Quantization format

UD_Q4_K_S

UD_Q4_K_S has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 28 files measured
Nominal bpw
from the block layout
Measured average
5.285
28 files
Range
4.521–12.367
varies by architecture
File sizes
4.67 GiB+
up to 548.12 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B19.46 GiB4.649
gemma-4-26B-A4B-it15.36 GiB4.969
Qwen-AgentWorld-35B-A3B19.46 GiB4.822
LTX-2.313.12 GiB12.257
GLM-5.2406.46 GiB4.635
Ornith-1.0-35B19.46 GiB4.822
Qwen3.5-122B-A10B68.39 GiB4.696
Qwen3-Coder-Next42.92 GiB4.627
LTX-210.83 GiB4.929
Laguna-S-2.163.88 GiB4.668
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1621.47 GiB5.585
Step-3.7-Flash106.32 GiB4.536
LFM2.5-8B-A1B4.67 GiB4.737
Ornith-1.0-397B213.94 GiB4.631
Qwen3.5-397B-A17B212.33 GiB4.521
North-Mini-Code-1.016.81 GiB4.736
MiniMax-M2.7121.97 GiB4.581
MiniMax-M3230.71 GiB4.641
NVIDIA-Nemotron-3-Super-120B-A12B-BF1673.59 GiB5.114
Mistral-Small-4-119B-260364.70 GiB4.654
Huihui-gemma-4-26B-A4B-it-abliterated15.27 GiB4.940
MiMo-V2.5166.56 GiB4.604
GLM-5.1404.10 GiB4.605
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16305.18 GiB4.677
MiMo-V2.5-Pro548.12 GiB4.601
From the filewhat these mean

Other quantizations