ggml-org has released llama.cpp b12090 with a fix for a failing fattn test in ggml-zdnn, listed as change #31583. The release page offers downloads for several systems, including an Ubuntu 26.04 s390x build with zDNN support and macOS builds for Apple Silicon and Intel. The notes do not describe other changes or say whether the fix affects end users beyond the test failure.
Why it matters
People using the Ubuntu 26.04 s390x zDNN build can install a release that includes the test fix, though the notes do not specify any change to runtime behavior.
