Does AutoGGUF support quantization for SD3 or Flux and other DIT drawing models?

[Amazon90]

Your GUI tool is very beginner-friendly for those new to model quantization, but I’m unsure if your tool supports quantizing drawing models. I hope you can provide an answer.

https://github.com/city96/ComfyUI-GGUF/blob/main/tools/README.md

[leafspark]

It should support quantizing DIT models like SD3 and Flux. Here's how:

1. Convert your safetensors model to GGUF format:

python convert.py --src model.safetensors

2. Build and install the patched llama.cpp version and copy its binaries into a new folder in the llama_bin folder in your AutoGGUF directory.

3. Launch AutoGGUF - the new folder should appear in the backend selection.

4. Quantize normally, but:

   • Skip text model parameters
   • Use the Extra Arguments box for vision model settings

If there's prebuilt binaries in a repo you'd like to use, you could also set AUTOGGUF_BACKEND_REPO to owner/repo to pull new releases from that repository instead (although the binaries have to be in one zip file, same as the llama.cpp repo).