idostyle

Might want to consider `GGML_CCACHE=OFF` instead of installing ccache inside the docker build. It gets you nothing in those isolated builds, as nothing will be cached.

We should push this further. As long as leejet is invited and included respectively going forward, I see no issue in focusing a more communal/organizational approach.

Nice job stduhpf. > Do you think transformer based models work better with importance, like ggml quants generally do? (eg. flux) Flux also seems to struggle with the lower bit...

`llama-imatrix --show-statistics` assumes that layer naming follows "blk.%d" instead of the `single_blocks.%d/double_blocks.%d` naming in flux and flux-lite. Would have to adjust `process_tensor_name` in that PR accordingly.

Isn't this missing the actual ggml sync? Something like in the screenshot should be shown in diff, right? 

Releases are build via Github CI. You should be able to inspect them under 'Actions' in this project. Direct link to latest master release windows vulkan build: https://github.com/leejet/stable-diffusion.cpp/actions/runs/16272721121/job/45944181356