| license: mit | |
| tags: | |
| - quantization | |
| - codebook | |
| - sf | |
| # SF grids | |
| The 27 Gaussian VQ codebook grids used by the published SF model maps. | |
| Binary format: `[uint32 n][uint32 d][float64 distortion]` followed by `n*d` float32 centroid values. | |
| Model-specific weight and KV sensitivity maps are published separately as `dlab-cmu/sf-map-<model-name>` and collected in **SF grids and maps**. | |