Quantization format

IQ4_NL

IQ4_NL has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 840 files measured
Nominal bpw
from the block layout
Measured average
4.779
840 files
Range
2.217–32.274
varies by architecture
File sizes
0.03 GiB+
up to 539.67 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-27B14.97 GiB4.628
Qwen3.6-35B-A3B19.33 GiB4.618
Qwen3.5-9B5.00 GiB4.451
gemma-4-26B-A4B-it13.69 GiB4.430
gemma-4-12B-it6.26 GiB4.493
Qwen3.5-4B2.40 GiB4.429
Hy3158.14 GiB4.546
gemma-4-E4B-it4.50 GiB4.838
Qwen3-Coder-30B-A3B-Instruct16.12 GiB4.536
gemma-4-31B-it16.10 GiB4.422
Qwen3-VL-30B-A3B-Instruct16.12 GiB4.457
Llama-3.2-1B-Instruct0.72 GiB5.004
gemma-4-E2B-it2.83 GiB4.749
Qwen3-8B4.46 GiB4.682
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP32.65 GiB10.095
Qwen3.5-0.8B0.47 GiB4.642
Qwen3-4B2.22 GiB4.736
Llama-3.1-8B-Instruct4.36 GiB4.660
Ornith-1.0-35B18.50 GiB4.584
Llama-3.2-3B-Instruct1.79 GiB4.774
ThinkingCap-Qwen3.6-27B15.20 GiB4.774
Qwythos-9B-v25.23 GiB4.654
Qwen3.5-122B-A10B67.26 GiB4.619
Qwen3-Coder-Next42.03 GiB4.532
gemma-3-1b-it0.67 GiB5.776
From the filewhat these mean

Other quantizations