b8258

The open-source project llama.cpp, maintained by ggml-org, has released a new commit (b8258) that expands its hardware support matrix and addresses a key server warning. The most notable addition is a new pre-built binary for Windows x64 systems with CUDA 13.1 DLLs, giving users another option for GPU-accelerated inference alongside existing CUDA 12.4 and Vulkan builds. This continues the project's mission to make running models like Meta's Llama 3 efficient and accessible across diverse consumer hardware.

The commit also includes a targeted fix for the server component, issuing a warning when the `--swa-full` argument is used with models that don't support Sliding Window Attention (SWA). This improves the developer experience by providing clearer feedback during configuration. The release follows llama.cpp's established pattern of providing a wide array of pre-compiled binaries for platforms including macOS (Apple Silicon and Intel), various Linux distributions (with CPU, Vulkan, and ROCm backends), and now an expanded Windows suite.