Quantization format
UD_IQ3_XXS
UD_IQ3_XXS has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 213 files measured
Nominal bpw
—
from the block layout
Measured average
3.298
213 files
Range
2.075–5.506
varies by architecture
File sizes
0.11 GiB+
up to 392.74 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3.6-27B | 11.17 GiB | 3.454 | — |
| Qwen3.6-35B-A3B | 12.30 GiB | 2.940 | — |
| Qwen3.5-9B | 3.74 GiB | 3.329 | — |
| gemma-4-26B-A4B-it | 10.63 GiB | 3.441 | — |
| gemma-4-12B-it | 4.32 GiB | 3.104 | — |
| DeepSeek-V4-Flash | 95.93 GiB | 2.832 | — |
| Qwen3.5-4B | 1.82 GiB | 3.346 | — |
| gemma-4-E4B-it | 3.46 GiB | 3.719 | — |
| Qwen3-Coder-30B-A3B-Instruct | 11.97 GiB | 3.367 | — |
| gemma-4-31B-it | 11.02 GiB | 3.028 | — |
| Qwen3-VL-30B-A3B-Instruct | 12.00 GiB | 3.318 | — |
| Qwen-AgentWorld-35B-A3B | 12.80 GiB | 3.172 | — |
| Llama-3.2-1B-Instruct | 0.54 GiB | 3.725 | — |
| gemma-4-E2B-it | 2.21 GiB | 3.705 | — |
| Qwen3-8B | 3.18 GiB | 3.331 | — |
| Qwen3.5-0.8B | 0.37 GiB | 3.647 | — |
| Qwen3-4B | 1.56 GiB | 3.330 | — |
| GLM-5.2 | 262.34 GiB | 2.991 | — |
| Llama-3.1-8B-Instruct | 3.09 GiB | 3.309 | — |
| Ornith-1.0-35B | 12.80 GiB | 3.172 | — |
| Llama-3.2-3B-Instruct | 1.28 GiB | 3.412 | — |
| Kimi-K2.7-Code | 351.00 GiB | 2.848 | — |
| Qwen3.5-122B-A10B | 41.67 GiB | 2.862 | — |
| Qwen3-Coder-Next | 26.53 GiB | 2.860 | — |
| gemma-3-1b-it | 0.55 GiB | 4.734 | — |
●From the filewhat these mean