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
As mentioned in ggml-org/llama.cpp#15344 (comment), whisper.cpp currently fails to build with Vulkan enabled when using the latest version of shaderc, just like llama.cpp did until recently. This happens on both Linux and macOS, on both x86_64 and aarch641.
The problem was recently fixed in llama.cpp's copy of ggml by disabling certain shader optimizations (ggml-org/llama.cpp#15352). Applying this change to whisper.cpp's copy of ggml seems to get it building and running again. Presumably the base ggml repo has the same issue, although I can't test that through Nixpkgs. I'd open PRs for this myself, but I'm not really clear on how the three copies of ggml relate to each other, and I don't want to cause unnecessary merge conflicts for you.