Quantization format

IQ3_XS

IQ3_XS has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 788 files measured
Nominal bpw
from the block layout
Measured average
3.562
788 files
Range
1.297–13.839
varies by architecture
File sizes
0.03 GiB+
up to 434.58 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B15.94 GiB3.807
gemma-4-26B-A4B-it11.58 GiB3.748
gemma-4-12B-it5.15 GiB3.696
Qwen3.5-4B2.24 GiB4.127
Hy3128.00 GiB3.680
gemma-4-31B-it12.89 GiB3.541
gemma-4-E2B-it2.89 GiB4.842
Qwen3-8B3.38 GiB3.542
Qwen3.5-0.8B0.46 GiB4.560
Llama-3.1-8B-Instruct3.28 GiB3.506
Ornith-1.0-35B15.10 GiB3.743
ThinkingCap-Qwen3.6-27B12.41 GiB3.898
Qwythos-9B-v24.37 GiB3.893
Qwen3.5-122B-A10B55.13 GiB3.786
Qwen3-Coder-Next30.76 GiB3.317
Qwen2.5-32B-Instruct12.76 GiB3.346
Qwen3.5-35B-A3B15.94 GiB3.807
Qwen2.5-Coder-7B-Instruct3.12 GiB3.515
Qwen2.5-7B-Instruct3.12 GiB3.515
Qwen3-0.6B0.35 GiB4.040
Qwen3-VL-2B-Instruct0.78 GiB3.137
Jan-v3-4B-base-instruct1.85 GiB3.593
Qwen2.5-1.5B-Instruct0.68 GiB3.792
Qwen2.5-VL-7B-Instruct3.12 GiB3.228
Ornith-1.0-9B4.25 GiB3.967
From the filewhat these mean

Other quantizations