Quantization format

UD_IQ2_XXS

UD_IQ2_XXS has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 199 files measured
Nominal bpw
from the block layout
Measured average
2.591
199 files
Range
1.596–5.374
varies by architecture
File sizes
0.17 GiB+
up to 312.21 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-27B8.74 GiB2.704
Qwen3.6-35B-A3B10.02 GiB2.394
Qwen3.5-9B2.97 GiB2.644
gemma-4-26B-A4B-it9.24 GiB2.990
DeepSeek-V4-Flash84.62 GiB2.498
Qwen3.5-4B1.42 GiB2.610
Qwen3-Coder-30B-A3B-Instruct9.62 GiB2.708
gemma-4-31B-it7.95 GiB2.183
Qwen3-VL-30B-A3B-Instruct9.63 GiB2.663
Qwen-AgentWorld-35B-A3B10.71 GiB2.654
Llama-3.2-1B-Instruct0.43 GiB3.006
Qwen3-8B2.43 GiB2.545
Qwen3.5-0.8B0.31 GiB3.098
Qwen3-4B1.17 GiB2.497
GLM-5.2222.08 GiB2.532
Llama-3.1-8B-Instruct2.33 GiB2.495
Ornith-1.0-35B10.71 GiB2.654
Llama-3.2-3B-Instruct0.97 GiB2.606
Kimi-K2.7-Code296.00 GiB2.402
Qwen3.5-122B-A10B34.12 GiB2.343
Qwen3-Coder-Next21.71 GiB2.341
gemma-3-1b-it0.53 GiB4.516
Qwen3.5-35B-A3B9.93 GiB2.371
Qwen3-14B4.16 GiB2.421
Qwen3-VL-4B-Instruct1.17 GiB2.263
From the filewhat these mean

Other quantizations