Quantization format
UD_TQ1_0
UD_TQ1_0 has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 50 files measured
Nominal bpw
—
from the block layout
Measured average
2.067
50 files
Range
1.810–2.771
varies by architecture
File sizes
1.81 GiB+
up to 229.90 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3-Coder-30B-A3B-Instruct | 7.46 GiB | 2.098 | — |
| Qwen3-VL-30B-A3B-Instruct | 7.60 GiB | 2.102 | — |
| Qwen3-Coder-Next | 17.64 GiB | 1.902 | — |
| GLM-4.7-Flash | 7.76 GiB | 2.135 | — |
| Qwen3-30B-A3B-Instruct-2507 | 7.54 GiB | 2.121 | — |
| Llama-4-Scout-17B-16E-Instruct | 27.25 GiB | 2.155 | — |
| Kimi-K2-Instruct | 226.87 GiB | 1.899 | — |
| GLM-4.5-Air | 35.63 GiB | 2.771 | — |
| Qwen3-Next-80B-A3B-Instruct | 19.06 GiB | 2.013 | — |
| MiniMax-M2.5 | 51.94 GiB | 1.951 | — |
| grok-2 | 76.17 GiB | 2.428 | — |
| GLM-4.7-Flash-REAP-23B-A3B | 6.09 GiB | 2.274 | — |
| DeepSeek-V3.1 | 158.79 GiB | 1.993 | — |
| Qwen3-Next-80B-A3B-Thinking | 18.83 GiB | 1.989 | — |
| granite-4.0-h-tiny | 1.81 GiB | 2.235 | — |
| GLM-4.6V | 34.27 GiB | 2.733 | — |
| DeepSeek-V3.2 | 150.20 GiB | 1.883 | — |
| Qwen3-VL-30B-A3B-Thinking | 7.59 GiB | 2.098 | — |
| Llama-4-Maverick-17B-128E-Instruct | 98.45 GiB | 2.106 | — |
| Qwen3-VL-235B-A22B-Instruct | 50.67 GiB | 1.847 | — |
| Qwen3-VL-235B-A22B-Thinking | 50.83 GiB | 1.853 | — |
| GLM-4.7 | 78.69 GiB | 1.886 | — |
| Qwen3-30B-A3B-Thinking-2507 | 7.59 GiB | 2.135 | — |
| DeepSeek-R1-0528 | 150.52 GiB | 1.889 | — |
| ERNIE-4.5-21B-A3B-PT | 6.06 GiB | 2.371 | — |
●From the filewhat these mean