The ggml-org/llama.cpp project has released version b9996, which includes a fix for log messages being lost before exit due to threading issues.

  • Flushes the log buffer before exiting to prevent message loss from multi-threaded LOG() calls.
  • Provides pre-built binaries for macOS (Apple Silicon and Intel), Linux (Ubuntu x64/arm64/s390x with CPU, Vulkan, ROCm, OpenVINO, SYCL), Android, Windows (CPU, OpenCL, CUDA 12/13, Vulkan, OpenVINO, SYCL, HIP), and openEuler.
  • Includes an updated UI build for general use.

This release ensures reliable logging output across supported platforms and hardware accelerators.