Baseline resolved configsubspace/config.json - Subspace resolved config