Quantization format
UD_Q3_K_S
UD_Q3_K_S has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 10 files measured
Nominal bpw
—
from the block layout
Measured average
4.737
10 files
Range
3.259–9.909
varies by architecture
File sizes
10.43 GiB+
up to 291.94 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3.6-35B-A3B | 14.30 GiB | 3.418 | — |
| LTX-2.3 | 10.43 GiB | 9.745 | — |
| Qwen3-Coder-Next | 31.03 GiB | 3.346 | — |
| Qwen3.5-397B-A17B | 153.04 GiB | 3.259 | — |
| MiniMax-M2.7 | 87.21 GiB | 3.275 | — |
| NVIDIA-Nemotron-3-Super-120B-A12B-BF16 | 57.47 GiB | 3.994 | — |
| Mistral-Small-4-119B-2603 | 46.23 GiB | 3.326 | — |
| Huihui-gemma-4-26B-A4B-it-abliterated | 11.67 GiB | 3.775 | — |
| GLM-5.1 | 291.94 GiB | 3.326 | — |
●From the filewhat these mean