Quantization format

IQ2_M

IQ2_M has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 745 files measured
Nominal bpw
from the block layout
Measured average
2.988
745 files
Range
1.178–10.009
varies by architecture
File sizes
0.02 GiB+
up to 321.02 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B12.07 GiB2.885
gemma-4-26B-A4B-it9.97 GiB3.225
gemma-4-12B-it4.60 GiB3.305
Qwen3.5-4B1.82 GiB3.354
Hy393.14 GiB2.678
gemma-4-31B-it11.78 GiB3.236
gemma-4-E2B-it2.38 GiB3.987
Qwen3-8B2.84 GiB2.981
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP22.16 GiB6.853
Qwen3.5-0.8B0.37 GiB3.651
Llama-3.1-8B-Instruct2.75 GiB2.937
Ornith-1.0-35B11.24 GiB2.785
Qwopus3.6-27B-Coder9.74 GiB3.011
ThinkingCap-Qwen3.6-27B10.13 GiB3.180
Qwythos-9B-v23.64 GiB3.237
Qwen3.5-122B-A10B41.39 GiB2.842
Qwen3-Coder-Next24.31 GiB2.620
Qwen2.5-32B-Instruct10.49 GiB2.751
Qwen3.5-35B-A3B12.07 GiB2.885
Qwen2.5-Coder-7B-Instruct2.59 GiB2.921
Qwen2.5-7B-Instruct2.59 GiB2.921
Qwen3-0.6B0.31 GiB3.531
Qwen3-VL-2B-Instruct0.65 GiB2.614
Jan-v3-4B-base-instruct1.56 GiB3.047
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking13.76 GiB2.990
From the filewhat these mean

Other quantizations