Quantization publisher

kai-os

kai-os publishes 7 quantizations across 2 models in our index, averaging 6.508 effective bits per weight. Their files differ in size from other publishers' builds of the same nominal quantization on 7 of the pairs we can compare — the same label does not mean the same file.

From the file· summed file bytes
Repositories
2
Quantizations
7
Models covered
2
Avg effective bpw
6.508
across their files

Same model, same quant label, different bytes

largest disagreements first
ModelQuantkai-osvsTheirsDifference
Carnice-V2-27bQ5_K_M17.91 GiBbartowski19.10 GiB-6.3%
Carnice-V2-27bQ4_K_M15.41 GiBbartowski16.33 GiB-5.6%
Carnice-V2-27bQ2_K9.98 GiBbartowski10.80 GiB-7.7%
Carnice-V2-27bIQ2_M9.32 GiBbartowski9.90 GiB-5.9%
Grug-12BQ4_K_M6.87 GiBbartowski7.14 GiB-3.7%
Carnice-V2-27bQ8_026.63 GiBbartowski26.70 GiB-0.2%
Carnice-V2-27bBF1650.11 GiBbartowski50.11 GiB-0.0%

A quantization label describes a target, not a recipe. Publishers make different choices about which tensors to keep at higher precision, and some apply an importance matrix while others don't — so two files both honestly labelled the same thing can differ measurably in size and in quality.

Models they publish

ModelQuantizationsSmallest
Grug-12B16.87 GiB
Carnice-V2-27b69.32 GiB