llama.cpp b10046 improves test reliability and expands platform coverage
New release fixes recurrent state rollback test, adds builds for ARM, CUDA, Vulkan, and more.
Deep Dive
The llama.cpp team's b10046 release, signed with GitHub’s verified signature, includes the commit "tests: actually exercise test-recurrent-state-rollback". Pre-built builds are available for macOS Apple Silicon (arm64)—note macOS Intel is disabled—Linux (Ubuntu x64, arm64, s390x, plus GPU backends Vulkan, ROCm, OpenVINO, SYCL FP32/FP64), Android arm64, Windows (x64, arm64, with CUDA 12/13, Vulkan, OpenVINO, SYCL, HIP), and an iOS XCFramework. Also included: UI Assets and openEuler builds (some disabled).
Key Points
- Fixes test-recurrent-state-rollback issue for better stability with recurrent neural network models
- Pre-built binaries available for macOS, Linux, Windows, Android, iOS, and openEuler across multiple architectures
- GPU support includes CUDA 12/13, Vulkan, ROCm 7.2, OpenVINO, SYCL, HIP, and OpenCL Adreno
Why It Matters
Simplifies deployment of local LLMs on diverse hardware, improving reliability for server and edge AI workloads.