Quantization format

IQ2_S

IQ2_S has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 452 files measured
Nominal bpw
from the block layout
Measured average
2.712
452 files
Range
2.145–9.233
varies by architecture
File sizes
0.02 GiB+
up to 311.72 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B11.09 GiB2.650
gemma-4-26B-A4B-it9.53 GiB3.083
gemma-4-12B-it4.39 GiB3.150
Hy386.43 GiB2.485
gemma-4-31B-it11.25 GiB3.089
Ornith-1.0-35B10.25 GiB2.541
ThinkingCap-Qwen3.6-27B9.59 GiB3.011
Qwen3.5-122B-A10B37.85 GiB2.599
Qwen3-Coder-Next21.76 GiB2.346
Qwen2.5-32B-Instruct9.67 GiB2.536
Qwen3.5-35B-A3B11.09 GiB2.650
Qwen3-30B-A3B8.59 GiB2.416
Qwen2.5-Coder-32B-Instruct9.67 GiB2.536
Qwen3.5-27B9.79 GiB3.026
Laguna-S-2.133.77 GiB2.467
Qwen2.5-Coder-14B-Instruct4.66 GiB2.710
Qwen3-30B-A3B-Instruct-25078.14 GiB2.291
Voxtral-Small-24B-25076.97 GiB2.466
Llama-3.3-70B-Instruct20.71 GiB2.522
DeepSeek-R1-Distill-Llama-70B20.71 GiB2.522
Step-3.7-Flash57.93 GiB2.471
gemma-3-27b-it8.18 GiB2.561
GLM-4.6V-Flash3.52 GiB2.941
DeepSeek-R1-Distill-Qwen-14B4.66 GiB2.710
Mistral-Small-24B-Instruct-25016.96 GiB2.538
From the filewhat these mean

Other quantizations