Quantization publisher

stepfun-ai

stepfun-ai publishes 7 quantizations across 1 models in our index, averaging 6.198 effective bits per weight. Their files differ in size from other publishers' builds of the same nominal quantization on 8 of the pairs we can compare — the same label does not mean the same file.

From the file· summed file bytes
Repositories
1
Quantizations
7
Models covered
1
Avg effective bpw
6.198
across their files

Same model, same quant label, different bytes

largest disagreements first
ModelQuantstepfun-aivsTheirsDifference
Step-3.7-FlashIQ3_XXS70.56 GiBbartowski78.38 GiB-10.0%
Step-3.7-FlashQ3_K_L95.46 GiBbartowski88.94 GiB+7.3%
Step-3.7-FlashQ4_K_S103.84 GiBbartowski109.05 GiB-4.8%
Step-3.7-FlashQ8_0195.04 GiBbartowski197.44 GiB-1.2%
Step-3.7-FlashIQ4_XS97.78 GiBbartowski99.94 GiB-2.2%
Step-3.7-FlashQ3_K_M87.36 GiBbartowski85.51 GiB+2.2%
Step-3.7-FlashQ8_0195.04 GiBunsloth195.04 GiB-0.0%
Step-3.7-FlashBF16366.96 GiBunsloth366.96 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
Step-3.7-Flash770.56 GiB