Quantization publisher

williamliao

williamliao publishes 30 quantizations across 10 models in our index, averaging 8.080 effective bits per weight. Their files differ in size from other publishers' builds of the same nominal quantization on 24 of the pairs we can compare — the same label does not mean the same file.

From the file· summed file bytes
Repositories
10
Quantizations
30
Models covered
10
Avg effective bpw
8.080
across their files

Same model, same quant label, different bytes

largest disagreements first
ModelQuantwilliamliaovsTheirsDifference
Qwen3.6-27B-NVFP4NVFP421.58 GiBneko-legends26.29 GiB-17.9%
Qwen3.6-27B-NVFP4NVFP419.99 GiBtngtech18.51 GiB+8.0%
Qwen3.6-35B-A3B-NVFP4NVFP421.10 GiBmudler22.21 GiB-5.0%
gemma-4-26B-A4B-it-DFlashQ8_00.44 GiBAlittlehammmer0.44 GiB+0.2%
gemma-4-26B-A4B-it-DFlashQ4_K_M0.25 GiBAlittlehammmer0.25 GiB+0.4%
gemma-4-26B-A4B-it-DFlashQ6_K0.34 GiBAlittlehammmer0.34 GiB+0.3%
gemma-4-26B-A4B-it-DFlashQ8_00.44 GiBAnbeeld0.44 GiB-0.0%
gemma-4-26B-A4B-it-DFlashQ6_K0.34 GiBAnbeeld0.34 GiB-0.0%
gemma-4-26B-A4B-it-DFlashQ5_K_M0.29 GiBAnbeeld0.29 GiB-0.0%
gemma-4-26B-A4B-it-DFlashQ4_K_M0.25 GiBAnbeeld0.25 GiB-0.0%
Qwen3.6-35B-A3B-DFlashQ6_K0.31 GiBAnbeeld0.31 GiB0.0%
Qwen3.6-35B-A3B-DFlashQ4_K_M0.22 GiBAnbeeld0.22 GiB0.0%
Qwen3.6-35B-A3B-DFlashQ5_K_M0.26 GiBAnbeeld0.26 GiB0.0%
Qwen3.6-35B-A3B-DFlashQ8_00.39 GiBAnbeeld0.39 GiB0.0%
Qwen3.6-27B-DFlashQ4_K_M0.96 GiBAnbeeld0.96 GiB0.0%
Qwen3.6-27B-DFlashQ6_K1.33 GiBAnbeeld1.33 GiB0.0%
Qwen3.6-27B-DFlashQ5_K_M1.14 GiBAnbeeld1.14 GiB0.0%
Qwen3.6-27B-DFlashQ8_01.72 GiBAnbeeld1.72 GiB0.0%
Qwen3.6-27B-DFlashQ6_K1.33 GiBAlittlehammmer1.33 GiB0.0%
Qwen3.6-27B-DFlashQ4_K_M0.96 GiBAlittlehammmer0.96 GiB0.0%
Qwen3.6-35B-A3B-DFlashQ6_K0.31 GiBAlittlehammmer0.31 GiB0.0%
Qwen3.6-35B-A3B-DFlashQ8_00.39 GiBAlittlehammmer0.39 GiB0.0%
Qwen3.6-35B-A3B-DFlashQ4_K_M0.22 GiBAlittlehammmer0.22 GiB0.0%
Qwen3.6-27B-DFlashQ8_01.72 GiBAlittlehammmer1.72 GiB0.0%

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
Qwen3.6-27B-DFlash60.87 GiB
Qwen3.6-27B-NVFP4119.99 GiB
Qwen3.6-27B-NVFP4121.58 GiB
Qwen3.6-35B-A3B-DFlash60.20 GiB
Qwen3.6-35B-A3B-NVFP4121.10 GiB
Ornith-1.0-35B-AEON-Ultimate-Uncensored-NVFP4121.32 GiB
gemma-4-26B-A4B-it-DFlash60.23 GiB
gemma-4-31B-it-NVFP4120.61 GiB
gemma4-12B-it-DFlash60.38 GiB
Gemma-4-26B-A4B-NVFP4116.46 GiB