# ggufy

A lightweight command-line and GUI tool, written in Zig, that converts safetensors model files into various GGUF quantizations, targeting image and video diffusion model architectures.

ggufy converts tensor formats for diffusion models, mainly from safetensors into GGUF at a chosen quantization level, including formats such as F32, BF16, F16, and several 8-bit and 4-bit float and integer variants; models can optionally carry quantization-sensitivity data for supported architectures. It currently targets image and video diffusion model families such as SD1.5, SDXL, Flux, HiDream, Cosmos and WAN, among others listed in its architecture support table.

It ships as a single-file executable for Linux, Windows and macOS on both arm64 and x86_64, in both a command-line and a graphical version, and is built primarily in Zig with the ggml C and C++ library linked in for the quantization routines themselves; an early pure-Zig quantization attempt is kept in the project's history.

The author discloses that AI assistance was used in writing parts of the code, alongside human review of changes before merging.

Website: https://github.com/qskousen/ggufy