Quantization format

UD_Q3_K_M

UD_Q3_K_M has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 37 files measured
Nominal bpw
from the block layout
Measured average
4.412
37 files
Range
3.369–11.747
varies by architecture
File sizes
3.67 GiB+
up to 431.78 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B15.46 GiB3.694
gemma-4-26B-A4B-it11.85 GiB3.836
DeepSeek-V4-Flash120.44 GiB3.556
Qwen-AgentWorld-35B-A3B15.53 GiB3.850
LTX-2.312.26 GiB11.455
GLM-5.2319.20 GiB3.640
Ornith-1.0-35B15.53 GiB3.850
Kimi-K2.7-Code431.78 GiB3.504
Qwen3.5-122B-A10B54.20 GiB3.722
Qwen3-Coder-Next33.47 GiB3.608
LTX-29.39 GiB4.271
Laguna-S-2.150.31 GiB3.676
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1618.19 GiB4.734
Step-3.7-Flash83.13 GiB3.546
LFM2.5-8B-A1B3.67 GiB3.722
Ornith-1.0-397B167.62 GiB3.629
Qwen3.5-397B-A17B165.23 GiB3.518
North-Mini-Code-1.013.24 GiB3.730
MiniMax-M2.794.29 GiB3.542
MiniMax-M3181.22 GiB3.645
NVIDIA-Nemotron-3-Super-120B-A12B-BF1657.47 GiB3.994
Mistral-Small-4-119B-260350.64 GiB3.643
Huihui-gemma-4-26B-A4B-it-abliterated11.67 GiB3.775
MiMo-V2.5130.41 GiB3.605
GLM-5.1315.19 GiB3.591
From the filewhat these mean

Other quantizations