llama.cpp b9826 release fixes SYCL norm, adds multi-platform builds

ggml-org's llama.cpp, the go-to open-source library for running large language models locally on consumer hardware, has shipped version b9826. This release primarily addresses a failing unit test for the norm operation under the SYCL backend (issue #25044), which is critical for Intel GPU users leveraging the SYCL abstraction layer. The fix ensures numerical correctness and stability for models using normalization layers, a core component in transformer architectures.

Beyond the bug fix, b9826 showcases llama.cpp's relentless expansion of build targets. The release ships prebuilt binaries or configuration for over 20 platform-backend combinations: macOS (Apple Silicon, Intel, iOS as XCFramework), Linux (x64 CPU/arm64/s390x, Vulkan, ROCm 7.2, OpenVINO, SYCL FP32/FP16), Android (arm64 CPU), Windows (x64 CPU, arm64 CPU, OpenCL Adreno, CUDA 12.4, CUDA 13.3, Vulkan, OpenVINO, SYCL, HIP), and openEuler (x86 and aarch64 with ACL Graph). This breadth reinforces llama.cpp's mission to democratize local AI inference, letting professionals run models like Llama, Mistral, or Gemma on everything from a Raspberry Pi to a multi-GPU workstation.