Quantization format
IQ1_S
IQ1_S has a nominal rate of — bits per weight. We have no measured files carrying this label yet.
From the file· 192 files measured
Nominal bpw
—
from the block layout
Measured average
2.075
192 files
Range
1.561–7.254
varies by architecture
File sizes
0.02 GiB+
up to 528.03 GiB
Real files
one per model, most downloaded first
| Model | Size● | Effective bpw● | vs nominal● |
|---|---|---|---|
| Hy3 | 59.65 GiB | 1.715 | — |
| gemma-4-E2B-it | 2.16 GiB | 3.616 | — |
| Qwen3.5-122B-A10B | 26.92 GiB | 1.849 | — |
| Qwen3-Coder-Next | 15.44 GiB | 1.665 | — |
| jina-embeddings-v5-text-small | 0.19 GiB | 2.792 | — |
| Laguna-S-2.1 | 23.15 GiB | 1.691 | — |
| Phi-3.5-mini-instruct | 0.78 GiB | 1.762 | — |
| gemma-2-2b-it | 0.78 GiB | 2.546 | — |
| Step-3.7-Flash | 40.03 GiB | 1.708 | — |
| Mistral-7B-Instruct-v0.3 | 1.50 GiB | 1.783 | — |
| Meta-Llama-3-8B-Instruct | 1.88 GiB | 2.013 | — |
| Llama-3.1-70B-Instruct | 14.29 GiB | 1.740 | — |
| jina-embeddings-v5-text-nano | 0.09 GiB | 3.756 | — |
| Ornith-1.0-397B | 76.16 GiB | 1.649 | — |
| Qwen3.5-397B-A17B | 82.64 GiB | 1.760 | — |
| Qwen2-7B-Instruct | 1.77 GiB | 2.000 | — |
| Mixtral-8x22B-v0.1 | 27.61 GiB | 1.687 | — |
| DeepSeek-V3-0324 | 124.38 GiB | 1.561 | — |
| Mistral-Small-Instruct-2409 | 4.50 GiB | 1.737 | — |
| Llama-3-8B-Instruct-32k-v0.1 | 1.88 GiB | 2.012 | — |
| Yi-Coder-1.5B-Chat | 0.46 GiB | 2.661 | — |
| Yi-Coder-9B-Chat | 1.88 GiB | 1.825 | — |
| Mathstral-7B-v0.1 | 1.50 GiB | 1.783 | — |
| solar-pro-preview-instruct | 4.46 GiB | 1.730 | — |
| llama-3-firefunction-v2 | 14.29 GiB | 1.740 | — |
●From the filewhat these mean