b8955

The llama.cpp project, maintained by ggml-org, has released version b8955, a significant update that refactors parameter handling for local AI model inference. This release introduces a cleaner separation of concerns by renaming the internal 'sparam' structure to 'sampling' and adding a dedicated sampling parameter category. The change simplifies configuration for developers running large language models on local hardware, making it easier to tune generation behavior without wading through legacy parameter names.

This release ships pre-built binaries across an extensive range of platforms, including macOS (Apple Silicon arm64, Intel x64), Linux (x64, arm64, s390x with CPU, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16), Windows (x64, arm64 with CPU, CUDA 12/13, Vulkan, SYCL, HIP), Android arm64, and iOS XCFramework. The update also includes a special KleidiAI-enabled build for Apple Silicon and openEuler builds for Ascend NPUs (310p and 910b). This broad support ensures that developers can run optimized local AI inference on nearly any modern device, from laptops to edge servers.