The llama.cpp project has released version b9877, which introduces a crash behavior when encountering multi-buffer inputs. This change is implemented via pull request #25276 to handle such cases explicitly.

  • macOS Apple Silicon builds are available for both standard and Intel architectures.
  • KleidiAI support on macOS Apple Silicon has been disabled in this release.
  • Linux binaries cover CPU, Vulkan, ROCm 7.2, OpenVINO, and SYCL variants for x64, arm64, and s390x.
  • Windows builds include CPU, CUDA 12/13, Vulkan, OpenVINO, SYCL, and HIP options.
  • Android arm64 and iOS XCFramework binaries are also provided.
  • openEuler support is partially disabled, with specific ACL Graph builds remaining available.