Quantization format

Q5_K_L

Q5_K_L has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 527 files measured
Nominal bpw
from the block layout
Measured average
5.964
527 files
Range
4.279–8.803
varies by architecture
File sizes
0.10 GiB+
up to 270.65 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B24.43 GiB5.836
gemma-4-26B-A4B-it18.16 GiB5.876
gemma-4-12B-it8.40 GiB6.033
Qwen3.5-4B3.35 GiB6.176
gemma-4-31B-it21.37 GiB5.871
Llama-3.2-1B-Instruct0.91 GiB6.312
gemma-4-E2B-it4.03 GiB6.753
Qwen3-8B5.81 GiB6.090
Qwen3.5-0.8B0.66 GiB6.482
Llama-3.1-8B-Instruct5.64 GiB6.034
Ornith-1.0-35B23.59 GiB5.847
Llama-3.2-3B-Instruct2.25 GiB6.020
ThinkingCap-Qwen3.6-27B20.06 GiB6.298
Qwythos-9B-v27.09 GiB6.313
Qwen3.5-122B-A10B84.66 GiB5.814
Qwen3-Coder-Next53.25 GiB5.741
Qwen2.5-32B-Instruct22.11 GiB5.797
Qwen3.5-35B-A3B24.43 GiB5.836
Qwen2.5-Coder-7B-Instruct5.38 GiB6.073
Qwen2.5-7B-Instruct5.38 GiB6.073
Qwen3-0.6B0.60 GiB6.891
Qwen3-VL-2B-Instruct1.24 GiB5.013
Jan-v3-4B-base-instruct3.16 GiB6.160
Qwen2.5-1.5B-Instruct1.10 GiB6.123
Qwen2.5-VL-7B-Instruct5.38 GiB5.577
From the filewhat these mean

Other quantizations