Quantization format

UD_Q4_K_M

UD_Q4_K_M has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 32 files measured
Nominal bpw
from the block layout
Measured average
5.947
32 files
Range
4.841–14.358
varies by architecture
File sizes
4.96 GiB+
up to 586.37 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B20.61 GiB4.925
gemma-4-26B-A4B-it15.78 GiB5.108
Qwen-AgentWorld-35B-A3B20.61 GiB5.109
LTX-2.315.19 GiB14.183
GLM-5.2433.83 GiB4.947
Ornith-1.0-35B20.61 GiB5.109
Qwen3.5-122B-A10B72.89 GiB5.005
Qwen3-Coder-Next45.92 GiB4.950
Laguna-S-2.168.10 GiB4.976
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1622.25 GiB5.788
Step-3.7-Flash113.71 GiB4.851
LFM2.5-8B-A1B4.96 GiB5.028
Ornith-1.0-397B228.19 GiB4.940
Qwen3.5-397B-A17B227.33 GiB4.841
North-Mini-Code-1.017.88 GiB5.040
MiniMax-M2.7130.54 GiB4.903
Nex-N2-mini20.60 GiB5.039
MiniMax-M3245.90 GiB4.946
NVIDIA-Nemotron-3-Super-120B-A12B-BF1676.87 GiB5.342
Mistral-Small-4-119B-260368.70 GiB4.942
Huihui-gemma-4-26B-A4B-it-abliterated15.71 GiB5.084
MiMo-V2.5177.81 GiB4.915
GLM-5.1432.60 GiB4.929
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16334.55 GiB5.127
ERNIE-Image-Turbo5.38 GiB5.755
From the filewhat these mean

Other quantizations