Quantization format

IQ2_XS

IQ2_XS has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 447 files measured
Nominal bpw
from the block layout
Measured average
2.614
447 files
Range
1.608–9.993
varies by architecture
File sizes
0.02 GiB+
up to 285.38 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B10.89 GiB2.602
gemma-4-26B-A4B-it9.43 GiB3.052
Hy384.81 GiB2.438
gemma-4-31B-it10.71 GiB2.942
Ornith-1.0-35B10.06 GiB2.492
Qwopus3.6-27B-Coder8.89 GiB2.748
ThinkingCap-Qwen3.6-27B9.30 GiB2.921
Qwen3.5-122B-A10B37.23 GiB2.556
Qwen3-Coder-Next20.69 GiB2.231
Qwen2.5-32B-Instruct9.27 GiB2.431
Qwen3.5-35B-A3B10.89 GiB2.602
Qwen3-30B-A3B8.51 GiB2.395
Qwen2.5-Coder-32B-Instruct9.27 GiB2.431
Qwen3.5-27B9.50 GiB2.937
Laguna-S-2.133.12 GiB2.420
Qwen2.5-Coder-14B-Instruct4.38 GiB2.548
Phi-3.5-mini-instruct1.07 GiB2.414
gemma-2-2b-it0.93 GiB3.068
Qwen3-30B-A3B-Instruct-25078.07 GiB2.270
Voxtral-Small-24B-25076.71 GiB2.377
Llama-3.3-70B-Instruct19.69 GiB2.397
DeepSeek-R1-Distill-Llama-70B19.69 GiB2.397
Step-3.7-Flash56.76 GiB2.421
gemma-3-27b-it7.86 GiB2.461
DeepSeek-R1-Distill-Qwen-14B4.38 GiB2.548
From the filewhat these mean

Other quantizations