Quantization format
UD_IQ2_M
UD_IQ2_M has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 206 files measured
Nominal bpw
—
from the block layout
Measured average
2.937
206 files
Range
1.874–5.853
varies by architecture
File sizes
0.10 GiB+
up to 329.27 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3.6-27B | 10.10 GiB | 3.123 | — |
| Qwen3.6-35B-A3B | 10.73 GiB | 2.564 | — |
| Qwen3.5-9B | 3.40 GiB | 3.024 | — |
| gemma-4-26B-A4B-it | 9.33 GiB | 3.018 | — |
| gemma-4-12B-it | 3.92 GiB | 2.818 | — |
| DeepSeek-V4-Flash | 84.68 GiB | 2.500 | — |
| Qwen3.5-4B | 1.64 GiB | 3.022 | — |
| gemma-4-E4B-it | 3.30 GiB | 3.547 | — |
| Qwen3-Coder-30B-A3B-Instruct | 10.09 GiB | 2.840 | — |
| gemma-4-31B-it | 10.01 GiB | 2.751 | — |
| Qwen3-VL-30B-A3B-Instruct | 10.10 GiB | 2.792 | — |
| Qwen-AgentWorld-35B-A3B | 10.77 GiB | 2.669 | — |
| Llama-3.2-1B-Instruct | 0.50 GiB | 3.472 | — |
| gemma-4-E2B-it | 2.13 GiB | 3.577 | — |
| Qwen3-8B | 2.90 GiB | 3.038 | — |
| Qwen3.5-0.8B | 0.35 GiB | 3.407 | — |
| Qwen3-4B | 1.43 GiB | 3.049 | — |
| GLM-5.2 | 222.19 GiB | 2.534 | — |
| Llama-3.1-8B-Instruct | 2.80 GiB | 2.992 | — |
| Ornith-1.0-35B | 10.77 GiB | 2.669 | — |
| Llama-3.2-3B-Instruct | 1.17 GiB | 3.128 | — |
| Kimi-K2.7-Code | 296.14 GiB | 2.403 | — |
| Qwen3.5-122B-A10B | 36.46 GiB | 2.504 | — |
| Qwen3-Coder-Next | 23.25 GiB | 2.506 | — |
| gemma-3-1b-it | 0.54 GiB | 4.625 | — |
●From the filewhat these mean