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
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Qwen3.6-35B-A3B | 10.89 GiB | 2.602 | — |
| gemma-4-26B-A4B-it | 9.43 GiB | 3.052 | — |
| Hy3 | 84.81 GiB | 2.438 | — |
| gemma-4-31B-it | 10.71 GiB | 2.942 | — |
| Ornith-1.0-35B | 10.06 GiB | 2.492 | — |
| Qwopus3.6-27B-Coder | 8.89 GiB | 2.748 | — |
| ThinkingCap-Qwen3.6-27B | 9.30 GiB | 2.921 | — |
| Qwen3.5-122B-A10B | 37.23 GiB | 2.556 | — |
| Qwen3-Coder-Next | 20.69 GiB | 2.231 | — |
| Qwen2.5-32B-Instruct | 9.27 GiB | 2.431 | — |
| Qwen3.5-35B-A3B | 10.89 GiB | 2.602 | — |
| Qwen3-30B-A3B | 8.51 GiB | 2.395 | — |
| Qwen2.5-Coder-32B-Instruct | 9.27 GiB | 2.431 | — |
| Qwen3.5-27B | 9.50 GiB | 2.937 | — |
| Laguna-S-2.1 | 33.12 GiB | 2.420 | — |
| Qwen2.5-Coder-14B-Instruct | 4.38 GiB | 2.548 | — |
| Phi-3.5-mini-instruct | 1.07 GiB | 2.414 | — |
| gemma-2-2b-it | 0.93 GiB | 3.068 | — |
| Qwen3-30B-A3B-Instruct-2507 | 8.07 GiB | 2.270 | — |
| Voxtral-Small-24B-2507 | 6.71 GiB | 2.377 | — |
| Llama-3.3-70B-Instruct | 19.69 GiB | 2.397 | — |
| DeepSeek-R1-Distill-Llama-70B | 19.69 GiB | 2.397 | — |
| Step-3.7-Flash | 56.76 GiB | 2.421 | — |
| gemma-3-27b-it | 7.86 GiB | 2.461 | — |
| DeepSeek-R1-Distill-Qwen-14B | 4.38 GiB | 2.548 | — |
●From the filewhat these mean