Quantization format

Q2_K_M

Q2_K_M has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 8 files measured
Nominal bpw
from the block layout
Measured average
3.451
8 files
Range
2.895–4.210
varies by architecture
File sizes
1.70 GiB+
up to 12.61 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
GRM-2.6-Plus-062812.61 GiB3.898
gemma-4-12B4.57 GiB3.280
Ministral-3-8B-Instruct-2512-BF163.44 GiB3.313
granite-3.1-8b-instruct2.85 GiB2.992
Magistral-Small-2509-Vision8.09 GiB2.895
orpheus-3b-0.1-pretrained1.70 GiB3.855
Pixtral-12B-24096.00 GiB4.210
MiniMax-H312.20 GiB3.164
From the filewhat these mean

Other quantizations