b9072

The latest tag b9072 for ggml-org/llama.cpp addresses a Vulkan SPIR-V shadowing bug (issue #22760) that could cause rendering artifacts or crashes in local LLM inference when using the Vulkan backend. This fix, verified with GPG signature, is crucial for users leveraging Vulkan for cross-platform GPU acceleration, as SPIR-V shadowing errors can lead to incorrect shader behavior. The release ensures smoother operation on systems without NVIDIA CUDA or AMD ROCm support, particularly on Linux and Windows with integrated or non-native GPUs.

Beyond the bug fix, b9072 significantly expands platform support. It provides pre-built binaries for macOS (Apple Silicon with optional KleidiAI, Intel x64), Linux (x64, arm64, even s390x mainframe), Windows (x64 and arm64 CPU, CUDA 12 & 13, Vulkan, SYCL, HIP), Android (arm64 CPU), and openEuler (custom x86 and aarch64 builds with ACL Graph). This breadth makes llama.cpp accessible on everything from smartphones to enterprise servers. For developers and researchers running models like LLaMA, Mistral, or Gemma locally, this release reduces setup friction and increases hardware compatibility, reinforcing llama.cpp's position as the go-to solution for on-device AI.