Views
No views yet
| Setting | Value |
|---|---|
| Model | LiquidAI/LFM2.5-230M |
| Dataset | FineWeb-Edu (streamed) |
| Hook points | model.layers.6, model.layers.10 |
| Training tokens | 100M per SAE |
| Context length | 1,024 tokens |
| Input dimension | 1,024 |
| SAE width | 16,384 (16× expansion) |
| BatchTopK $K$ | 64 |
| Training batch size | 4,096 tokens |
| Learning rate | 10^-4 |
| Model precision | FP16 |
| SAE precision | FP32 |
| Activation normalisation | None |
| Random seed | 42 |