Quantization format

Q2_K_L

Q2_K_L has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 764 files measured
Nominal bpw
from the block layout
Measured average
3.379
764 files
Range
2.095–7.075
varies by architecture
File sizes
0.08 GiB+
up to 348.68 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B13.04 GiB3.116
gemma-4-26B-A4B-it10.37 GiB3.355
gemma-4-12B-it4.96 GiB3.560
Qwen3.5-4B2.18 GiB4.027
Hy399.81 GiB2.870
Qwen3-Coder-30B-A3B-Instruct10.55 GiB2.969
gemma-4-31B-it12.08 GiB3.319
Qwen3-VL-30B-A3B-Instruct10.55 GiB2.918
Llama-3.2-1B-Instruct0.54 GiB3.760
gemma-4-E2B-it3.43 GiB5.756
gpt-oss-20b10.95 GiB4.373
Qwen3-8B3.19 GiB3.348
Qwen3.5-0.8B0.49 GiB4.816
Qwen3-4B1.55 GiB3.320
Llama-3.1-8B-Instruct3.08 GiB3.290
Ornith-1.0-35B12.21 GiB3.027
Llama-3.2-3B-Instruct1.27 GiB3.396
ThinkingCap-Qwen3.6-27B12.18 GiB3.825
Qwythos-9B-v24.84 GiB4.305
Qwen3.5-122B-A10B43.90 GiB3.015
LFM2.5-1.2B-Instruct0.45 GiB3.304
Qwen3-Coder-Next26.52 GiB2.859
Qwen2.5-32B-Instruct12.18 GiB3.192
gemma-3-1b-it0.64 GiB5.519
Qwen3.5-35B-A3B13.04 GiB3.116
From the filewhat these mean

Other quantizations