The llama.cpp project has released version b9811, which includes a fix for a compiler bug affecting the conv2d coopmat2 path in Vulkan. This workaround is also applied to the CONV_3D implementation based on suggestions from NVIDIA engineer Jeff Bolz.

  • macOS Apple Silicon (arm64) and Intel (x64) binaries are available, with KleidiAI support disabled.
  • Linux builds cover CPU (x64, arm64, s390x), Vulkan, ROCm 7.2, OpenVINO, and SYCL FP32/FP16 variants.
  • Android arm64 CPU binaries are provided for mobile devices.
  • Windows releases include CPU, OpenCL Adreno, CUDA 12/13, Vulkan, OpenVINO, SYCL, and HIP builds.
  • openEuler support is disabled for x86 and aarch64 architectures.
  • A standalone UI binary is also included in the release.

This update ensures stability for users running convolution operations on Vulkan hardware by addressing specific compiler issues.