You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Release master-945-a1ded76 pins the leejet/ggml submodule to a revision declaring ggml version 0.25.3. Upstream ggml has moved to 0.26.0. Llama and Whisper are using the new backend. I can't load stable-diffusion together with llama and whisper on a native window machine. I am considering the problem is the backend mismatch.
As I understand sdcpp devs are active in ggml development, especially pushing changes that involve features and improvements that impact sdcpp itself and update to versions that suit the requirements of sdcpp whenever best suited.
Activity
Enlightnd commentedon Oct 13, 2026
As I understand sdcpp devs are active in ggml development, especially pushing changes that involve features and improvements that impact sdcpp itself and update to versions that suit the requirements of sdcpp whenever best suited.
wbruna commentedon Oct 13, 2026
stable-diffusion.cpp builds with mainline 0.26.0; just define
SD_USE_UPSTREAM_GGMLto avoid depending on the specific features from leejet's fork.Another (more complex) option could be creating a local merge between 0.26.0 and the currently pinned version.
ardan-bkennedy commentedon Oct 13, 2026
So we need this project to rebase their fork to 0.26.0 and then apply their specific change?
Enlightnd commentedon Oct 15, 2026
Like I said, Stable Diffusion cpp updates their GGML pin regularly. You can literally see in the github that the last one was only a week ago.
You can update manually with git or use SD_USE_UPSTREAM_GGML like @wbruna said, or just wait until sdcpp repins GGML next.